The 7-Step List Build That Works

1. Start with the right brief

Most firms start sourcing with a sector and a revenue and/or headcount range. That’s not enough.

We work with clients to define the true must-haves vs. nice-to-haves, then translate that into actual search logic across multiple platforms.

Good inputs = better discovery.

2. Build the base list
We typically start with Grata, Sourcescrub, and Apollo.
Each has strengths and blind spots so we often use them in combination to build the initial universe.

3. Layer in deep research
This is where tools like Perplexity (in Deep Research mode) add value.
We use it to surface companies that don’t appear in databases (but are still squarely within scope).

Here’s an example of a real starter prompt:

You are a researcher at a mid-market financial advisory firm.
Build a list of privately held last-mile food distribution companies in the US.
Exclude companies with over 500 employees or recent PE backing.
Prioritise those active post-2020, especially in the Northeast and Midwest.

This kind of input gives us a fresh perspective on adjacent players and fills in the gaps missed by traditional search tools.

4. Run lookalike searches (intelligently)
We use Clay to find lookalike companies — but only run lookalikes on one company at a time.
If you batch multiple companies, the logic tries to average everything and the results drop off fast.

This helps us expand around confirmed winners, without drifting off-thesis.

5. Use AI for first-pass filtering
We use ChatGPT-4o, Clay, and custom logic to triage large lists.

But we don’t trust the first run.

We start with a batch of 30 companies, run a manual pass alongside, and compare.
If the AI misses clear fits (or misclassifies irrelevant ones), we refine the prompt.
Then (and only then) do we scale the run.

6. Apply human QA
Every “yes” and “maybe” still gets reviewed by a real person.
Is the company still active?
Does it actually align with the thesis?
Would a principal take the meeting?

This is where precision kicks in…and where AI has to hand over the reins.

7. Enrich, validate, and upload
We pull contact data from Apollo, ZoomInfo, and LinkedIn (or use Clay as an aggregator), then validate and clean it before uploading into Salesforce.

Every record is tagged, segmented, and CRM-ready.

But Doesn’t AI Do All This?

Not really.

Yes, we use AI across multiple points in the workflow:
- ChatGPT-4o for classification and prompt logic
- Clay for enrichment and filtering
- Perplexity for expanding the search universe

But it doesn’t replace structure, iteration, and judgment.

If you skip the QA, skip the pilot batches, and skip refining the brief, you’ll end up with a list that looks right in theory…but fails the second a deal team picks it up.

Our Go-To Stack

1. Grata or Sourcescrub for initial discovery
2. Clay for 1:1 lookalikes and contact enrichment
3. Perplexity to expand the surface area of the thesis
4. ChatGPT-4o for classification and filtering
5. Manual QA
6. Push to CRM

This gives us speed, coverage, and control
without drowning in noise or wasting expensive time.

The Takeaway

You don’t need more tools.
You need a tighter process.

And you need people who know where to trust AI — and, more crucially, where to override it.