Does using AI for sales prospecting work? After sending close to a million cold emails using AI, we can confidently say yes — but only if you know how to do it right.
AI for online sales prospecting and automation isn't a magic wand. To be successful with AI, you need to learn when to use it (hint: for chunks of emails, not full emails!) and how to prompt well. That's what we'll focus on in this guide.
We'll cover:
• A framework for creating AI prompts for any use case you can dream up
• basic & advanced AI prompting frameworks
• case studies and prompt templates from our own experience
Our results speak for themselves: AI has saved our sales team hundreds of manual research hours—and doubled or tripled positive response rates. The time savings alone is worth the investment in the learning.
Sales teams have already used this guide to set up campaigns and land meetings in a matter of hours. With this guide, you can quickly get ahead of the curve and uplevel your outbound motion using AI.
If you need convincing to use AI for sales emails, read this…
From our experience sending millions of prospecting emails, AI will:
- Increase response rates: AI email campaigns routinely get us better response rates.
- In a recent campaign, using AI halved the number of emails we sent and doubled our positive responses. That's a 5x boost in positive response rate.
- e.g: We got 473 emails for 12 responses (2.5%) not using AI vs. 162 emails for 21 (13%) using AI.
- Leverage time better: We estimate that AI saves us at least 4 hours a day** on emails that would have otherwise needed manual research + writing.
- Boost email deliverability: Email providers are cracking down on mass templated emails. If you only use "first name + company name" with a phrase, you could get penalized, or worse, go to spam. Some people use spintax to tackle this; with AI, you can get unique personalized variants for your emails every time.
Here's proof that this stuff works!
Don't take our word for it. Before you move on and spend time learning about how to wield AI to improve your prospecting, look at real examples of what our AI-generated emails have done in the wild.
Real email responses from AI-generated campaigns 👇
.png)
.avif)
.avif)
Real reactions from other sales teams 👇
.avif)
Without further ado, let's get started!
1. Getting Started with AI Sales Prospecting
Emails written with the help of AI can yield better results than those manually written by SDRs. But only if you do it right.
In this section, we'll cover what you should use AI to do.
Tl;dr - use AI to write parts of your email. Ideally, the ones that vary based on personal inputs specific to your prospect.
Use AI to write customized lines based on inputs
AI shines when you want to write customized lines for emails based on dynamic or personal inputs (e.g - company descriptions, news, etc.)
Do use AI to generate highly creative & personalized copy, like custom value propositions or new marketing ideas. AI can also help clean unstructured datasets that are messy or variable in format—more on that in the our expert prompting section.
Don't use AI to write things that you can accomplish better with web scrapers. Static information, like simply inserting a job title or company location into an email, fall into this bucket. If you want to mention a prospect's job title in an email, just use Clay to scrape the information. (If you want to clean complex and varied sets of job titles, that's another story! AI can be very useful for data cleaning tasks like this).
First, write a template for your emails
Salespeople who complain about the quality of AI-generated emails are usually making one common mistake: asking AI to write full emails, without proper guidance. In these cases, the AI often “hallucinates” lines or ends up sounding overly sales-y— because of the copy that it has been trained on.
Instead, start with an email template using a few static pieces, and a few variable pieces. You can come up with your template manually or using AI (see this video for more on this).
Then, use AI to customize *parts* of your email
For all future emails, the AI will write just one specific part: the hook that is personalized to your receiver. This is the part you'd need a human to research and write otherwise. You're going to give your AI inputs, and ask it to make inferences based on those inputs.
Here's an example of an email template with both static and variable (i.e. written with AI) sections.
Static email with custom variables ready for AI generated text:
.avif)
Email with all of the text filled in:
.avif)
2. Basic AI Sales Prospecting prompts
If you want to use AI to help with your email workflows, writing a good prompt is the most important skill to learn. The better your prompts are, the more the AI will be able to help you.
The core concept of AI prompting is constrained creativity. You need to give the AI enough guardrails to get the kind of result you want.
For basic prompting, providing the AI with a set of instructions is sufficient. For advanced prompting, you'll want to give the AI instructions as well as a series of input and output pairs as examples to learn from. In this section, we'll cover basic prompting!
General rule: Be as specific as possible
Your prompts should be simply phrased and highly specific. Imagine you're explaining what you want to a fifth grader. Here's a general line by line plan you can use:
Summarize what you want and what input(s) to draw upon.
E.g. Tell me what this company does, based on the “company description” input.
Specify the input(s) in Clay.
E.g. Here is the “company description” input: [insert input].
Describe your desired output format (including word count, phrasing, etc).
E.g. Write a one sentence summary in under 10 words.
Provide a prefix, if applicable.
E.g. Start each summary with the prefix, “I noticed that your company”
Tip: AI prompting is a process of trial and error. AI can do an infinite amount of tasks and you need to constrain it to only accomplish your task successfully.
If you see the output is being too sales-y, tell it to write casually. If the output is coming out with too many sentences, specify the sentences. The framework above will help you start your prompt—but watch the outputs carefully and keep refining your prompt to get the output that you desire.
Examples of Basic AI Sales Prompts
Summarize Company Description
This prompt summarizes a company's description to use in an email, in a line like “I was on your site and saw that you…”
Prompt:
Using the input, generate a sentence summarizing what a company does.
This is the input: "{{Company Description}}".
Keep the output under 8 words and use specific keywords from the input.
Complete the sentence with my prefix. This is my prefix: "I was on your site and saw you…”
Infer Company Needs from Open Jobs
This prompt infers what a company might need based on what open roles it is hiring for.
Prompt:
Using the input, infer why a company is hiring for the role. The input that I will give you is the company description and the job title they are hiring for.
This the company description input: “ {{Company Description}}”. This is the job title input: “{{Job Title}}”.
Take your time to think about what the mission of the company is and how hiring this role would help them accomplish their mission. When completing the output, generate text about how hiring this role would solve a problem.
Complete the output with my prefix. This is my prefix: “I saw you on your website that you are hiring for a {{Job Role}}. In my experience, companies hire for this role to”
3. Expert AI Sales Prompts
✉️ Sometimes, one input won't be enough to guide your AI. For more complex use cases, where you want an output in a specific voice, style, or format, you should train the AI through a conversation, providing both general instructions and several input → output examples.
Here are use cases where this would be a good fit:
- Brainstorm creative marketing ideas
- Clean data
- Generate long form copywriting that requires large amounts of context
Read on to learn about how to give your AI training examples in order to accomplish more advanced prompts in the voice you want!
Advanced prompting has the same framework as basic prompting. Remember, the framework is:
Summarize the general goal:
E.g. Using the input, create a sentence summarizing this company's mission. (If you have multiple inputs, say something like: I'm going to give you multiple inputs. Using them, write a summary of X.)
Provide the core data
E.g. This is the input(s): ___. If any inputs are empty, ignore them.
Constrain creativity
. E.g. Keep the output <15 words, keep your tone casual, talk as if you were an acquaintance, use specific keywords, etc.
Specify what the message should look like
E.g. Everything should start with this prefix: “___.” Or the output should look like this: ___.
Example: Generative Creative Custom AI Use Cases
In this case, we used AI to generate creative ideas for how sales managers in a company could start using AI in their business in order to sell an agency's services.
Prompt
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "{{Description}}"
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice.
-Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
-To train the AI, we gave the AI three pairs of inputs and outputs.
Training Pair #1 - Apollo.io:
Input →
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "Apollo.io combines a buyer database of over 250M contacts and powerful sales engagement and other sales process automation tools in one, easy to use platform. Trusted by over 160,000 companies including Autodesk, Rippling, Deel, Jasper.ai, Divvy, and Heap, Apollo has more than one million users globally. By helping sales professionals find their ideal buyers and intelligently automate outreach, Apollo helps go-to-market teams sell anything."
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice. Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
Output →
-Train AI to onboard new staff members to learn the ins and outs of outbound prospecting faster.
-Use AI to help customers come up with copywriting they can use inside of Apollo to get better response rates.
-Use AI to predict which deals are going to close and when and then teach your customers how you did it.
Training Pair #2 - Red Bull:
Notice that the input is entirely the same except for the company example
Input →
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "Red Bull | 786,465 followers. Red Bull Gives Wings to People and Ideas. | Red Bull Gives Wings to People and Ideas. This drives us - and all we do - since 1987. Today, Red Bull operates in over 170 countries selling more than 7.9 billion cans annually and growing!"
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice. Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
Output →
-Use AI to come up with new marketing ideas to test and get Red Bull out to more people.
-Train AI to onboard new employees so they learn the ins and outs of the logistics infrastructure that supports Red Bull.
-Use AI to analyze demand and see where you should be charging more or giving discounts for Red Bull.
Training Pair #3 - Kajabi:
Input →
-This is the input: "Kajabi | 7,077 followers. Empowering knowledge entrepreneurs who are serious about their business to achieve success online. | Since its founding in 2010, Kajabi has empowered knowledge entrepreneurs who are serious about their business to achieve success online. With its robust, ever-expanding set of digital tools, and features, Kajabi's all-in-one knowledge commerce platform leads the industry and helps people to turn what they know into what they do."
Output →
-Add AI into your product to help course creators write their courses faster.
-Use AI to understand which courses are growing the fastest and onboard more coaches for those industries.
-Train an AI model to teach your team how to become experts at giving customers support with Kajabi.
-This is what the prompt and training pairs would look like in Clay. The input in Example 1 and Example 2 is the same as the full prompt we started with above.
.avif)
.avif)
.avif)
Example: Cleaning Job Titles
When you scrape job titles, people often put extra information in the titles that makes it pretty clear that the title was scraped and used as a custom variable. We can clean them using AI in just three steps.
Input →
Prompt:
Clean these job titles from extra information. Often times, people put extra information that is unnecessary in their job titles. Remove this information and simplify the job titles. Here are the titles I want you to clean:
CEO || World ChangerFounder YC 22
CRO/Chief Strategy Officer
Chief Operating Officer and Chief of Staff
Human Resources Director
Engineering, Director
Account Executive - Enterprise Sales - APAC
Output →
CEO
Founder
CRO
Chief Operating Officer
Director of Human Resources
Director of Engineering
Account Executive
This is how it'll look in Clay:
.avif)
.avif)
4. Our 13 Best AI Sales Prompt Templates
Including unique and relevant personalizations in your emails can make or break your prospecting campaigns. The best personalizations communicate a simple but powerful message: I researched you, I understand your problem, and I can show you how to fix it.In this section, we'll show you our most successful AI prompts and personalizations. These have routinely 2x'd response rates and led to hundreds of meetings booked for companies. Try them out!
You can browse our full prompt library below, or check out this video to learn how to implement several of them at once.
🍎 Simple, Personalized Lines for Emails
Summarize a company's mission
This personalization uses AI to summarize the mission statement of any company. This is one of our most useful personalizations and reliably boosts positive response rates.
Many companies employ SDRs to manually write these lines, but you can fully automate this process with AI. This personalization helped one of Eric's clients, a marketing agency, more than double their response rates from prospective clients (from 1.5% to 3.2%).
To get started, we simply input any company description into Clay and use the ChatGPT integration to summarize its mission statement. We specify constraints (e.g., “use less than eight words and use specific words from the company description”). We format our results into the following sentence: “I was on your website, and it looks like you help [X].”
Prompt:
Using the input, summarize the company mission in one sentence. This is the input: “{{Company Description}}”.
Keep the output under 12 words and use keywords from the input. Complete each output with my prefix.
This is my prefix: “I was on your site and it looks like you help”
Here's an example:
.png)
The strategy applies to anything you might want to summarize. Another idea would be to use AI to summarize what makes a company different from its competitors. Try it out!
Summarize a Person's Profile
This prompt summarizes someone's profile, highlighting anything that's important to you. This is fun to do in Clay, using GPT3.5 to keep costs down. You can check someone's role history, seniority, current role's focus, and more.
Prompt:
Using the input, summarize the profile in a couple of sentences. The input I will give you is in JSON format. Here is the input: “{{Input a whole social profile}}”. Summarize the profile for me and look out for specific parts about their profile.
Besides giving me a general summary, also tell me:
- Has the person held their current role multiple times or is the first time they are in this seniority?
- What is their current role's focus?
- Do they move jobs often?
For all of this data look at thei profile and think methodically before answering. Summarize the output in 4-6 sentences focusing on what I want to know about them.
Change Messaging Depending on a Person's Role
This is a great workflow if your product has different value propositions for different titles (e.g., growth marketers vs. CMOs).
You can use Clay's persona flow and formulas to say something different but this requires an exact keyword match. You can use this if sometimes the titles change slightly and you want to match up the selling propositions.
The best way to approach this is by using two different AI integrations:
The first one is to infer what the title cares about.
The second one is to write the messaging.
First, prompt AI to make decisions about the problems that a title would be facing.
Prompt 1:
Using the input, write out 2-3 problems a person is responsible for solving in their role. This is the input: “{{Job Title}}”.
Be specific about the particular problems a person in this role faces that a person in a more generalized role would not face.
Keep the output to 2-4 sentences and be very specific about the problems they face.
The output is going to give you a summary of what each particular title would be facing as a problem. We then take that summary and write problem based messaging to everyone.
Prompt 2:
The output we are going to create is a sentence that casually communicates a value proposition given the problems a person faces in their role. The input I'm going to give you is the problems and responsibilities of the role.
This is the input: “{{Summary from last AI prompt}}”. Considering these problems and responsibilities, pick out what problem my company solves that would resonate most with this person.
These are the problems my company solves: “{{Insert your value propositions}}”.
Complete the output with my prefix. This is my prefix: “When I speak to other {{titles}}, they tell me they see problems like”
This would lets you speak directly to the customer relationship management problem they see as it relates to your product or service. You can follow up with how you have helped other companies like them!
Mention a Podcast that Someone Was Recently On
You can achieve this by doing a site search on Listen Notes. First, we can use Clay's Google Search integration to search [listennotes.com](http://listennotes.com) for any podcast mentions of the person's name. The search would look like this in Google.
site:listennotes.com “Person Name”
Then, the Google Search integration will give you data on the podcast name and a little bit about what they discussed. Use those inputs for your prompt!
Prompt:
Using the input, complete a sentence about a podcast this person was on. I will give you the title of the podcast and a small snippet about the podcast. Here are the inputs: “{{Podcast Title}}”. The other input is here: “{{Podcast Snippet}}”. Keep the output under 12 words and use keywords from the inputs. Complete each output with my prefix. This is my prefix: “I saw your podcast about”.
Summarize a Recent News Event for a Custom First Line
Tying outreach to recent events can be a great way to do trigger-based targeting. This prompt summarizes a recent news event for a custom first line.
Prompt:
Using the input, summarize a recent news article in one sentence. This is the input: {{News Article Headline}}”. Keep the output under 12 words and use keywords from the input. Complete each output with my prefix. This is my prefix: “I saw the recent article about”
🔮 More Complex and Creative AI Sales Prompts
Generate Creative Marketing Ideas for Prospects
This strategy uses AI to generate useful marketing ideas for any company. The marketing ideas could include outbound campaign ideas, topics to post about on social media, shortform videos to create, etc., depending on what you want to sell.
Eric uses this strategy to send about 300 emails a day and gets ~3 positive responses for his outbound agency. It's easy to tweak this for anything that you're selling.
In Clay, we used the company description as input and asked ChatGPT to use it to think of three creative marketing ideas that are possible with outbound marketing. We set up an AI assistant so that we could provide up to ten examples of inputs and great answers—and then we were ready to go.
Here's an example email generated entirely with AI in Clay:
Hi X,
It looks like [Company Name] helps manage and protect Microsoft infrastructure. Have you considered running any of the following outbound campaigns to gain new customers?
- Target IT managers involved in cloud migration, offering tailored solutions to simplify their transition
- Reach out to CISO or security managers, highlighting your security and compliance benefits
- Email Microsoft certified professionals, emphasizing your compatibility with Microsoft environments and how you can increase their efficiency.
Are you doing any of these right now, or would you need to help automating or initiating any of them?
This strategy represents a stark departure from conventional outbound prospecting, since you have no social proof, offers, or guarantees. What you do have, however, are concrete and imaginative ideas for your prospect that are relevant to what they're trying to sell or achieve—and an easy way to offer to help.
Here's our prompt:
Using the input, create three bullet points of creative outbound marketing ideas for a company. The ideas have to be possible with cold email marketing. This is the input: [company description]. Analyze who their target audience would be by job title and think of creative ways to market to them. Be specific and give unique advice. Keep each bullet under 15 words and use strategies that use outbound marketing automation.
Watch this video to set it up, step by step.
Generate Ideas for How Someone Could Use a Virtual Assistant
The goal of this campaign is to use AI to summarize the mundane tasks a business owner might have, which they might be able to use a virtual assistant's help with. We also summarize the fulfilling tasks that they get to do. This resulted in a 100% increase in meetings booked for the company that used it!
Here is an example two-step sequence:
Email 1:
.png)
Email 2:
.png)
We need two AI prompts here. The first one writes a line about the mundane, repetitive tasks a business has to do. The second one writes the lines about the fulfilling tasks that they'd rather be doing.
Prompt 1 (generate mundane tasks)
Using the input, give me three ideas for a small business on how they could use a virtual assistant to help them.
This is the input: {company description}.
Keep the output under 15 words per idea and be sure to output the ideas in one line separated by a comma. Be creative and use keywords from the input.
To help train the AI to give you the outputs you want, use a few input/output examples. Here is an example output we wrote manually to summarize mundane tasks a certain company owner would have: “staying in contact with vendors and collecting quotes for your clients to review, collecting remaining invoices and posting your wins on social media, and helping your 8 senior event consultants move twice as fast by offloading manual tasks and administrative tasks together.”
Prompt 2 (generate fulfilling tasks):
Using the input, create a 6 word output with some of the most rewarding tasks within a business. This is the input: {company description}.
Keep the output under 6 words. Keep the output in all lower case and be specific about the task. Complete each output with my prefix. This is my prefix: "to add some context, I figured you might have parts of your business like”
We used input/output pairs here as well to teach the AI. Here's an example output we wrote manually: “to add some context, I figured you might have parts of your business like adding personal touches that make things feel memorable.”
Change Messaging Based on Company Mission & A Recent Social Media Post
Here's an example of two campaigns we sent to invite folks to a dinner hosted by a business influencer. The AI campaign used less time, sent drastically less emails, and generated a better open rate and more responses.
RESULTS:
Without AI: 473 emails, 80% open rate, 12 replies
.avif)
With AI: 162 emails sent, 85% open rate, 21 responses
.avif)
The campaign that used AI even sparked responses from people who thanked us for our email!
.avif)
.avif)
.avif)
Here's the non-AI campaign email format:
.avif)
Here's the AI email format:
.avif)
And this is an example of an actual AI-generated email:
.avif)
This is what the prompt looked like for the first line.
Using the input, complete a sentence summarizing what someone's last social media post was about. This is the input: "{{Social Media Post Text}}" Keep the output under 12 words and use specific keywords from the input. Complete each output with my prefix and do not incude quotations or hashtags (#) in the output. This is my prefix: "I saw your recent post about"
This is what the prompt looked like for the second line:
Using the input, complete a sentence summarizing what someone's company does: "Company Description" Keep the output under 12 words and use specific keywords from the input. Complete each output with my prefix and do not include quotations or hashtags (#) in the output. This is my prefix: "Further than that, I saw how you've been helping people"
Infer What Job Titles a B2B Company Sells To
You can use this prompt to infer what job titles or type of people a company sells to. This works great if your product helps your potential customers to sell to a certain client more effectively.
Prompt:
Using the input, determine the job titles or type of people that this company sells to. This is the input: “{{Company Description}}”. Keep the output to just a list of 3 types of people. Be specific about who they sell to or what their job title is.
Here is an example that has nothing to do with the input, it just is an example of formatting. VPs of Sales, Chief Revenue Officers, and SDR Leaders.
Output who the company in the output most likely sells to. Only name their titles, don't include anything else.
5. Automate Entire Campaigns with Clay
Once you've mastered using any kind of AI prompt, why not automate your entire sales cycle with multi-step email campaigns?
Read on to see three examples of how we've:
- Personalized an entire 4-step sequence within Clay, earning 10%+ positive response rates.
- Automate outreach to prospects based on seven different triggers
- Filter a list of ICPs into 1 of 6 best-fit email campaigns based on their situation</aside>
Personalize an Entire 4-Step Email Sequence
When we worked with Taplio, a tool that helps people schedule posts and announcements, we personalized an entire 4-step sequence using AI. This campaign generated a 10% positive response rate.
Here's a video walkthrough of the entire campaign.
Here's each email of the campaign:
Email 1
We generated this email by scraping their last 3 social profile post results, and then calling out the time difference between post 2 and post 3.
{{First name}} – Saw one of your recent posts about {{AI summary of posts}}. Seems like the last time you posted was {{# of days}} days before that though.
If you're trying to build your brand, have you considered using a scheduling tool?
Our scheduling tool can help write your posts and help you monitor social trends that you can riff on as well.
Let me know if this would be useful to you and I can send over access!
Email 2
This is a follow-up that adds value to the sequence by following up with the prospect's most recent post.
{{First Name}} – I just saw your post about {{AI generation}}. It reminded me to follow up and see if you saw this!
Email 3
Taplio helps create content with AI generated suggestions. We took person's recent posts and their company description to automatically create post ideas that they could bring to their social media.
{{First Name}} – I keep seeing your content online so I made some more post ideas considering what your company does and how you promote it.
{{Creative Ideas}}
This is just some examples of what our AI can generate for you (and you haven't even given it your insights yet!)
Let me know if I could send over some access for you to try it out.
Email 4
Here, we used AI to extract out and normalize their followers so that we can mention in an email.
{{First Name}} – just like how Facebook pages killed their organic reach, it can be expected that your current social media will do the same one day.
I saw your company page had {{# of Followers}} followers and you had about {{# of followers}} followers.
As you are building your following, are you maximizing the platform's organic reach before it gets oversaturated?
We helped companies like Lemlist and Outreach double down on their social content with our analytics, scheduling, lead generation, management and content generation tool.
Could I send you access to try it out?
Reach Out to Prospects Based on 7 Triggers
This workflow analyzes companies for seven different triggers and why we should reach out to them. This automates process automation using a methodology called "Why You, Why Now" from John Barrows. This workflow is great for customer support tech or service companies.
You can see the full process in this video, including how to set each trigger up on Clay, step by step. Once you've set up these triggers, you can use them in any order you'd like to fill out a personalized trigger-based sequence.
Here's each trigger from the John Barrows methodology:
Trigger 1
Look for a new customer support team lead and check what tool they used at their last company. (This wedges you into the company when a new hire might be shaking things up!)
Example of first line:Hey Suchita, noticed you recently were promoted to Principal Customer Success Manager in March. Before working at 6sense, I saw you worked at Yext. Noticed both companies are using Zendesk and Drift right now. How's your experience with them?
Trigger 2
Look for the % growth of the sales productivity of customer team staff. This helps you understand if the team might be outgrowing past processes!
Example of first line:I noticed that 6sense added 20 customer support people in the last 3 months. Are your current processes still working with the team growth?
Trigger 3
Look for new product launches (which always come with new support demands). Use AI to discuss what support questions might come up with the new product.
Example of first line:
I figured customers might reach out for support around things like utilizing the new AI Writer tool in their own own sales process automation work and automation process.
Trigger 4
Look at sales data for companies with no support tech on site. We can even compare this to how many employees the company has and how long they have been in business.
Example of first line (for a different company):
Noticed you aren't using any support tech on the site. I'd imagine that since the business was founded in 2014 and you just broke 100 employees that your old sales processes might not be working any more.
Trigger 5
See if a company is using a competitor's support tech on site.
Example of first line:
Noticed you're using Zoho Support, have you found their multichannel inbox to be useful in support or are things falling through the cracks?
Trigger 6
Notice if a company is hiring for customer support. Use AI to summarize tasks mentioned in the job description that your tool could optimize or replace!
Example of first line:
Noticed you're hiring for as Customer Support Associate. Took a look at the job description and it looked like they will need to respond to emails with appropriate answers to customers. Are you thinking about how AI can be leveraged to speed up this process?
Trigger 7
See whether the company has a generic support email and either no support tech on site OR they have support tech on site (which let's us know that they need to have everything in one place).
Example of first line:
Noticed on the site you weren't using a support software but I found support@domain.com is up there. Are you finding that you can answer all of your support questions without dropping things with just an inbox?
In a follow up email, we could say something like below to keep the personalization going.
"I figured customers might reach out for support around things like leveraging 6sense's AI for creating and converting high-quality pipelines."
Filter Your ICP into 6 Unique Campaigns
This is one of the craziest workflows we've ever built. Here, we used Clay to take one Apollo export and turn that data into six different campaigns depending on industry, employee headcount, technology—with AI lines for each prospect. This was used to save over 6 hours per week of manual effort and there was no loss in positive response rate.
This campaign was selling the very best sales automation software and engagement tools to marketing leaders in American B2B companies with under 100 employees. This video goes over everything in depth!
Step 1: Verify emails
We used Debounce to verify prospects' emails to make sure we can contact them. If we didn't find a verified email, we used Clay's “get email” function to guess our prospect's work email. We put our best results into a master email column.
Step 2: Pull Company Description
We checked if the companies on our list are B2B using the “company description” tag on Apollo. You can also get this information using Clay's “enrich company data” feature. We put our best results into a master “company description” column.
Step 3: Use OpenAI to Check if the Company is B2B
We set up a column using OpenAI using the following prompt:
Tell me if the company from my given input is B2B or B2C. A B2B company is a company that sells to other companies. A B2C company is a company that sells directly to consumers. The input is [insert “Company Full Description” column from table]. Only respond with “B2B” or “B2C.
Nothing continued being analyzed in the table unless it was a B2B company.
Step 4: Sort the Company Into a Campaign
Check the “Industry” column to sort the company into a campaign for management consulting, financial services, and recruiting businesses. If the company doesn't fit any of those four main buckets, we check if employee count is under ten and add it to our Facebook campaign.
Otherwise, we check their website to see if the company has any CRMs we can integrate with or replace. In our final, catch-all campaign, we simply use ChatGPT to draft a custom starting line about the company. We also add a column to check if emails are sent and to add prospects to a running Google sheet of people we've contacted.
Step 5: Refill the Campaign
Once you've set up your campaign, you can import any list of CSVs to add to the current table. Click “add to current table” and carefully map each piece of data that you need for your table to run. Watch the magic happen! (Even if these leads are already in the campaign, Instantly.ai will automatically dedupe them).
With that, you can start with a list of qualified leads, sort them, and automate them into different campaigns with unique copy based, and track the emails you've sent!
6. How Clay Uses AI... Within Clay
For the finale of our guide, we're going to show you how we use Clay…at Clay!
We've used AI to fully automate a 4 step campaign for our own outbound sequences, getting a 5.1% positive response rate.
Every email in the sequence used multiple levels of personalization. This table took under an hour to put together and is easily one of our most powerful outbound campaigns running today.
Let the Numbers Speak for Themselves
Before we dive in to the methodology, let's show you some results:
.avif)
These stats are for every single email in the sequence. Below, we've included copy of every email in the sequence. Each email was sent 5-7 days after the previous one. We used many layers of personalization, and also made sure to spread them out over time. (One common mistake is to spend a lot of time on personalization and relevance, but to use it all on the first email.)
Email 1️
{{First Name}} – I was on your site and it looks like you sell to {{job titles}} titles. It looked like you are helping {{AI summary of company mission}}.
I had to assume you are using outbound tactics considering {{Full Name of SDR}} is an {{SDR/BDR}} on the team.
Are you thinking about how they could be leveraging better data and GPT-4 to do their prospecting?
The reason I ask is because we are helping companies like Rippling and Ramp with our tool that collects the data and writes the messaging for them.
Would this be worth a chat to build a custom outbound table for you?
Email 2️
{{First Name}} – as you're helping people {{AI summary of company mission}}}, I'd imagine you are using outbound marketing tactics like the following.
{{Creative Ideas}}
Have you cracked how to do this persona targeting and data enrichment automatically or is your team still using tools like Sales Nav to do manual research?
If so, would it be worth taking some time to show you how we can automate persona and account research for your outbound team?
Email 3️
{{First Name}} – I sent an email a while back with the subject “Outbound edge” that probably didn't do a good enough job showing you how we could help your outbound team.
Our tool combines over 40 databases into one spreadsheet so that your team can see all the benefits of our sales automation technology with the data in front of them instead of manually researching accounts.
For example at your company
- It looks like your sales team has seen a {{percent change}}% change in the past 3 months.
- Your company account has gained {{# of Followers}} in the past 3 months.
- The company looks like it has {{# of sales people}} people with “Sales” in their title based in the United States today.
Besides things like this, we also monitor for recent news, open job positions, tech on site and mentioned in job descriptions, and a lot more.
On a call, we can build a custom workflow for your team to automate their outbound research. Wanna make it happen?
Email 4️
{{First Name}} – Perhaps this is not your area of focus. Should I potentially reach out to {{Other sales rep/marketing person in the company that's NOT them}} given their role as a {{Job title cleaned with AI}}.
💫 Recreate this campaign in Clay
To recreate this campaign within Clay, just watch this video!
Here is each step we go through in the video:
Video: Email 1️
- Import your data or lead list
- Verify email addresses
- Use “Find Work Email” integration to generate a master email column
- Normalize company names
- Get company descriptions
- Use AI to figure out who the company sells to
- Prompt: Using the input, what types of people or specific job titleswould a company sell to? This is the input: [ ] Who gets the most value orproduct of the service add specific job titles or a type of person return up tothree job titles do not include any numbers or extra information an examplewould look like this VPS or sales and two like all that stuff right and sothey sell to CEO CFOs and chief investment officers they sell to contentcreators filmmakers and video editors
^ All of that creates our first line: {{First Name}} – I was on your site and it looks like you sell to {{job titles}}. It looked like you are helping {{company mission}}.
- Use the “Find People” enrichment to find the full name of the SDR or BDR on the team
In our formula, if someone title includes “sales” we call then an SDR; if it contains “business,” we call them a BDR. If we can't find either, we don't reach out to the company at all. We'd rather send fewer emails that land with hyper targeted personalizations.
^ This is the ingredient for the next part of the email: I had to assume you are using outbound tactics considering {{Full Name}} is an {{SDR/BDR}} on the team.
Video: Email 2️
- Generate and insert creative outbound campaign ideas for these companies.
Using AI, we generate outbound ideas. As an example from an education company, we generated:
- Targeting School principles and district manager administrators highlightingstreamlined remote learning with class tools on Zoom
- Reaching out to university educators showcasing classes for effective higher education virtual classrooms and lectures
- Connecting with professional trainers offering online courses promoting productivity and engagement tools for learning
Our email uses these ideas to ask:
As you're helping people {{company mission}}, I'd imagine you are using outbound marketing tactics like the following.
{{Creative Ideas}}
Have you cracked how to do this persona targeting and data enrichment automatically or is your team still using tools like Sales Nav to do manual research?”
Video: Email 3️
In this email, we get use the Harmonic integration to get information about the sales funnel, team growth and company account growth (remember to normalize all percentages!) Lastly, to get the number of salespeople at the company, we use the “Employee Headcount by Title” integration.
We only email the people for whom we can find all this data. We'd rather send fewer emails, but make them the best, most personalized emails that we can come up with!
Video: Email 4️
For this email, we want to find other top sales reps and marketing people at the company to reach out to. We insert a bunch of example titles (such as “Director of Marketing,” “Director of Customer Initiatives,” etc) and exclude the ones we don't want (such as “development,” “representative,” or “engineer.”)
💡 For more tips on using AI in your outbound campaigns, sign up for our newsletter!
Does using AI for sales prospecting work? After sending close to a million cold emails using AI, we can confidently say yes — but only if you know how to do it right.
AI for online sales prospecting and automation isn't a magic wand. To be successful with AI, you need to learn when to use it (hint: for chunks of emails, not full emails!) and how to prompt well. That's what we'll focus on in this guide.
We'll cover:
• A framework for creating AI prompts for any use case you can dream up
• basic & advanced AI prompting frameworks
• case studies and prompt templates from our own experience
Our results speak for themselves: AI has saved our sales team hundreds of manual research hours—and doubled or tripled positive response rates. The time savings alone is worth the investment in the learning.
Sales teams have already used this guide to set up campaigns and land meetings in a matter of hours. With this guide, you can quickly get ahead of the curve and uplevel your outbound motion using AI.
If you need convincing to use AI for sales emails, read this…
From our experience sending millions of prospecting emails, AI will:
- Increase response rates: AI email campaigns routinely get us better response rates.
- In a recent campaign, using AI halved the number of emails we sent and doubled our positive responses. That's a 5x boost in positive response rate.
- e.g: We got 473 emails for 12 responses (2.5%) not using AI vs. 162 emails for 21 (13%) using AI.
- Leverage time better: We estimate that AI saves us at least 4 hours a day** on emails that would have otherwise needed manual research + writing.
- Boost email deliverability: Email providers are cracking down on mass templated emails. If you only use "first name + company name" with a phrase, you could get penalized, or worse, go to spam. Some people use spintax to tackle this; with AI, you can get unique personalized variants for your emails every time.
Here's proof that this stuff works!
Don't take our word for it. Before you move on and spend time learning about how to wield AI to improve your prospecting, look at real examples of what our AI-generated emails have done in the wild.
Real email responses from AI-generated campaigns 👇
Real reactions from other sales teams 👇
Without further ado, let's get started!
1. Getting Started with AI Sales Prospecting
Emails written with the help of AI can yield better results than those manually written by SDRs. But only if you do it right.
In this section, we'll cover what you should use AI to do.
Tl;dr - use AI to write parts of your email. Ideally, the ones that vary based on personal inputs specific to your prospect.
Use AI to write customized lines based on inputs
AI shines when you want to write customized lines for emails based on dynamic or personal inputs (e.g - company descriptions, news, etc.)
Do use AI to generate highly creative & personalized copy, like custom value propositions or new marketing ideas. AI can also help clean unstructured datasets that are messy or variable in format—more on that in the our expert prompting section.
Don't use AI to write things that you can accomplish better with web scrapers. Static information, like simply inserting a job title or company location into an email, fall into this bucket. If you want to mention a prospect's job title in an email, just use Clay to scrape the information. (If you want to clean complex and varied sets of job titles, that's another story! AI can be very useful for data cleaning tasks like this).
First, write a template for your emails
Salespeople who complain about the quality of AI-generated emails are usually making one common mistake: asking AI to write full emails, without proper guidance. In these cases, the AI often “hallucinates” lines or ends up sounding overly sales-y— because of the copy that it has been trained on.
Instead, start with an email template using a few static pieces, and a few variable pieces. You can come up with your template manually or using AI (see this video for more on this).
Then, use AI to customize *parts* of your email
For all future emails, the AI will write just one specific part: the hook that is personalized to your receiver. This is the part you'd need a human to research and write otherwise. You're going to give your AI inputs, and ask it to make inferences based on those inputs.
Here's an example of an email template with both static and variable (i.e. written with AI) sections.
Static email with custom variables ready for AI generated text:
Email with all of the text filled in:
2. Basic AI Sales Prospecting prompts
If you want to use AI to help with your email workflows, writing a good prompt is the most important skill to learn. The better your prompts are, the more the AI will be able to help you.
The core concept of AI prompting is constrained creativity. You need to give the AI enough guardrails to get the kind of result you want.
For basic prompting, providing the AI with a set of instructions is sufficient. For advanced prompting, you'll want to give the AI instructions as well as a series of input and output pairs as examples to learn from. In this section, we'll cover basic prompting!
General rule: Be as specific as possible
Your prompts should be simply phrased and highly specific. Imagine you're explaining what you want to a fifth grader. Here's a general line by line plan you can use:
Summarize what you want and what input(s) to draw upon.
E.g. Tell me what this company does, based on the “company description” input.
Specify the input(s) in Clay.
E.g. Here is the “company description” input: [insert input].
Describe your desired output format (including word count, phrasing, etc).
E.g. Write a one sentence summary in under 10 words.
Provide a prefix, if applicable.
E.g. Start each summary with the prefix, “I noticed that your company”
Tip: AI prompting is a process of trial and error. AI can do an infinite amount of tasks and you need to constrain it to only accomplish your task successfully.
If you see the output is being too sales-y, tell it to write casually. If the output is coming out with too many sentences, specify the sentences. The framework above will help you start your prompt—but watch the outputs carefully and keep refining your prompt to get the output that you desire.
Examples of Basic AI Sales Prompts
Summarize Company Description
This prompt summarizes a company's description to use in an email, in a line like “I was on your site and saw that you…”
Prompt:
Using the input, generate a sentence summarizing what a company does.
This is the input: "{{Company Description}}".
Keep the output under 8 words and use specific keywords from the input.
Complete the sentence with my prefix. This is my prefix: "I was on your site and saw you…”
Infer Company Needs from Open Jobs
This prompt infers what a company might need based on what open roles it is hiring for.
Prompt:
Using the input, infer why a company is hiring for the role. The input that I will give you is the company description and the job title they are hiring for.
This the company description input: “ {{Company Description}}”. This is the job title input: “{{Job Title}}”.
Take your time to think about what the mission of the company is and how hiring this role would help them accomplish their mission. When completing the output, generate text about how hiring this role would solve a problem.
Complete the output with my prefix. This is my prefix: “I saw you on your website that you are hiring for a {{Job Role}}. In my experience, companies hire for this role to”
3. Expert AI Sales Prompts
✉️ Sometimes, one input won't be enough to guide your AI. For more complex use cases, where you want an output in a specific voice, style, or format, you should train the AI through a conversation, providing both general instructions and several input → output examples.
Here are use cases where this would be a good fit:
- Brainstorm creative marketing ideas
- Clean data
- Generate long form copywriting that requires large amounts of context
Read on to learn about how to give your AI training examples in order to accomplish more advanced prompts in the voice you want!
Advanced prompting has the same framework as basic prompting. Remember, the framework is:
Summarize the general goal:
E.g. Using the input, create a sentence summarizing this company's mission. (If you have multiple inputs, say something like: I'm going to give you multiple inputs. Using them, write a summary of X.)
Provide the core data
E.g. This is the input(s): ___. If any inputs are empty, ignore them.
Constrain creativity
. E.g. Keep the output <15 words, keep your tone casual, talk as if you were an acquaintance, use specific keywords, etc.
Specify what the message should look like
E.g. Everything should start with this prefix: “___.” Or the output should look like this: ___.
Example: Generative Creative Custom AI Use Cases
In this case, we used AI to generate creative ideas for how sales managers in a company could start using AI in their business in order to sell an agency's services.
Prompt
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "{{Description}}"
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice.
-Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
-To train the AI, we gave the AI three pairs of inputs and outputs.
Training Pair #1 - Apollo.io:
Input →
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "Apollo.io combines a buyer database of over 250M contacts and powerful sales engagement and other sales process automation tools in one, easy to use platform. Trusted by over 160,000 companies including Autodesk, Rippling, Deel, Jasper.ai, Divvy, and Heap, Apollo has more than one million users globally. By helping sales professionals find their ideal buyers and intelligently automate outreach, Apollo helps go-to-market teams sell anything."
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice. Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
Output →
-Train AI to onboard new staff members to learn the ins and outs of outbound prospecting faster.
-Use AI to help customers come up with copywriting they can use inside of Apollo to get better response rates.
-Use AI to predict which deals are going to close and when and then teach your customers how you did it.
Training Pair #2 - Red Bull:
Notice that the input is entirely the same except for the company example
Input →
-Using the input, create three bullet points of creative ideas for a company to start using AI in their processes. The ideas have to be possible with artificial intelligence.
-This is the input: "Red Bull | 786,465 followers. Red Bull Gives Wings to People and Ideas. | Red Bull Gives Wings to People and Ideas. This drives us - and all we do - since 1987. Today, Red Bull operates in over 170 countries selling more than 7.9 billion cans annually and growing!"
-Take your time and think about what the company does, who they help, and what operations in their business could be streamlined with AI. Be specific and give unique advice. Keep each bullet under 15 words and give suggestions on how the business can leverage AI.
Output →
-Use AI to come up with new marketing ideas to test and get Red Bull out to more people.
-Train AI to onboard new employees so they learn the ins and outs of the logistics infrastructure that supports Red Bull.
-Use AI to analyze demand and see where you should be charging more or giving discounts for Red Bull.
Training Pair #3 - Kajabi:
Input →
-This is the input: "Kajabi | 7,077 followers. Empowering knowledge entrepreneurs who are serious about their business to achieve success online. | Since its founding in 2010, Kajabi has empowered knowledge entrepreneurs who are serious about their business to achieve success online. With its robust, ever-expanding set of digital tools, and features, Kajabi's all-in-one knowledge commerce platform leads the industry and helps people to turn what they know into what they do."
Output →
-Add AI into your product to help course creators write their courses faster.
-Use AI to understand which courses are growing the fastest and onboard more coaches for those industries.
-Train an AI model to teach your team how to become experts at giving customers support with Kajabi.
-This is what the prompt and training pairs would look like in Clay. The input in Example 1 and Example 2 is the same as the full prompt we started with above.
Example: Cleaning Job Titles
When you scrape job titles, people often put extra information in the titles that makes it pretty clear that the title was scraped and used as a custom variable. We can clean them using AI in just three steps.
Input →
Prompt:
Clean these job titles from extra information. Often times, people put extra information that is unnecessary in their job titles. Remove this information and simplify the job titles. Here are the titles I want you to clean:
CEO || World ChangerFounder YC 22
CRO/Chief Strategy Officer
Chief Operating Officer and Chief of Staff
Human Resources Director
Engineering, Director
Account Executive - Enterprise Sales - APAC
Output →
CEO
Founder
CRO
Chief Operating Officer
Director of Human Resources
Director of Engineering
Account Executive
This is how it'll look in Clay:
4. Our 13 Best AI Sales Prompt Templates
Including unique and relevant personalizations in your emails can make or break your prospecting campaigns. The best personalizations communicate a simple but powerful message: I researched you, I understand your problem, and I can show you how to fix it.In this section, we'll show you our most successful AI prompts and personalizations. These have routinely 2x'd response rates and led to hundreds of meetings booked for companies. Try them out!
You can browse our full prompt library below, or check out this video to learn how to implement several of them at once.
🍎 Simple, Personalized Lines for Emails
Summarize a company's mission
This personalization uses AI to summarize the mission statement of any company. This is one of our most useful personalizations and reliably boosts positive response rates.
Many companies employ SDRs to manually write these lines, but you can fully automate this process with AI. This personalization helped one of Eric's clients, a marketing agency, more than double their response rates from prospective clients (from 1.5% to 3.2%).
To get started, we simply input any company description into Clay and use the ChatGPT integration to summarize its mission statement. We specify constraints (e.g., “use less than eight words and use specific words from the company description”). We format our results into the following sentence: “I was on your website, and it looks like you help [X].”
Prompt:
Using the input, summarize the company mission in one sentence. This is the input: “{{Company Description}}”.
Keep the output under 12 words and use keywords from the input. Complete each output with my prefix.
This is my prefix: “I was on your site and it looks like you help”
Here's an example:
The strategy applies to anything you might want to summarize. Another idea would be to use AI to summarize what makes a company different from its competitors. Try it out!
Summarize a Person's Profile
This prompt summarizes someone's profile, highlighting anything that's important to you. This is fun to do in Clay, using GPT3.5 to keep costs down. You can check someone's role history, seniority, current role's focus, and more.
Prompt:
Using the input, summarize the profile in a couple of sentences. The input I will give you is in JSON format. Here is the input: “{{Input a whole social profile}}”. Summarize the profile for me and look out for specific parts about their profile.
Besides giving me a general summary, also tell me:
- Has the person held their current role multiple times or is the first time they are in this seniority?
- What is their current role's focus?
- Do they move jobs often?
For all of this data look at thei profile and think methodically before answering. Summarize the output in 4-6 sentences focusing on what I want to know about them.
Change Messaging Depending on a Person's Role
This is a great workflow if your product has different value propositions for different titles (e.g., growth marketers vs. CMOs).
You can use Clay's persona flow and formulas to say something different but this requires an exact keyword match. You can use this if sometimes the titles change slightly and you want to match up the selling propositions.
The best way to approach this is by using two different AI integrations:
The first one is to infer what the title cares about.
The second one is to write the messaging.
First, prompt AI to make decisions about the problems that a title would be facing.
Prompt 1:
Using the input, write out 2-3 problems a person is responsible for solving in their role. This is the input: “{{Job Title}}”.
Be specific about the particular problems a person in this role faces that a person in a more generalized role would not face.
Keep the output to 2-4 sentences and be very specific about the problems they face.
The output is going to give you a summary of what each particular title would be facing as a problem. We then take that summary and write problem based messaging to everyone.
Prompt 2:
The output we are going to create is a sentence that casually communicates a value proposition given the problems a person faces in their role. The input I'm going to give you is the problems and responsibilities of the role.
This is the input: “{{Summary from last AI prompt}}”. Considering these problems and responsibilities, pick out what problem my company solves that would resonate most with this person.
These are the problems my company solves: “{{Insert your value propositions}}”.
Complete the output with my prefix. This is my prefix: “When I speak to other {{titles}}, they tell me they see problems like”
This would lets you speak directly to the customer relationship management problem they see as it relates to your product or service. You can follow up with how you have helped other companies like them!
Mention a Podcast that Someone Was Recently On
You can achieve this by doing a site search on Listen Notes. First, we can use Clay's Google Search integration to search [listennotes.com](http://listennotes.com) for any podcast mentions of the person's name. The search would look like this in Google.
site:listennotes.com “Person Name”
Then, the Google Search integration will give you data on the podcast name and a little bit about what they discussed. Use those inputs for your prompt!
Prompt:
Using the input, complete a sentence about a podcast this person was on. I will give you the title of the podcast and a small snippet about the podcast. Here are the inputs: “{{Podcast Title}}”. The other input is here: “{{Podcast Snippet}}”. Keep the output under 12 words and use keywords from the inputs. Complete each output with my prefix. This is my prefix: “I saw your podcast about”.
Summarize a Recent News Event for a Custom First Line
Tying outreach to recent events can be a great way to do trigger-based targeting. This prompt summarizes a recent news event for a custom first line.
Prompt:
Using the input, summarize a recent news article in one sentence. This is the input: {{News Article Headline}}”. Keep the output under 12 words and use keywords from the input. Complete each output with my prefix. This is my prefix: “I saw the recent article about”
🔮 More Complex and Creative AI Sales Prompts
Generate Creative Marketing Ideas for Prospects
This strategy uses AI to generate useful marketing ideas for any company. The marketing ideas could include outbound campaign ideas, topics to post about on social media, shortform videos to create, etc., depending on what you want to sell.
Eric uses this strategy to send about 300 emails a day and gets ~3 positive responses for his outbound agency. It's easy to tweak this for anything that you're selling.
In Clay, we used the company description as input and asked ChatGPT to use it to think of three creative marketing ideas that are possible with outbound marketing. We set up an AI assistant so that we could provide up to ten examples of inputs and great answers—and then we were ready to go.
Here's an example email generated entirely with AI in Clay:
Hi X,
It looks like [Company Name] helps manage and protect Microsoft infrastructure. Have you considered running any of the following outbound campaigns to gain new customers?
- Target IT managers involved in cloud migration, offering tailored solutions to simplify their transition
- Reach out to CISO or security managers, highlighting your security and compliance benefits
- Email Microsoft certified professionals, emphasizing your compatibility with Microsoft environments and how you can increase their efficiency.
Are you doing any of these right now, or would you need to help automating or initiating any of them?
This strategy represents a stark departure from conventional outbound prospecting, since you have no social proof, offers, or guarantees. What you do have, however, are concrete and imaginative ideas for your prospect that are relevant to what they're trying to sell or achieve—and an easy way to offer to help.
Here's our prompt:
Using the input, create three bullet points of creative outbound marketing ideas for a company. The ideas have to be possible with cold email marketing. This is the input: [company description]. Analyze who their target audience would be by job title and think of creative ways to market to them. Be specific and give unique advice. Keep each bullet under 15 words and use strategies that use outbound marketing automation.
Watch this video to set it up, step by step.
Generate Ideas for How Someone Could Use a Virtual Assistant
The goal of this campaign is to use AI to summarize the mundane tasks a business owner might have, which they might be able to use a virtual assistant's help with. We also summarize the fulfilling tasks that they get to do. This resulted in a 100% increase in meetings booked for the company that used it!
Here is an example two-step sequence:
Email 1:
Email 2:
We need two AI prompts here. The first one writes a line about the mundane, repetitive tasks a business has to do. The second one writes the lines about the fulfilling tasks that they'd rather be doing.
Prompt 1 (generate mundane tasks)
Using the input, give me three ideas for a small business on how they could use a virtual assistant to help them.
This is the input: {company description}.
Keep the output under 15 words per idea and be sure to output the ideas in one line separated by a comma. Be creative and use keywords from the input.
To help train the AI to give you the outputs you want, use a few input/output examples. Here is an example output we wrote manually to summarize mundane tasks a certain company owner would have: “staying in contact with vendors and collecting quotes for your clients to review, collecting remaining invoices and posting your wins on social media, and helping your 8 senior event consultants move twice as fast by offloading manual tasks and administrative tasks together.”
Prompt 2 (generate fulfilling tasks):
Using the input, create a 6 word output with some of the most rewarding tasks within a business. This is the input: {company description}.
Keep the output under 6 words. Keep the output in all lower case and be specific about the task. Complete each output with my prefix. This is my prefix: "to add some context, I figured you might have parts of your business like”
We used input/output pairs here as well to teach the AI. Here's an example output we wrote manually: “to add some context, I figured you might have parts of your business like adding personal touches that make things feel memorable.”
Change Messaging Based on Company Mission & A Recent Social Media Post
Here's an example of two campaigns we sent to invite folks to a dinner hosted by a business influencer. The AI campaign used less time, sent drastically less emails, and generated a better open rate and more responses.
RESULTS:
Without AI: 473 emails, 80% open rate, 12 replies
With AI: 162 emails sent, 85% open rate, 21 responses
The campaign that used AI even sparked responses from people who thanked us for our email!
Here's the non-AI campaign email format:
Here's the AI email format:
And this is an example of an actual AI-generated email:
This is what the prompt looked like for the first line.
Using the input, complete a sentence summarizing what someone's last social media post was about. This is the input: "{{Social Media Post Text}}" Keep the output under 12 words and use specific keywords from the input. Complete each output with my prefix and do not incude quotations or hashtags (#) in the output. This is my prefix: "I saw your recent post about"
This is what the prompt looked like for the second line:
Using the input, complete a sentence summarizing what someone's company does: "Company Description" Keep the output under 12 words and use specific keywords from the input. Complete each output with my prefix and do not include quotations or hashtags (#) in the output. This is my prefix: "Further than that, I saw how you've been helping people"
Infer What Job Titles a B2B Company Sells To
You can use this prompt to infer what job titles or type of people a company sells to. This works great if your product helps your potential customers to sell to a certain client more effectively.
Prompt:
Using the input, determine the job titles or type of people that this company sells to. This is the input: “{{Company Description}}”. Keep the output to just a list of 3 types of people. Be specific about who they sell to or what their job title is.
Here is an example that has nothing to do with the input, it just is an example of formatting. VPs of Sales, Chief Revenue Officers, and SDR Leaders.
Output who the company in the output most likely sells to. Only name their titles, don't include anything else.
5. Automate Entire Campaigns with Clay
Once you've mastered using any kind of AI prompt, why not automate your entire sales cycle with multi-step email campaigns?
Read on to see three examples of how we've:
- Personalized an entire 4-step sequence within Clay, earning 10%+ positive response rates.
- Automate outreach to prospects based on seven different triggers
- Filter a list of ICPs into 1 of 6 best-fit email campaigns based on their situation</aside>
Personalize an Entire 4-Step Email Sequence
When we worked with Taplio, a tool that helps people schedule posts and announcements, we personalized an entire 4-step sequence using AI. This campaign generated a 10% positive response rate.
Here's a video walkthrough of the entire campaign.
Here's each email of the campaign:
Email 1
We generated this email by scraping their last 3 social profile post results, and then calling out the time difference between post 2 and post 3.
{{First name}} – Saw one of your recent posts about {{AI summary of posts}}. Seems like the last time you posted was {{# of days}} days before that though.
If you're trying to build your brand, have you considered using a scheduling tool?
Our scheduling tool can help write your posts and help you monitor social trends that you can riff on as well.
Let me know if this would be useful to you and I can send over access!
Email 2
This is a follow-up that adds value to the sequence by following up with the prospect's most recent post.
{{First Name}} – I just saw your post about {{AI generation}}. It reminded me to follow up and see if you saw this!
Email 3
Taplio helps create content with AI generated suggestions. We took person's recent posts and their company description to automatically create post ideas that they could bring to their social media.
{{First Name}} – I keep seeing your content online so I made some more post ideas considering what your company does and how you promote it.
{{Creative Ideas}}
This is just some examples of what our AI can generate for you (and you haven't even given it your insights yet!)
Let me know if I could send over some access for you to try it out.
Email 4
Here, we used AI to extract out and normalize their followers so that we can mention in an email.
{{First Name}} – just like how Facebook pages killed their organic reach, it can be expected that your current social media will do the same one day.
I saw your company page had {{# of Followers}} followers and you had about {{# of followers}} followers.
As you are building your following, are you maximizing the platform's organic reach before it gets oversaturated?
We helped companies like Lemlist and Outreach double down on their social content with our analytics, scheduling, lead generation, management and content generation tool.
Could I send you access to try it out?
Reach Out to Prospects Based on 7 Triggers
This workflow analyzes companies for seven different triggers and why we should reach out to them. This automates process automation using a methodology called "Why You, Why Now" from John Barrows. This workflow is great for customer support tech or service companies.
You can see the full process in this video, including how to set each trigger up on Clay, step by step. Once you've set up these triggers, you can use them in any order you'd like to fill out a personalized trigger-based sequence.
Here's each trigger from the John Barrows methodology:
Trigger 1
Look for a new customer support team lead and check what tool they used at their last company. (This wedges you into the company when a new hire might be shaking things up!)
Example of first line:Hey Suchita, noticed you recently were promoted to Principal Customer Success Manager in March. Before working at 6sense, I saw you worked at Yext. Noticed both companies are using Zendesk and Drift right now. How's your experience with them?
Trigger 2
Look for the % growth of the sales productivity of customer team staff. This helps you understand if the team might be outgrowing past processes!
Example of first line:I noticed that 6sense added 20 customer support people in the last 3 months. Are your current processes still working with the team growth?
Trigger 3
Look for new product launches (which always come with new support demands). Use AI to discuss what support questions might come up with the new product.
Example of first line:
I figured customers might reach out for support around things like utilizing the new AI Writer tool in their own own sales process automation work and automation process.
Trigger 4
Look at sales data for companies with no support tech on site. We can even compare this to how many employees the company has and how long they have been in business.
Example of first line (for a different company):
Noticed you aren't using any support tech on the site. I'd imagine that since the business was founded in 2014 and you just broke 100 employees that your old sales processes might not be working any more.
Trigger 5
See if a company is using a competitor's support tech on site.
Example of first line:
Noticed you're using Zoho Support, have you found their multichannel inbox to be useful in support or are things falling through the cracks?
Trigger 6
Notice if a company is hiring for customer support. Use AI to summarize tasks mentioned in the job description that your tool could optimize or replace!
Example of first line:
Noticed you're hiring for as Customer Support Associate. Took a look at the job description and it looked like they will need to respond to emails with appropriate answers to customers. Are you thinking about how AI can be leveraged to speed up this process?
Trigger 7
See whether the company has a generic support email and either no support tech on site OR they have support tech on site (which let's us know that they need to have everything in one place).
Example of first line:
Noticed on the site you weren't using a support software but I found support@domain.com is up there. Are you finding that you can answer all of your support questions without dropping things with just an inbox?
In a follow up email, we could say something like below to keep the personalization going.
"I figured customers might reach out for support around things like leveraging 6sense's AI for creating and converting high-quality pipelines."
Filter Your ICP into 6 Unique Campaigns
This is one of the craziest workflows we've ever built. Here, we used Clay to take one Apollo export and turn that data into six different campaigns depending on industry, employee headcount, technology—with AI lines for each prospect. This was used to save over 6 hours per week of manual effort and there was no loss in positive response rate.
This campaign was selling the very best sales automation software and engagement tools to marketing leaders in American B2B companies with under 100 employees. This video goes over everything in depth!
Step 1: Verify emails
We used Debounce to verify prospects' emails to make sure we can contact them. If we didn't find a verified email, we used Clay's “get email” function to guess our prospect's work email. We put our best results into a master email column.
Step 2: Pull Company Description
We checked if the companies on our list are B2B using the “company description” tag on Apollo. You can also get this information using Clay's “enrich company data” feature. We put our best results into a master “company description” column.
Step 3: Use OpenAI to Check if the Company is B2B
We set up a column using OpenAI using the following prompt:
Tell me if the company from my given input is B2B or B2C. A B2B company is a company that sells to other companies. A B2C company is a company that sells directly to consumers. The input is [insert “Company Full Description” column from table]. Only respond with “B2B” or “B2C.
Nothing continued being analyzed in the table unless it was a B2B company.
Step 4: Sort the Company Into a Campaign
Check the “Industry” column to sort the company into a campaign for management consulting, financial services, and recruiting businesses. If the company doesn't fit any of those four main buckets, we check if employee count is under ten and add it to our Facebook campaign.
Otherwise, we check their website to see if the company has any CRMs we can integrate with or replace. In our final, catch-all campaign, we simply use ChatGPT to draft a custom starting line about the company. We also add a column to check if emails are sent and to add prospects to a running Google sheet of people we've contacted.
Step 5: Refill the Campaign
Once you've set up your campaign, you can import any list of CSVs to add to the current table. Click “add to current table” and carefully map each piece of data that you need for your table to run. Watch the magic happen! (Even if these leads are already in the campaign, Instantly.ai will automatically dedupe them).
With that, you can start with a list of qualified leads, sort them, and automate them into different campaigns with unique copy based, and track the emails you've sent!
6. How Clay Uses AI... Within Clay
For the finale of our guide, we're going to show you how we use Clay…at Clay!
We've used AI to fully automate a 4 step campaign for our own outbound sequences, getting a 5.1% positive response rate.
Every email in the sequence used multiple levels of personalization. This table took under an hour to put together and is easily one of our most powerful outbound campaigns running today.
Let the Numbers Speak for Themselves
Before we dive in to the methodology, let's show you some results:
These stats are for every single email in the sequence. Below, we've included copy of every email in the sequence. Each email was sent 5-7 days after the previous one. We used many layers of personalization, and also made sure to spread them out over time. (One common mistake is to spend a lot of time on personalization and relevance, but to use it all on the first email.)
Email 1️
{{First Name}} – I was on your site and it looks like you sell to {{job titles}} titles. It looked like you are helping {{AI summary of company mission}}.
I had to assume you are using outbound tactics considering {{Full Name of SDR}} is an {{SDR/BDR}} on the team.
Are you thinking about how they could be leveraging better data and GPT-4 to do their prospecting?
The reason I ask is because we are helping companies like Rippling and Ramp with our tool that collects the data and writes the messaging for them.
Would this be worth a chat to build a custom outbound table for you?
Email 2️
{{First Name}} – as you're helping people {{AI summary of company mission}}}, I'd imagine you are using outbound marketing tactics like the following.
{{Creative Ideas}}
Have you cracked how to do this persona targeting and data enrichment automatically or is your team still using tools like Sales Nav to do manual research?
If so, would it be worth taking some time to show you how we can automate persona and account research for your outbound team?
Email 3️
{{First Name}} – I sent an email a while back with the subject “Outbound edge” that probably didn't do a good enough job showing you how we could help your outbound team.
Our tool combines over 40 databases into one spreadsheet so that your team can see all the benefits of our sales automation technology with the data in front of them instead of manually researching accounts.
For example at your company
- It looks like your sales team has seen a {{percent change}}% change in the past 3 months.
- Your company account has gained {{# of Followers}} in the past 3 months.
- The company looks like it has {{# of sales people}} people with “Sales” in their title based in the United States today.
Besides things like this, we also monitor for recent news, open job positions, tech on site and mentioned in job descriptions, and a lot more.
On a call, we can build a custom workflow for your team to automate their outbound research. Wanna make it happen?
Email 4️
{{First Name}} – Perhaps this is not your area of focus. Should I potentially reach out to {{Other sales rep/marketing person in the company that's NOT them}} given their role as a {{Job title cleaned with AI}}.
💫 Recreate this campaign in Clay
To recreate this campaign within Clay, just watch this video!
Here is each step we go through in the video:
Video: Email 1️
- Import your data or lead list
- Verify email addresses
- Use “Find Work Email” integration to generate a master email column
- Normalize company names
- Get company descriptions
- Use AI to figure out who the company sells to
- Prompt: Using the input, what types of people or specific job titleswould a company sell to? This is the input: [ ] Who gets the most value orproduct of the service add specific job titles or a type of person return up tothree job titles do not include any numbers or extra information an examplewould look like this VPS or sales and two like all that stuff right and sothey sell to CEO CFOs and chief investment officers they sell to contentcreators filmmakers and video editors
^ All of that creates our first line: {{First Name}} – I was on your site and it looks like you sell to {{job titles}}. It looked like you are helping {{company mission}}.
- Use the “Find People” enrichment to find the full name of the SDR or BDR on the team
In our formula, if someone title includes “sales” we call then an SDR; if it contains “business,” we call them a BDR. If we can't find either, we don't reach out to the company at all. We'd rather send fewer emails that land with hyper targeted personalizations.
^ This is the ingredient for the next part of the email: I had to assume you are using outbound tactics considering {{Full Name}} is an {{SDR/BDR}} on the team.
Video: Email 2️
- Generate and insert creative outbound campaign ideas for these companies.
Using AI, we generate outbound ideas. As an example from an education company, we generated:
- Targeting School principles and district manager administrators highlightingstreamlined remote learning with class tools on Zoom
- Reaching out to university educators showcasing classes for effective higher education virtual classrooms and lectures
- Connecting with professional trainers offering online courses promoting productivity and engagement tools for learning
Our email uses these ideas to ask:
As you're helping people {{company mission}}, I'd imagine you are using outbound marketing tactics like the following.
{{Creative Ideas}}
Have you cracked how to do this persona targeting and data enrichment automatically or is your team still using tools like Sales Nav to do manual research?”
Video: Email 3️
In this email, we get use the Harmonic integration to get information about the sales funnel, team growth and company account growth (remember to normalize all percentages!) Lastly, to get the number of salespeople at the company, we use the “Employee Headcount by Title” integration.
We only email the people for whom we can find all this data. We'd rather send fewer emails, but make them the best, most personalized emails that we can come up with!
Video: Email 4️
For this email, we want to find other top sales reps and marketing people at the company to reach out to. We insert a bunch of example titles (such as “Director of Marketing,” “Director of Customer Initiatives,” etc) and exclude the ones we don't want (such as “development,” “representative,” or “engineer.”)
💡 For more tips on using AI in your outbound campaigns, sign up for our newsletter!
.avif)
%20(1).avif)
.avif)
.avif)
