Reward income is the wrong yardstick for a pilot. Hold it at $0 in the base case and judge the pilot on what the platform can observe within a quarter: whether founders found the conversation useful, and what the work cost in platform and founder hours.
Why a pilot cannot be judged on rewards
The scarce input in a platform team is attention, and compensation arrives last, if at all. The reward, when one is earned, equals the greater of $50,000 plus 5% and 15% of the Actual Supplier Payout the referred company receives from SourceX, with no cap; a lead, meeting or signed agreement alone earns nothing payable, and a qualifying reward falls due within 60 calendar days of the later of two events, completed handoff and the point at which the Acquisition Partner's acceptance of the complete, validated dataset is final and in writing. Upstream, the company's own published path includes an assessment of initially up to 30 days after it signs SourceX's Master agreement, an acceptance review of up to 30 days and payment within 60 calendar days of a valid invoice after Closing. A one-quarter pilot can end with every company still moving and every step working. Keep three ledgers, and never let the third stand in for the first two.
| Ledger | What it records | Where the data comes from | How to read it |
|---|---|---|---|
| Founder usefulness | Whether each CEO or founder who heard the idea found it worth their time | A two-line note after each conversation, coded useful, neutral or a cost, in the founder's own words | The pilot's main result |
| Operating effort | Platform hours and founder hours, by stage | A weekly time log, kept as you go | The price of that result |
| Compensation | Rewards actually paid | Dashboard status Paid, shown only after an actual payout is recorded | Observed, never forecast; $0 in the base case |
Five failure types and two false alarms
Amy Edmondson's Harvard Business Review article on learning from failure sorts failures into three categories: preventable failures in predictable operations, unavoidable failures in complex systems, and intelligent failures at the frontier, and argues that each needs its own strategy. A referral pilot produces all three. Fix the preventable ones, accept complex ones without drawing big conclusions, and keep intelligent ones small.
| Label | Where it shows | Evidence that confirms it | Category | What it teaches |
|---|---|---|---|---|
| F1 Empty cohort | Screening: few companies meet the published fit facts | The screen sheet: full-time headcount at peak, years of the company's own records, rights, an authorized CEO | Intelligent if the pilot was built to test the cohort; preventable if nobody screened | Where fit companies sit in your network, not how well you asked |
| F2 No willing owner | The ask: a company fits, but whoever holds the CEO relationship declines or stalls | The ask log, with each owner's answer and reason | Complex: goodwill, timing and politics you cannot plan away | What a relationship owner will spend goodwill on |
| F3 Misread offer | The conversation: a CEO declines on a misunderstanding, such as selling customer data, a middleman or a sale of the company | Founder notes in the CEO's words, coded by a second person | Intelligent the first time; preventable once seen | How your explanation lands |
| F4 SourceX decline | Review: records do not match current data needs, rights block a license, or SourceX already knew the company | The outcome as SourceX communicates it | Complex: outside the platform's control | Something about that company today, not about the sector |
| F5 Slippage | Anywhere: follow-ups dropped, the operator reassigned, notes not kept | Planned against logged hours; gaps in the ask log | Preventable | Whether the platform has the capacity it assumed |
| N1 Still moving | After application: the company is in review or a deal when the pilot ends | Dated stage changes per company | Not a failure: the window was shorter than the process | How long to keep observing |
| N2 Informed no | The conversation: the CEO understood the offer and declined | The founder note shows the CEO described the offer correctly | Not a failure: the founder kept control | That the explanation worked |
Two rules keep the labels honest. Each company that did not finish gets one primary label, at the furthest stage it reached. A company your records cannot place is labeled unplaceable, and unplaceable rows go first in the report, because they mean the pilot failed at measurement before it failed at anything else.
F4 needs restraint: each SourceX decision applies one specific program's criteria, so a company declined under one program can still suit a later one. A duplicate is also an F4; companies with an earlier application, relationship or valid introduction with SourceX are not credited to a later partner, and the platform cannot see SourceX's records in advance.
The work at each stage, with income held at $0
The platform's job ends early: partners never export, upload or describe confidential records, and the company applies itself through a short assessment with no uploads. The hours are a planning template, not a benchmark; no figures from real pilots are published, so replace them with your own logs after a month.
| Stage | Who does the work | Illustrative hours per company | What you record |
|---|---|---|---|
| Screen | Platform, from what it already knows, at metadata level | 0.5 platform | Passed, failed on a named fact, or unknown |
| Ask the relationship owner | Platform explains; the owner decides whether to raise it | 0.5 platform, 0.25 owner | Agreed, or declined with the reason |
| CEO conversation | The owner asks; the CEO decides | 1 platform, 0.5 founder | Wants a conversation, informed no, or misread |
| Introduction and application | Platform submits; the company completes the assessment | 0.5 platform, 1 founder | Verified application, or none |
| Review and deal | The company works with SourceX; platform answers questions, never handles records | 0.25 platform a month | Stage changes from the dashboard |
| Reward | Nobody, until the conditions are met | 0 | Paid, observed only |
Credit follows the published attribution rule: once a company submits a verified application, the credit goes to whichever valid introduction of it came first in the 90-day window, and that credit is never shared between partners. Whoever submits must first have signed the SourceX Channel Partner Agreement; anyone who signed it under its older name, Channel Partner Non-Circumvention and Non-Disclosure Agreement, is covered by that agreement, which remains in force as signed.
Same zero, different lesson (Illustrative)
Illustrative. Two invented funds, invented companies and invented counts, built to show the method. Both pilots ran one quarter and both paid $0 in rewards.
Pilot A. An early-stage fund screened 12 portfolio and former portfolio companies and logged 22 platform hours and 9 founder hours.
Pilot B. A seed fund screened 14 portfolio companies and logged 11 platform hours and 2 founder hours.
| Label | Pilot A (12 companies) | Pilot B (14 companies) |
|---|---|---|
| F1 Empty cohort | 4 | 11 |
| F2 No willing owner | 1 | 1 |
| F3 Misread offer | 3 | 0 |
| F4 SourceX decline | 1 | 0 |
| F5 Slippage | 0 | 0 |
| N1 Still moving | 2 | 0 |
| N2 Informed no | 1 | 2 |
| Founders who heard it: useful, neutral, cost | 4, 2, 1 | 0, 1, 1 |
| Decision | Redesign the message, keep observing, no new companies yet | Stop |
Pilot A's zero is mostly timing and wording. Two companies are still in review, so the quarter ended first; the stage duration benchmark shows how to count them as open, not lost. Three CEOs heard "AI data" and pictured their customers' records being sold, a misreading the guide to what founders think data licensing means anticipates and a preventable failure from now on. One SourceX decline describes one company's records, nothing more. The team rewrites its explanation, tests it on the next three fit companies and watches the two open ones without adding hours.
Pilot B is the case where the team should stop. Eleven of its fourteen companies had never reached 50 full-time employees, so the zero comes from the cohort, not from how anyone asked. Of the three that fit, one relationship owner declined and both CEOs who heard the idea understood it and said no, one calling it a distraction from fundraising. Another quarter with this cohort would repeat the result and spend more founder goodwill. That the pilot was cheap, 11 platform hours, is no argument for repeating it: a cheap test that answered its question has done its job. If the fund still believes in the idea, the next test is a new hypothesis with its own budget, such as LP-owned companies from the map of relationships outside the portfolio, not a continuation.
The one-page learning report
Write the report within two weeks of the cutoff, from records, and send it to the people who ran the asks before anyone else reads it.
| Section | What goes in it | Rule |
|---|---|---|
| Hypothesis | One sentence written before launch, such as "at least 3 of 12 CEOs will find a licensing conversation useful" | Dated before the first ask and never edited afterward |
| Cohort | How the companies were chosen, how many, on what date | Every company stays in the denominator |
| Stage counts | Companies at each stage, with dates | From the ask log and dashboard, not memory |
| Labels | One primary label per company, unplaceable rows first | A second person checks every F3 and N2 |
| Founder usefulness | Useful, neutral or a cost, with a short quote | Coded by someone who did not make the asks |
| Effort | Platform and founder hours by stage | Copied from the weekly log |
| Compensation | Rewards paid | Observed only; no estimates in this report |
| Decision | Expand, redesign, pause or stop, and the rule that triggered it | The rule existed before launch |
| What we will not conclude | At least one tempting conclusion the data cannot carry | For example, sector demand from a single decline |
Decision rules to write before launch
Apply these at the cutoff. They read the labels, the founder ledger and the hours, never an expected payout.
- Stop when F1 covers most of the cohort and most founders who heard the idea rated it neutral or a cost. The cohort is the cause; repeating it repeats the result.
- Stop when F5 recurs after one fix and nobody can be named to own the work; the time question for platform teams covers that trade-off.
- Redesign when F2 or F3 dominate among companies that fit. Change who asks or what they say, then test on a small batch before adding companies.
- Pause when N1 dominates: stop new asks, keep observing, and set a second cutoff.
- Expand only when fit companies exist, most founders rated the conversation useful, platform hours per CEO conversation stayed within the cap you set, and at least one company has moved past review. A paid reward is not one of the conditions.
Any single F4 is ignored for strategy. Three for the same reason in the same kind of record are a pattern to ask SourceX about.
Where this taxonomy can mislead
- Small numbers. With a dozen companies, one reclassified row can flip a label's rank. Report counts, not percentages, below 20 companies.
- Coding the no. "Not now" can be an informed no or a misread offer. When two coders disagree, the row is unplaceable.
- Demand moves. SourceX's company site says value depends on actual Acquisition Partner demand and due diligence, not size alone, so one pilot's period says little about the next.
- Rights. A company can pass every visible fact and fail on rights in review, where client confidentiality, employee privacy and third-party contracts surface.
- Late rewards. A reward that arrives after the report is a dated addendum, not a reason to rewrite the decision.
Methods note. The labels come from the published referral stages, the program's fit facts and attribution rule, and Edmondson's three failure categories, checked on 2026-10-26. They are untested on real pilots, SourceX has published no pilot or conversion data, and every count here is invented. They record where and why companies stopped, not record quality or demand, which only SourceX's review assesses. To try them, label every company from your last pilot; if more than a quarter are unplaceable, fix the logging before running anything new. The first ten conversations guide and the overview for venture platform teams cover the work before a pilot starts.
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.
When your learning report points to a company that fits and a CEO who wants the conversation, get a partner link, test that company once more against the published fit criteria and work through the program terms before the introduction goes out.
Running a referral pilot and assessing SourceX as a company are separate decisions; readers weighing the second 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.