Learn and master Clay with Clay university

Next, you'll watch an AI agent build two enrichment workflows from one prompt, one for PE firms and one for their operators, and you'll learn how to read what it built before you trust it.

🍵 From FETE to Target, Enrich, Act

If you took the old Clay 101, you learned FETE: Find, Enrich, Transform, Export. Those steps still happen, and they're still a good way for a beginner to think. In the agentic era, a tighter framework is: Target, Enrich, Act.

• Target replaces Find. We're targeting a list, like the search plus segment we just built, and that list is dynamic: always on, refreshing on its own.
• Enrich combines enrich and transform. One prompt from the CLI can do both, in one workflow. The nodes still differ: transform steps look like code, normalization snippets, or regular expressions, and enrichment steps look like data provider waterfalls and AI web research.
• Act replaces Export. Export mattered when Clay was never the place you acted on your data. You can still send data to outside tools, but more of the action now happens inside Clay, and keeping your data there means more context for every play after this one.

🎯 The Strategy Before the Prompt

Before prompting anything, decide what the play needs. For these PE firms and their operators:

• Basic enrichment on every company and person
• Portfolio size: how many portfolio companies each firm has
• Portfolio mix: what share is tech or SaaS (where we're strongest), and what share is healthcare, real estate, or finance (newer verticals)
• Existing customers: are any portfolio companies already customers? That makes a partnership pitch much stronger
• GTM research: what the firm has done to operationalize sales, marketing, and ops at its portfolio companies

All of it goes into one enrichment workflow per side of the play.

💬 The Prompt

Here's the prompt from the video:

I want you to help me build workflows for these new companies and people segments on the PE side.

On the company side, I want to first check:
1. Do they have a U.S. headquarters? I see your message up above that we were doing U.S. office, not headquarters. I'd like to keep it headquartered in the U.S. and the people also based in the U.S.
2. I'd like to see what percentage of their portfolio is in a tech or SaaS industry, and what percentage is in a real estate, finance, or healthcare industry. Give me the percentage breakdowns and an understanding of what "other" is. Label any industry that's greater than 15% of a company's portfolio.
3. Check whether any of their portfolio companies are already existing Clay customers. Use CRM data, or find the information on the web if the CRM connection isn't live.
4. Return the total portfolio count for each firm.
5. Research what this firm has done around GTM operationalization of its portfolio companies: sales, marketing, ops, customer success.

On the people side: basic enrichment, plus work email using a waterfall, with a Claygent as the last step that finds emails based on patterns if the waterfall finds nothing. Also find anything GTM-relevant about these people: interviews about sales, marketing, or GTM optimization, and any GTM experience in a prior role, in a separate field.

For both flows, understanding the play I'm trying to make, ask me any questions that will better help you help me before getting started.


⚠️ If you're new to this, you don't have to start with a mega prompt like this!

You can start with something simpler, and go back and forth with the AI, or lean on a friend or on experiments to develop your taste:

I want to go after portfolio companies because we're building a PE channel partnership play. I want to look for warm intros, figure out what unique data points I can use to strengthen the proposition for partnership, and get your help fleshing out my strategy. Ask me any questions to better help you help me before getting started. Give me at least three to five creative ideas of what I could do from a strategic perspective.


Claude's Questions

📝 Test before you run everything, because it's always smart from a credit perspective. And also, connect your CRM in your own workspace; we’ll skip that for now to keep customer data off screen.

🤖 A Quick Look at Claygents

While the build runs, look at Claygents in the sidebar. Claygents have gotten much better.

Think of one as a small computer you can run in the cloud: you define its outputs, and you can give it skills, web search, tools, account context, and business context. You'd normally build one from the CLI or with a prompt.

We’ll go deeper on these in a bonus lesson.

📖 Reading What the Agent Built

The build took Claude about ten and a half minutes, because it tested its own work and fixed what failed. Read the agent's summary before you open anything. Plenty of people skip it and just say "it worked."

The summary is your first hint at what happened, the work itself is the next one, and the results are the final outcome.

Here's what the log showed, in order:

1. No CRM connection, as discussed
2. New Audiences fields for the outputs, reusing fields the people side already had
3. The company workflow: the ISO country code from enrichment powers the US HQ check, and Claude built the Claygent steps on GPT-5.4 nano, the model the skill recommends
4. Testing and fixing: the portfolio step only counted the first page of one firm's portfolio, so Claude tightened the prompt and moved to a stronger model
5. The people workflow: a lookup into Audiences, a Claygent research step, the Work Email waterfall, a branch, and a Claygent email-pattern fallback, then write-back
6. Two new segments that fill as the workflows run

Two takeaways before going further: check how well the portfolio Claygent is really performing and which model it uses, and remember that a CRM connection would let you skip a Claygent for the customer check.

Then ask for the links: "Can you link me both of the workflows?"

🔎 Inside the Two Workflows

The Company Workflow (U.S. PE Firms – HQ, Portfolio Mix & GTM Research)

• Segment trigger: the U.S PE companies segment, loading the right number of records
• Enrich company: Clay's native enrichment action
• US HQ check: a code node. You don't have to read every line yourself. Here it takes the enrichment as input, reads the country or locality, strips whitespace, and writes the HQ, its source, and a label. When you have time, paste code like this into your AI and ask what it does.
• Write HQ back to Audiences, because the prompt asked for it
• Portfolio research: a Claygent, still on GPT-5.4, which you can prompt the AI to change
• GTM value creation: a second Claygent, then write-back

Checking the Test Runs

Open the two test runs, then click into one record's run details. The portfolio-count Claygent call still wasn't capturing every active portfolio company. The GTM value-creation call (look for the blue icon on the node) came back with good sources.

The People Workflow (U.S. PE Operators – Enrich, Email & GTM Research)

• Enrich person, then code that identifies the person's PE firm
• Look up the firm and its HQ flag in Audiences
• GTM research on the person
• Work Email waterfall, Clay's native one, with multiple email providers in sequence. Open it and confirm it's using as many providers as it should.
• A regular expression that captures the email from the waterfall's output
• A Claygent email-pattern guess as the last source, then write-back

Publish, then run. Do the same for the company workflow.

🔁 Ask the AI to Improve Its Own Work

While the people workflow runs, go back to the agent and ask for anomalies and improvements:

Could you help me investigate model quality issues? Could you help test on a couple of stronger models? I'm okay with using something like Opus 5.5 as a daily driver for this. If it returns better results, please make sure that the list of portfolio companies we're collecting is exhaustive, or as close to exhaustive as possible, via the methodologies you're using. Don't worry if it's not accurate, we can add in data on the customer side as well.


⭐ Looking to build IRL? → If you want to build the next play like this with a group and someone to ask, that's what cohorts are: taught live, with other people building at the same time. 100% optional but super helpful for taking your skills to the next stage.

🚀 What's Next

You've now built enrichment for both sides of a play from one prompt, and you know how to read the build, the nodes, and the run details before you trust the results.

Next is the last step, Act: sending the data out, syncing it to ads, and running it through the Sequencer.