Using AI for Lead Generation in Finance: What Public Filings Can and Can't Tell You

Most lead gen advice starts in the middle. It assumes you already know who your buyer is and just need more of them, cheaper.

There’s a failure that happens earlier, and people don’t talk about it much: you have an offering and you genuinely do not know who to contact for it.

The problem

A small finance company raises capital for investment strategies. Identify institutions that could allocate, get someone on the phone, have the conversation. That worked for years.

Then they took on a concentrated U.S. micro-cap equity strategy, and the process stopped producing. Not in the usual “conversion is down” way. For this one offering they had no reliable way to generate leads, because they didn’t know who to contact in the first place.

That’s not a lead shortage. That’s not having a definition of a lead yet.

Get the definition right first

Alex Hormozi has the cleanest version of this I’ve found:

• a lead is a person you can contact
• an engaged lead is a person you can contact who has shown interest in what you sell

Everybody wants the second one. You can’t skip the first. Without contactable people there’s no outreach, no campaign, no funnel. Nothing to point the machine at.

So step one wasn’t a campaign. It was a research question: which U.S. institutions could plausibly buy a concentrated micro-cap strategy, and who works there?

That’s market research. For a long time it meant either paying $16,500 a year for a database subscription, or handing an analyst a spreadsheet and a few months.

Why finance

Because in finance, the answer is often already published.

Institutions that manage money have to report to the government. That’s just how securities regulation works. And it means the raw material for a prospect list is free, public, updated quarterly, and mostly ignored by sales teams.

Every institution managing over $100 million in U.S. stocks files a quarterly 13F showing what it holds. The SEC publishes them in bulk: 8,741 filers in a single 99 MB download. Most of those were asset managers, not the asset owners we actually wanted, so we filtered them out using another SEC dataset, the adviser registry. That removed 6,791.

Then the part that actually mattered. A 13F shows what an institution holds. A second SEC file maps holdings to issuers. A third maps issuers to public float. Chain those together and we can see, from public data, whether an institution actually buys small companies, not just whether it says it does.

The IRS gave us a second universe. Private foundations file a 990-PF: total assets, itemized stock holdings, who they pay to manage the money, officers and trustees, and the signer’s phone number. We pulled 5.3 GB of those filings and parsed 1,609.

About 45 minutes of compute. Zero dollars spent on data. Out came 809 institutions and 4,697 named contacts, each tied back to the filing it came from.

What we actually got

Leads. Not engaged leads.

Nobody on that list had raised their hand. They’d been identified as able to buy, which is half the definition.

And it’s thinner than 4,697 makes it sound. Of those, 4,495 are inference: foundations whose tax return names a paid investment manager. That tells us they allocate through managers, which is how they’d buy this strategy. It does not tell us they want it.

The rows with position-level evidence came to 202 contacts across 33 institutions.

Endowments produced zero — the ones large enough to file a 13F held funds and large caps, not the individual small companies we needed as proof. Pensions produced zero too.

Single family offices are the strange one. On paper, they’re probably the best-fit buyer. The SEC’s own estimates say there are thousands of them. In free public data, a few dozen are visible. That’s it.

They mostly don’t have to file, so they don’t.

Version one. Not solved.

What it changed

Before this there was no real way to generate leads for the offering. Afterward:

• 33 institutions we could prove hold the kind of stock this strategy buys
• named people and direct phone numbers for them
• a larger inferred tier to work through
• clarity on which channels free data will never reach

That changes the job. Now the problem becomes turning leads into engaged leads. That’s outreach, positioning, and having something worth responding to. None of it could start until someone answered the basic question: who do we call?

What generalizes

If you sell into a regulated industry, your prospect list usually has a public shadow. Finance is the extreme version, but healthcare, insurance, construction, and government contracting all leave trails too.

Before you buy a database, find out which public record it was built from. Sometimes the vendor really does have something you can’t get on your own — “who is actively searching right now” is real information, and no filing is going to hand you that. A lot of the time, you’re renting a cleaned-up rearrangement of public data.

Build it once yourself anyway, even if you end up paying later. Not because free is automatically better. It isn’t. Building it is how you learn which parts of your market are visible at all.

That was the real win here: we came out knowing exactly which two channels need a different approach, instead of spending a year on a subscription before realizing it was never going to cover them.