AI agents: a practical application for marketing & sales
AI agents sound great in theory. Here's a real world application of them
Introduction: a practical application for AI agents
I recently wrote about AI agents, and I suggested that Unix pipelines are a good framework for thinking about them. AI agents basically let you pipe data from one system to another, manipulate the data in that second system, and use the output as input for yet another process. And in principle you can chain these agents infinitely.
But what is a practical application of this? It’s great to theorize about similarities between AI agents and Unix pipelines, but theory doesn’t do much for us in the real world. How can you use AI agents to make your life more productive? How can you use AI agents to make more money more quickly? Well, I can’t answer these questions for you, but I can present a practical application of AI agents which should help you think about how this technology works, and how you might apply it to your own use cases.
We’re going to use an AI agent to prioritize sales leads. To do that, we first need to capture the sales leads. A very high level overview of this process is:
Build a content marketing engine to capture sales leads
Enrich the sales leads data
Feed the enriched data into an AI agent to run a scoring algorithm on the sales leads
Sell to your highest priority leads
Retire to a grand estate on St. Barts
An important aside before we continue: in this post, I reference, and link to, a number of companies’ products. I mention and link to these companies’ products only because I am familiar with them—there are many different products out there which will allow you to achieve similar functionality to what is described in this post. No endorsement is implied, and I’ve no affiliation with any of the companies mentioned in this post.
Additionally, there are a number of constraints and considerations in what I propose below, which I discuss in more detail at the end of the post.
Build the content marketing engine to capture sales leads
To capture sales leads we’re going to build a content marketing engine for LinkedIn. This content marketing engine is adapted from one that I found here.
Here’s how the content engine works for our B2B SaaS company:
Develop an ideal customer profile (ICP), so we know who we want to sell to.
Identify the relevant subject matter expert (SME) at the company. For an early stage B2B SaaS company this is usually the CEO. For a later stage company it might well be someone other than the CEO. But the general idea is that you want to identify the person in your company who can speak most authoritatively about the pain points that your company solves for your customers.
Develop a content plan that piques prospects’ interest:
Product: What pain points does your company’s product solve for your customers?
Thought leadership: What topics do your customers care about?
Do a video interview with the SME on a regular basis.
Prepare questions ahead of time, ensuring that they focus on Product and Thought Leadership.
Use a tool like Descript to record the video and create a transcript. Descript allows you to slice the interview into bite-sized chunks for easy viewing.
Post bite-sized chunks of the interview to LinkedIn.
Track performance of these posts using a tool like Shield.
Use your best-performing posts as targeted ads on LinkedIn: this is where you source leads from. The ads will point the lead to a registration page for you to capture email and other contact information. Use a multi-layered ad targeting approach:
top-down job title targeting
bottom-up job function targeting
account-based targeting
And that’s the content engine. It can run for as long as it is productive. And because you quantitatively track the performance of your posts with a tool like Shield, you will know when (not if) LinkedIn’s algorithm changes to the point that this is no longer an effective lead generation strategy.
It’s important to note that if your target market is a vertical niche, you may want to forgo job title and job function targeting, and instead simply target the companies that operate in your market.
Enrich your lead data
Now that you have leads coming in, you want to learn more about the leads and their employer: how large the company is, where it is located, what industry it operates in, etc. You want to know all of this information because you want to prioritize leads which match your ideal customer profile (ICP). Developing your ICP is the first step that you do for the content engine, for which see the previous section.
One way to enrich your lead data is with Clearbit. Here’s a list of all the data attributes that Clearbit claims you can use to enrich your lead data. The particular attributes that you are interested in will depend in large part on your ICP. If you’re targeting a broad market of customers, your ICP will describe a high level view of your ideal customer. If you’re targeting a niche customer, on the other hand, your ICP will be very specific and granular. How you want to enrich your data, then, depends vitally on the scope of the ICP and the breadth of the market you’re pursuing.
Fire up your AI agent and score your leads
Here’s where our AI agent comes into play. We’re going to take our enriched data and run it through an algorithm to score leads by priority. We’re going to use AgentHub’s tools to build this AI agent. AgentHub’s AI agents are basically flow charts, as you can see from this screen shot:
If you want to interact with this template, register for a free account. The template is available here.
In order to make sense of this flow chart, and to understand how the logic for the AI agent works, we’ll start at the top and work our way down:
The user uploads a CSV with enriched data on sales leads. The CSV includes: company name, # of employees, industry, contact name, contact position, contact email. Note that the template can be modified per your requirements; you don’t need to use the canned data structure.
The data are passed to GPT-4, and the text is parsed using the following criteria (also known as a prompt):
Small to medium sized companies (medium is ideal)
Companies in primary resource heavy industries are ideal, ex. manufacturing, agriculture, etc.
The higher up the contact position, the better. VP or Executive level is preferred.
Again you can modify these criteria (prompt) to suit your needs.
You can also change the model from GPT4 to another one.
Each lead is assigned a score as a result of the text being parsed by GPT-4, and this score is appended to the lead data.
A new file, with scored leads, is generated.
You could extend this AI agent by combining it with other agents that AgentHub offers. For example, here’s their sales email agent.
Bringing it all together
Let’s review what you’ve done:
Built a content engine on LinkedIn
Acquired a steady pipeline of leads
Used an AI agent to score the leads by priority
Received a list of scored leads
Now your sales team is ready to pursue the high priority leads first.
Constraints and considerations
The process that I’ve outline in this post is a highly idealized one. In order for it to work, it depends on a number of assumptions, some of which may not hold for your circumstance. Therefore, I’d take stock of the following considerations, and adjust accordingly.
This process requires that your marketing group and your sales group have an effective working relationship. To say that this is not always the case is to traffic in understatement. This kind of process only works if your sales team is receptive to receiving scored leads. If they prefer to chart their own course, you’ll find getting them to work in concert with you is a hard road to hoe.
This process assumes that your prospects are active on LinkedIn. While this assumption holds for many markets, it does not hold for all.
This process assumes that Clearbit’s data enrichment is more or less accurate. Depending on the particular market you serve, this may not be the case.
🍪