Before you pay for a lead database, spend a week building thirty names by hand. Not as a warm-up exercise. As the actual test of whether the segment in your head is real, answered with public information that costs nothing but attention.
Start with the buyers who already told you something
Any startup with even a handful of users has a small pile of unpaid research sitting in a support inbox or a signup form: the person who asked a sharp question before buying, the one who churned and explained why, the one who emailed asking for a feature that already exists. Read that pile before you read anything else. It is the only buyer language you did not have to guess at, because a real person under no obligation to you chose those exact words.
Then look for more people who sound like that. Not a lookalike model, a plain search. If three churned users all named the same competing habit or the same internal spreadsheet they were replacing you with, that phrase is now a search term, and searching it in public places surfaces other companies with the identical habit.
A job posting is a dated, specific signal a filter can't produce
A company hiring for a particular role is announcing, in public, that a problem it doesn't currently have covered is about to get budget. If what you sell is something a first ops hire, a first support hire, or a first person running outbound would reach for on day one, that posting is a better signal than almost any headcount filter, because it is timestamped rather than a static snapshot. Read the posting itself, not just the title. It usually admits what the company is currently bad at, in its own words, and that sentence belongs in your email more than anything you would have invented on their behalf.
Somebody else's customer page is a partial map of your market
Any product solving a problem adjacent to yours, rather than identical to it, has usually published a page of customer logos and a few case studies. Read the case studies, not just the logos. They describe, in the customer's own words, the situation that pushed them to buy something. Where that situation overlaps with what you solve, you have found a company already primed for a category of tool, and often a named, quoted person with a title attached.
The same trick works on review sites built around an entirely different category of software than yours. Reviewers list their job title and company size next to a paragraph explaining exactly what was broken before they bought whatever they're reviewing. None of that costs anything to read, and it hands you the target and the opening line in the same paragraph.
App marketplaces and integration pages sort a list for you
Plenty of software publishes a directory of the other tools its customers connect it to, or a marketplace listing showing who has installed an add-on. Somebody who installed a scheduling plugin, a payments add-on, or an inventory sync has told you, without knowing it, something true and specific about how their business runs. That is a filter no headcount range gives you, and it's sitting in a public directory rather than behind a subscription.
Communities teach you the words before they teach you the names
A narrow forum, a niche group chat, a subreddit built around one job function: lurking there for a week does two things. It shows you real people complaining about the real problem, in language you can borrow honestly instead of paraphrasing into marketing copy. And it exposes which of your assumptions about the buyer were wrong, before you have written to a single one of them, which is a far cheaper place to be wrong than three weeks into a sending schedule.
Doing it by hand is the point, not the inconvenience
Thirty names built one at a time forces you to actually read each one: the company, the role, whether the story you told yourself about your buyer survives contact with an actual person. Mistakes in targeting tend to surface around name twelve, when you notice half the list doesn't have the internal function you assumed existed everywhere. A list of five thousand bought on day one hides that same mistake inside a number that feels like progress. You find out later, from silence, which is a far more expensive way to learn it, and a worse one for whatever sending domain you burned finding out.
Know when the hand-built list has finished its job
Once thirty names produce a handful of conversations you'd honestly call promising, you have evidence a segment exists, not just a hunch you're fond of. The next question is whether it's large enough to sustain a year of outbound, and that's worth checking before a cent changes hands. Most tools in this category, ours included, let you turn a plain description of that same buyer into a count and a look at the next batch of names for free, so you find out whether the market behind your thirty is forty companies or forty thousand before any credit moves.
That's the honest order: prove the reason by hand, check the size for free, and only pay to reach people once both steps have told you it's worth reaching them at all.
This will feel slower than it is
Thirty names by hand sounds like a week you don't have when there's a product to ship. Weigh it against the alternative honestly. A purchased list arrives fast and then costs you the following month discovering, one bounce and one blank reply at a time, that half of it was never the right audience. The hand-built version front-loads the discomfort into a single week instead of spreading it across a quarter disguised as a sending problem. Founders who skip this step aren't saving time. They're borrowing it from later, at a worse rate, once a domain reputation is attached to the mistake.
What a subscription buys that a week of reading can't
None of this argues against ever paying for data. Scale is a real thing worth buying once you know what you're scaling. The argument is against buying it on day one as a substitute for looking, because a subscription answers "how many" long before it answers "are these actually the right thirty," and the second question decides whether the first number was ever worth having.
A startup that has done the hand-built version first walks into that purchase knowing exactly what it's asking the subscription to confirm: not "who might buy this," but "how many more of the thirty I already understand are out there." That is a much smaller, much cheaper question, and it's the only version of the purchase worth making.