Should a fund run a broad survey or a targeted referral pilot? A decision matrix that picks by what you already know, and the case where neither is right

Short answer

Run a broad survey only when you cannot yet tell which portfolio companies pass the published screen, and keep it to a few metadata questions. Run a targeted pilot when a handful already pass and each has someone who will ask its CEO. A survey records what founders say; a pilot records what happens. When the screen facts are missing and nobody can ask, run neither: screen from what the fund already holds, then look beyond the portfolio.

Visual briefFull-size diagram
A two-by-two matrix with screen facts known on one axis and confirmed paths to the CEO on the other; its quadrants read survey, pilot, map owners and neither, with Illustrative Fund A placed in the pilot quadrant and Fund B in the survey quadrant.
Illustrative funds placed by the decision matrix on this page; the quadrant labels follow its stated criteria, not measured outcomes.

The choice is not about effort. A survey reaches every founder, including those who cannot qualify, and each recipient may read it as a sign that a program is coming. A pilot reaches a few, but tests the part a survey never can: whether a real person asks a real CEO and the CEO wants the conversation. Funding status decides neither; the screen is the same for a portfolio company and for one the fund never backed.

What each format can learn, and what it cannot

A survey answer is a claim. Anthropic's engineering guide to evaluating AI agents draws the same line for software: it separates what an agent says it did from what changed in its environment, checking a flight-booking agent's claim of success against whether a reservation actually exists. A founder's "about 60 staff, records since 2015" is the claim; a CEO agreeing in writing to a conversation, or a company submitting a verified application, is the reservation.

QuestionBroad surveyTargeted pilot
What it recordsSelf-reported screen factsDated events: the owner agreed, the CEO agreed, the introduction was recorded, the company applied
Who it reachesEvery portfolio company, including those that cannot qualifyThree to eight companies chosen in advance
Typical errorNon-response and misreported headcount, such as contractors counted as staffToo few rows to estimate any rate
Expectation riskEvery recipient may assume something will followLimited to companies that already pass
What you have at the endA classified list and a set of unknownsA tested path, or a specific reason it failed

Pin each row to an operating company and its controller

Both formats fail the same way if a row names the wrong entity. A founder asked about "your company" may answer for a holding company, a spinout or a business they ran before. Every survey question and every pilot row should name the operating company that employs the staff and created the records, its controller as of a stated date, and its CEO, who decides whether it discusses licensing and signs for it. The screening workbook and the no-upload screening form both carry those fields.

The survey-versus-pilot decision matrix

Place the fund on two axes before choosing: how many companies' screen facts you already know (50+ full-time employees at peak, several years of the company's own records, rights to license, an eligible location), and how many of those companies have a confirmed path to the CEO.

Screen factsNo confirmed paths to CEOs yetConfirmed paths for some companies
Known for most companiesMap relationship owners first; neither format helps until someone can askTargeted pilot of the companies that pass
Unknown for most companiesNeither: screen from board data and public facts, then look outside the portfolioShort metadata survey to the founders you can reach

Five criteria break a tie when the fund sits near a line.

CriterionPoints toward a surveyPoints toward a pilotPoints toward neither
Portfolio stageGrowth stage: most companies are large, records unknownMixed: a few large companies, well knownSeed stage: few or none near 50 full-time at peak
Facts already heldHeadcount known, rights and English share notAll four screen facts known for a few companiesAlmost nothing beyond funding data
Relationship ownersFounders answer the platform team directlyA named person will ask each target CEONobody will ask
Founder fatigueNo survey in the past two quartersRecent surveys; spare the restFounders already over-asked
What you need to learnHow many companies might passWhether the path from ask to application worksNothing actionable yet

Method note. The axes and criteria are our design, built from the published screen (who qualifies and the program terms, Version 2026-10-22.2, since replaced by version 2026-11-04, both checked 2026-10-26) and from standard survey practice. They do not measure demand for any company's records, record quality or payout, which only SourceX's review assesses. No data on survey or pilot outcomes in referral programs has been published, so the matrix sorts by what you know, not by expected results.

Write the survey so it screens instead of promising

Eight questions, all metadata, none needing a file:

  1. Legal name and country of the operating company that employs your staff.
  2. Full-time employees at peak, contractors excluded.
  3. The year your company's own records begin in its own systems.
  4. Whose records they are: the company's own, or mostly held for customers.
  5. If you are outside the U.S., Canada and the EU, roughly what share of the records is in English. The rule, verbatim: Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English.
  6. Your CEO's name, if it is not you.
  7. Whether the company has ever applied to or corresponded with SourceX; a company SourceX already knows is reviewed as a duplicate, not credited to a new partner.
  8. Whether the CEO would want a conversation: yes, no or not now.

Open the survey with two sentences: answering is not an application, and SourceX decides fit only in its own review after a company applies. Program criteria can move while a survey is open, so stamp it with the date of the criteria you used, as described in the risk of using old mandate criteria in founder outreach.

Then classify every company, not just the ones that answered. Survey methodologists sort final outcomes into four groups: completed, eligible non-respondents, unknown eligibility and not eligible. Adapted to a portfolio, those become answered; known from board data to pass but silent; silent with screen facts unknown; and known to fail a published test. Report all four counts. The same standards warn that a response rate alone is not enough to tell how much non-response error a survey carries, or whether it carries any, and in a portfolio that error is the whole question: the busiest CEOs of the largest companies may be the ones who never reply.

Design a pilot small enough to finish

Choose only companies that pass the screen and have a named person who will ask the CEO. Fix the outcomes before the first ask, in order: owner agreed to ask; CEO informed and wants the conversation; whoever submits has signed the SourceX Channel Partner Agreement; introduction recorded in writing; verified application within 90 days; SourceX's review outcome when it is known. Set a stop date, such as 60 days from the first ask, and a stop rule, such as three CEOs declining for the same reason. A pilot of a few companies cannot estimate a conversion rate, and the ten-introduction experiment explains what so small a test can still reject.

Two funds, two answers (Illustrative)

Fictional funds, companies and counts; no outcome implied.

Fund A, early stage, 48 portfolio companies. Board data shows five with 50+ full-time employees at peak, and the platform team speaks to all five founders monthly. Matrix: facts known, paths confirmed, so a targeted pilot. Four go in. The fifth, a 70-person software company, is rejected before the pilot starts: nearly all its records sit in customers' accounts, and client-held material is usually not a fit. A survey of the other 43 would have asked founders who cannot qualify to spend time on nothing.

Fund B, growth stage, 22 portfolio companies. Most employ 150 to 900 people, but the fund knows little about their records, rights or English share, and several operate in India and Latin America. Matrix: facts unknown, founders reachable, so a short survey to all 22.

Outcome group (Fund B)CountWhat happens next
Answered14Each answer is checked against the screen; passing rows go to their relationship owner
Known to pass, silent3The partner who sits on each board asks once in person
Silent, facts unknown3Left unclassified; never counted as either a pass or a fail
Known to fail a published test2Told plainly, with the dated reason

One of the two that fail is a Brazilian logistics company whose records are mostly in Portuguese; under the 70% English rule it does not fit today, however strong its records.

Close the loop without implying acceptance

OutcomeWhat the fund writesWhat it never writes
Passes, CEO interested"Your answers match the published screen as of [date]; SourceX decides after your CEO applies""You qualify"
Passes, CEO not now"Noted; we will not raise it again unless you ask""We'll check back next quarter"
Fails a published test"Under the criteria dated [date], [test] is not met"Any workaround for the test
UnclassifiedNothing further unless the founder repliesA second and third chaser

What could make this matrix the wrong tool

  • Non-response hides the answer you need. A response rate alone cannot tell you how much non-response error the survey carries, or whether the silent companies differ from those that replied.
  • Claims are not outcomes. Self-reported headcount and rights are the founder's view; only SourceX's review tests rights and relevance.
  • Criteria move. Requirements differ by program, so a survey that runs for months can classify companies against criteria that have since changed.
  • Small pilots overreach. Four introductions can show a path works or fails; they cannot show how often it would work across a portfolio.
  • Relationship cost is unmeasured. No data on how founders react to data licensing surveys or asks has been published; record your own.

Funds whose portfolios mostly fail the size test may find more candidates around the portfolio than in it, as mapped in your next AI data introduction may sit outside the portfolio. A company that wants to assess itself first can use the company data readiness survey.

Next step

Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

Once the matrix points to a pilot and a named person will ask each CEO, register for a partner link, recheck every pilot company against who qualifies and read the program terms before the first ask; other portfolio playbooks are in guides.

Choosing between a survey and a pilot is referral work, and deciding whether SourceX itself deserves an investment is a separate question; investor materials are available on request through the SourceX investor overview.

This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.

  1. Step 1Share your linkSend your personal link to a company you know.
  2. Step 2Company appliesThe company applies itself at /apply.
  3. Step 3Company signs and is paidThe company signs, delivers its data and is paid by SourceX under its agreement.
  4. Step 4You get your rewardthe greater of $50,000 plus 5% and 15% of the actual supplier payout, with no cap. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details.

Common questions

Can a fund survey founders about data licensing without collecting anything sensitive?

Yes, if every question asks for metadata: full-time headcount at peak, the year the company's own records begin, whose records they are and the CEO's name. Ask for no files, samples or system access; a fit screen never needs them.

How many companies belong in a first referral pilot?

As many as already pass the published screen and have someone willing to ask the CEO, which for many early-stage funds may be only a handful. A pilot that small tests whether the path works; it cannot estimate a rate, and SourceX has not published conversion rates to compare against.

Does a high survey response rate mean the answers can be trusted?

Not on its own. Survey methodologists note that a response rate alone is not enough to tell how much non-response error a survey has, or whether it has any. Report all four outcome groups, including the companies you could not classify.

Sources

  1. refer.sourcex.si/terms
  2. refer.sourcex.si/who-qualifies
  3. aapor.org/wp-content/uploads/2023/05/Standards-Definit…
  4. anthropic.com/engineering/demystifying-evals-for-ai-agents
  5. sourcex.si/investors

Free resources

Know a company with valuable proprietary data?

Become a referral partner, get your link and introduce the company’s CEO (in an owner-run company, usually the owner). Referral partners can join from most countries. We can't work with partners in countries or regions under U.S. sanctions or on U.S. restricted-party lists. The companies you introduce should be based in the U.S., Canada and the EU, or in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia with at least 70% of their data in English.

Refer a company →

I own a business

Explore licensing your company's data directly to SourceX for AI training and evaluation. Start a short assessment; no uploads needed.

Start an assessment