Evaluate deal sourcing software on what it does for the companies in your mandate, not on the size of its database. Eight questions settle it on a demo: coverage below the transaction floor, where the financials come from, contact accuracy, what happens after export, seats versus team access, refresh cadence, what the score predicts, and who makes the call.
A software demo in this category is built to show breadth. The vendor searches a sector, the map fills with dots, the count in the corner says several thousand. Your mandate is not several thousand companies. It is a few hundred founder-owned businesses in three or four sectors, most of which have never raised, sold or filed, and the question that matters is how many of those are in the product and how much of what it says about them is true.
This post is the checklist for that question, and it is about the software seat specifically. The wider choice between the five ways to buy sourcing, from aggregators and intermediated networks through to a mandate-driven team, is laid out on BizNexus's comparison of deal sourcing platforms by model; this post assumes a data seat is already part of the answer.
Each question comes with the answer that should worry you. Grata and PitchBook are named where their own published material is the source, with the date we read it.
1. How many of my companies are in here?
The headline number is the easiest thing to find and the least useful. Grata describes a universe of 22M+ private companies including bootstrapped and founder-owned businesses (grata.com, observed 2026-09-20). PitchBook publishes coverage of 12.9M+ companies (pitchbook.com/data, observed 2026-09-20). Both are true, and neither tells you anything about a $3M EBITDA HVAC contractor in Ohio.
What to ask: how many companies in my sector, in my geography, below my EBITDA ceiling, and how do you know? Grata publishes no breakdown of its coverage by revenue or EBITDA band (grata.com/pricing, observed 2026-09-20). PitchBook's own comparison pages describe its coverage in terms of transacted and sponsor-backed companies.
The answer that should worry you: a total. If the vendor cannot produce a count for your band, produce it yourself on a trial. The method is in the last section.
2. Where do the financials come from?
A US private company that has never filed anything has no reported revenue. Whatever figure sits in the record is an estimate, modeled from headcount, web footprint, sector ratios or some blend. It means the number has an error band, and the vendor either knows what it is or does not.
Grata lists EU reported financials as a distinct feature (grata.com, observed 2026-09-20), which is a useful tell: where filings exist the product uses them, and where they do not, it models. For most of a US lower middle market mandate, it models.
What to ask: how was this revenue figure derived, and what is your measured error against companies whose revenue you later learned? Then check twenty companies whose numbers you already know.
The answer that should worry you: "proprietary model," with no band. An estimate without a stated error is a guess with a decimal point.
3. How accurate is the owner contact, and when was it verified?
For a founder-owned company the CEO field is the owner or it is nobody. There is no investor relations desk, no second address to try. A contact record that is eighteen months stale is a bounced email and a wasted week.
Data accuracy is the most-cited complaint in Grata's G2 reviews, at 18 of 79 as tallied by a third-party review site (prospeo.io, observed 2026-09-20). That is a third-party count, not a verdict, and the same complaint trails every product in the category.
What to ask: what is the date on this contact record, and what share of exported owner emails bounce on first send?
The answer that should worry you: "we verify continuously," with no date visible on the record. A process you cannot see is a process you cannot audit.
4. What happens after export?
This is the boundary of the product, and the vendor will usually be honest about it if you ask plainly. The seat ends at the CSV, or at the CRM sync, which is a CSV with a nicer pipe. What happens next is an email from a real domain, a follow-up on day five, a judgment about whether the owner is open or being polite, and an NDA ask that somebody owns. None of that is in the product.
The sibling post on data tools versus sourcing services takes that line apart with sources. The short form: a data tool sells information, a sourcing service sells capacity, and a firm short of the second routinely buys the first.
What to ask: after we export, who sends, from what domain, and what happens on day five when nobody has replied?
The answer that should worry you: "it integrates with your CRM." That is an answer to a different question.
5. Is it a seat or team access?
The quoted seat count is rarely the working one. The partner who sat through the demo is not the associate who will run the screen every morning.
Morningstar's FY2025 10-K describes PitchBook's pricing as primarily based on the number of user seats (filed 2026-02-13, observed 2026-09-20). Grata sells three tiers, all demo-gated (grata.com/pricing, observed 2026-09-20). Neither vendor publishes a price. The only dollar figures on Grata's own site are its referral page's worked examples of a $20,000 contract and a $30,000 contract (grata.com/referral, observed 2026-09-20). Third parties report a single PitchBook seat at around $12,000 to $20,000 a year (startupyeti.com, 2026-04-28), unconfirmed by the vendor.
What to ask: how many seats does a firm of our size typically run, and what does an additional one cost?
The answer that should worry you: a single-seat quote for a three-person BD function. Price the seats you will actually use.
6. What does "refreshed" mean?
Every vendor refreshes. The word covers everything from a nightly re-crawl of a website to a quarterly re-model of every financial estimate, and the difference matters when the signal you are buying is that a company changed.
What to ask: when did this record last change, what changed, and can I see that history? A product that shows the last three edits has a refresh process. One that shows a single "last updated" stamp has a timestamp.
The answer that should worry you: "real-time." Nothing in private-company data is real-time. The honest answer is a cadence and a method.
7. What does the score predict, and for which deal size?
The most interesting layer in the category is the intent signal, the score that says a company is more likely than its peers to enter a process. It is also where the scope of the claim matters most.
Grata's Seller Intent publishes recall figures of 98% in the US and 89% in EMEA on back-tested 2025 transactions, and Grata scopes those figures to mid-market and large-cap deals (grata.com, observed 2026-09-20). That scope is part of the claim. A recall number measured on transactions large enough to leave a public trail says nothing yet about a $4M EBITDA distributor whose sale will never be announced. No vendor we have read publishes an accuracy figure for intent signals below $10M of EBITDA.
What to ask: on what deal sizes was this score validated, and what is the base rate of companies in my band that transact in a year?
The answer that should worry you: the number quoted without the scope.
8. Who makes the call?
Nobody at the vendor. That is the category, not a criticism, and both major vendors are clear about which side of the line they sit on. Nothing on either surface describes contacting an owner, confirming a listed deal is still available or delivering a CIM as something the product does.
It matters because the seat's return is a function of hours. A database in front of a team with time produces conversations. In front of a team without time it produces a larger list of companies nobody contacted. If the associate has four hours a week for outbound, the seat is not the first purchase, and a second seat makes it worse.
The answer that should worry you: your own, when you count the hours the firm actually has for the work the seat leaves behind.
Run the evaluation after the demo, not during it
A trial answers your questions, and it needs a list built before it starts. Take the last hundred companies your team identified by hand for the current mandate and run three tests:
- Presence. How many of the hundred are in the product at all? That is your coverage figure for your band, and the only one that exists.
- Estimate accuracy. For the twenty whose revenue you know, how far off is the product's figure, and in which direction?
- Deliverability. Export the owner contacts and send one plain email from a real address. Count bounces on day one and replies by day ten.
Two to three weeks is enough if the list is ready on day one. Decide on those three numbers. If the vendor will not run a trial against a list you supply, that is a fourth.
Where OmniSource sits
We sit near this category, so read this knowing that. OmniSource is not a database and publishes no company count to set against the ones above. It is a platform paired with a human origination team: targets identified, assessed and qualified against a documented mandate, approved by the client's deal team in a review queue, then worked through the outreach, the NDA ask and the CIM chase. Coverage spans Off-Market, Pre-Market and On-Market in one pipeline, because a mandate does not care which channel a deal arrives through.
If question 1 is your gap, buy the seat. If question 8 is, the seat will not fix it, and the two purchases are complements. The Grata alternatives page sets out where each kind of product fits a lower middle market mandate.
Who this post is not for
This is written to the buyer's BD function. Owners considering a sale and advisors with a deal to place are served on biznexus.com, not here.
The short version
The database count is the vendor's number. Coverage of your band, the error on the financials, the bounce rate on owner contacts and the hours you have for what the seat leaves behind are yours. Bring a hundred companies to the trial, run three tests, and buy on the results.