This page is for VC platform leads and operating partners planning a small referral pilot across a portfolio or wider network. It covers one sampling question the published pilot designs leave open.
The company a pilot never sees
Every network screens twice. Once on the published fit facts, which is necessary. And once, silently, on the owner's sense of who would say no: the founder who "hates anything AI", the CEO who "never has time", the company that "would never share records". That second screen is usually kind and often right. It is also never tested, because the companies it removes never produce a result. A pilot built only from likely-yes companies reports a yes rate for favorites and invites the reader to treat it as a yes rate for the network.
Including likely decliners turns the second screen into something you can check. It also fits what a pilot is for. NCCIH's guidance on pilot studies says: "The goal of pilot work is not to test hypotheses about the effects of an intervention, but rather, to assess the feasibility/acceptability of an approach to be used in a larger scale study." Here the feasibility questions are concrete: will owners actually ask CEOs they expect to refuse, and do those CEOs react badly when asked well?
Three groups are easy to confuse, and only one of them belongs in the likely-no stratum.
| Group | What it is | In this pilot? |
|---|---|---|
| Likely to decline | The owner's prediction about a company that passes every fit fact known so far | Yes: these are the negative cases |
| Has declined | A CEO who already said no | No; whether to reopen is decided under reconnecting after a company declined |
| Likely to fail fit | A company the owner's data thesis ranks low | Not here; the thesis-contrast design samples that contrast |
The negative-case sampling plan
Fixed in writing before anyone is asked. The numbers are study settings, labeled Illustrative.
| Step | Rule | Why |
|---|---|---|
| 1. Frame | Every company with a named relationship owner that passes all fit facts known today, list frozen on a date | Strata must differ in predicted willingness, not in fit |
| 2. Exclusions | Remove any company whose CEO has said no, asked not to be contacted, is in a crisis, is wound down or in a formal process, or is known to have had earlier contact with SourceX | A known fact outranks a prediction; counted by reason, never replaced |
| 3. Sealed prediction | Each owner marks every remaining company likely yes, unsure or likely no; each likely no gets one reason code (no time this year, wary of AI, records feel sensitive, relationship too new, other) | The prediction is fixed before anything can change it |
| 4. Sealing | Predictions go to the platform lead before the draw and stay unopened until the read date | Nobody adjusts an ask, or a memory, to match |
| 5. Draw | By lot: six likely yes, four unsure, six likely no; if fewer than six likely no exist, take them all; below four, drop the contrast and record why | Equal yes and no groups make the comparison readable |
| 6. One ask | The same message for every CEO: why this company, that the CEO decides, the owner's reward disclosed, a no that takes one reply, at most one follow-up | The stratum must never shape the tone |
| 7. Owner opt-out | An owner may refuse to ask a drawn company; it is recorded as not asked and stays in its stratum | Reluctance is a result, not a reason to resample |
| 8. Final dispositions | On the read date, 120 days after the last ask, each company gets exactly one: not asked, not reached, CEO declined, CEO deferred to a date, CEO yes with a fit fact failing, CEO yes with none failing, duplicate | Every drawn company is accounted for, and each ask has passed its 90-day window |
Step 8 follows survey practice. AAPOR's Standard Definitions, in paraphrase, give every sampled case a final disposition, non-respondents and cases of unknown eligibility included, derive response, refusal and contact rates from those codes, and treat interim codes for a single contact attempt as unfit for final use. A company that "hasn't replied yet" on the read date gets a final code, not a hopeful blank.
Section 4 of the terms helps the single ask stay honest: the reward "is paid separately by SourceX and is never deducted from the supplier's agreed compensation," and SourceX has confirmed the company receives the same amount with or without a partner. Saying so is the same sentence in every stratum.
The exclusion in step 2 matters most for companies in a formal process. SourceX's screening page says that for a company wound down or in a formal process the CEO may not be enough, because whoever controls the estate decides; a pilot about CEO willingness has no place for it.
What is compared, and what is only described
The primary count is CEOs who said yes with no published fit fact failing, shown per stratum as counts, side by side.
The prediction table sets each sealed prediction against what happened: yes, no, deferred or no answer. It shows whether the owners' sense of reluctance carries information in this network.
Reasons stay descriptive. The sealed reason code can be set beside any reason a CEO chose to give, but nobody asks a CEO why they declined for the pilot's sake; collecting reasons properly needs consent and anonymity, which the decline-reason survey design provides.
Draws are random within each stratum, so the sample is fair to the owners' list. The strata are the owners' own judgments, though. A gap between likely yes and likely no shows whether those judgments predict answers here; it does not show why, and it says nothing about another network.
Decision rules written before the first ask
| Result, strata of six | Reading | What changes | What it never justifies |
|---|---|---|---|
| Likely-no yeses level with, above or one below likely-yes yeses | The silent screen is discarding conversations | Screen on the published fit facts only; stop excluding companies for predicted reluctance | Asking any CEO more than once |
| Likely-no yeses at least three below likely-yes yeses | Owners' predictions carry information | Use predictions to order the queue, and log every company skipped | Treating a prediction as a fact about the company |
| Likely-no yeses exactly two below | Inconclusive | Run a second cohort with new companies before changing practice | Asking the same companies again |
| Owners refuse more than half of the drawn likely-no asks | Unreadable as a comparison | Report the refusals; owners' reluctance is the finding | Any rate at all |
| Any complaint from a likely-no CEO about the ask | Stop rule | Asks to that stratum end and the script is reviewed | Continuing because the yes count looked good |
Sixteen companies, sixteen final codes
All sixteen drawn companies appear in the result, each with its final disposition. Excluded companies appear above the table as counts by reason. Deferred companies and applications still in progress on the read date are reported as open, not as losses, and a company that turns out to have earlier contact with SourceX is reported apart: under section 3 it is a duplicate and no partner is credited. Companies declined in SourceX's own review are reported as that, never folded into "CEO declined".
Six companies per stratum can show only a large gap. The report gives counts, never percentages, prints every zero, and lists what went wrong: owners who would not ask, predictions that went unsealed, asks that strayed from the script. If nothing completes in the window, the learning review for pilots with no completed transactions explains how to read the open rows later.
Where the sampling plan can go wrong
- Goodwill. Asking CEOs you expect to refuse spends some. The single ask, the one-reply no and the complaint stop rule limit it; they do not reduce it to zero.
- Leaking expectations. An owner who expects a no may write more stiffly despite the script. Reviewing a few asks from each stratum against the script before sending catches the worst of it.
- Predictions that drift. An owner may learn news after sealing. The sealed version is the one compared; the update is noted.
- Fit decided elsewhere. A yes with no failed fit fact can still be declined in SourceX's review, and the screening page says criteria follow each program and can move.
- Thin strata. Many networks will hold fewer than six honest likely-no companies; the plan says what to do then, and "too few to compare" is a valid finding.
- No market reading. Nothing here measures demand for any company's records or the size of any market.
For scoring one owner's forecasts company by company, rather than comparing groups, see the ten-introduction experiment.
Methods note. The plan rests on SourceX's terms and screening page, AAPOR's Standard Definitions and NCCIH's guidance on pilot studies, all checked on 2026-10-28. The strata sizes, reason codes and decision thresholds are our study settings and have not been run. Not measured: any company's records, payouts, SourceX's review decisions or how CEOs outside the sample would answer.
What every ask must keep true
The likely-no label never changes what a company must be to fit. It needs 50 or more full-time employees at peak, contractors left out; years of records it created; rights to license them; and a CEO willing to authorize the discussion. Funding plays no part. 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. Owners ask, at most, what kind of records exist; files, samples and logins stay with the company.
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.
Seal the predictions before the draw. When a drawn CEO in any stratum says yes, the owner who will submit can register, hold that company up to who qualifies a last time and go through the program terms before sending anything.
Running this pilot is referral work. Judging SourceX as a possible investment is a different task, and anyone doing it can request investor materials 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.