I built my first agent (I’m a marketer)

Lead enrichment agent. Step-by-step walkthrough. Prompts included.

In today’s newsletter, Eric (Keyplay’s marketing lead) is sharing a step-by-step walkthrough on how we created our first marketing agent.

Want more examples of real AI use cases in marketing, subscribe!

---

We replaced our whole marketing team with 47 AI Agents who created $50m in pipeline in 6 months!! 🎉

Just kidding LOL.

There’s a lot of hype on LinkedIn about AI Agents. Rightly so. They’re giving marketers who understand them massive leverage.

This newsletter isn’t about replacing your team with agents or driving huge pipeline numbers.

It’s about my real journey building one agent that we actually use. And how you can do the same (without hitting the dead ends I had to go down to get to something that actually works).

Step-by-step walkthrough below. Prompts included.

Automating Call Prep: Our Pre-Call Lead Enrichment Agent

Our AE, Katie, was spending ~15-30 minutes/call manually researching to prep for demos. At our scale, that meant she was spending dozens of hours a month doing call research.

With this agent, she gets all the information she needs (and information she may have missed before) in a couple of clicks.

Building this agent boiled down to three parts: data, analysis, and delivery. I’ll walk through each piece in the following sections, or you can watch this quick video walkthrough I recorded.

Full video here.

Data: Company Information & Scoring

I used Keyplay to get clean account-level data, a fit score, and to answer more qualitative fit questions with their AI Agents.

*Obviously, we already have our ICP Model set up in Keyplay. If you want to get yours set up, start here: How to Build an ICP Model and AI Agents for Account Research.*

I built account research agents in Keyplay to get us:

• Target market type. What type of market does this company sell into? (broader, more complex markets is better for us).

• ABM Maturity. Based on the tech this company uses, their team dynamics, and other details we can find online, how mature is this organizations ABM function?

• RevOps & AE team size. Signals of RevOps maturity, and how sales-led the company is.

Once you’re set up with Keyplay, you can use Zapier to get all of this data on command.

Analysis: AI + Data = Insights

Keyplay is an ICP Modeling & Targeting company. We help companies stop wasting spend on non-ICP accounts.

So, the more deeply Katie can understand a company’s ICP before the meeting, the more valuable the call will be for everyone involved.

So, I layered on some AI to help Katie better grok their ICP more quickly. We had chatGPT get us:

• Simple company description. Explain what this company does to me like I’m a 5th grader.

• Use Cases & ICP. Katie needs to understand a company’s ICP before the call. ChatGPT gives her a starting point.

• Case Studies. Case studies are a super helpful way to understand what the company actually does and who their best customers (who they’ll replicate with Keyplay) are.

• AI Agent Ideas. We build account research agents for companies. I had ChatGPT suggest some ideas. While these are not perfect, they are a nice starting point.

Delivery: Wiring it all Together

After some iterations, we landed on using Zapier for all of this (simple, flexible, and reliable).

We found that Katie would be doing two jobs with this data:

• A) quickly reading the fit score/company details to determine quality.

• B) doing deeper research to understand the company’s ICP.

For A, piping this data into our existing #demo-requests Slack channel was perfect. For B, we found that a project actually made more sense because she could ask follow-on questions and go deeper.

I built her a simple Claude project that she can chat with. All she has to do is say, “tell me about [company domain].”

My wishlist of agents

Naturally, this sparked a lot of ideas. I figured I’d share a few. Let me know if you want me to build any of these (or other ideas!).

Takeaways & Learnings

A couple of my favorite takeaways from diving in and actually building something:

• So many dead ends. It takes more work than you think to get something, even something simple like this, to work. Lots of edge-case management & prompt tweaking.

• AI automation should be used as a starting point. Automations are rigid. So, it’s only worth automating processes when they’ve proven repeatable. Anywhere there’s variance, or more human judgment, a simple project in Claude might make the most sense.

• Data = alpha. Trash in, trash out. We need better inputs if we want great outputs.

It was fun building my first agent! Let me know what AI in Marketing use case you want me to explore next.

- Eric