The companies a network overlooks, and the introductions that go nowhere, are often the most useful records in the set. They show partners where not to spend next month. A report that leaves them out describes the wins and teaches very little.
What is known, what is conditional and what is only a question
| Type | Statement | Basis |
|---|---|---|
| Program fact | Each company has a single credited partner, the one whose valid introduction of a new company came first within the 90-day attribution window, and credit attaches when that company submits a verified application | Program terms, Version 2026-10-22.2, since replaced by version 2026-11-04 |
| Program fact | Companies that already had an application, relationship or earlier valid introduction are reviewed as duplicates and not credited to a later partner | Program terms |
| Program fact | When a reward is reduced, reversed or canceled, the reason is recorded | Rewards page |
| Conditional | Whether an introduced company is accepted, licenses data or produces a reward | SourceX review, a signed agreement and the payment conditions in the terms |
| Proposed thesis | Most of what a partner can learn sits in declined and unanswered cases rather than paid ones | A question for the first edition, not a finding |
| Term | How this page uses it |
|---|---|
| Census | Every introduction submitted in the year, not a sample of them |
| Disposition | The status each case holds at the cutoff, from a fixed list |
| Pre-registration | A dated public record of questions, variables and analysis, written before anyone looks at outcomes |
| Suppressed cell | A table cell withheld because it rests on fewer than ten cases |
| Edition | One annual release, reporting a named entry cohort at a named cutoff |
Why an introductions report cannot be built like an opinion survey
Most "state of" reports in partnerships are surveys. PartnerStack's State of Partnerships in GTM 2026, produced with Wynter, rests on a survey of 100 senior B2B SaaS leaders and reports, for example, that 69% of them plan to increase investment in partnerships. That format answers what leaders think and intend. It cannot say what happened to a set of introduced companies, because nobody's recollection supplies a denominator.
A report on company data introductions has to start from dated outcome records: when each introduction was submitted, whether a verified application followed, where review stopped it, and how it ended. Surveys of partners and owners belong in a second layer that explains the records, never as the source of an outcome.
The refer site already carries four single-question study designs, each with its own cohort: referral quality by partner role, rejection reasons, conversion denominators and stage durations. The annual report does not replace them. It joins them on one cohort, one cutoff and one set of denominators, and adds three groups none of them covers: partners who registered and never introduced anyone, companies a partner screened out before any introduction, and owners who received an introduction and declined it.
The sampling frame: four populations, four consent bases
| Population | Who is in it | Census or sample | Consent basis | Never collected |
|---|---|---|---|---|
| Introductions | Every introduction submitted between 1 January and 31 December | Census | Program records, used only in aggregate; the privacy basis is confirmed before the study starts, and if it does not cover research use, consent is asked | Company names, records, contract terms |
| Partners | Every registered partner, including those with no introductions | Census for counts, opt-in sample for the survey | Counts from registration; survey answers only with separate opt-in consent | Individual reward amounts |
| Screened-out companies | Companies a partner considered and chose not to introduce | Opt-in sample of partners who keep a screening log | Partner opt-in; each company described only by size band, industry and reason | Company identity |
| Owners | Decision makers who received an introduction, whatever the outcome | Opt-in sample | Separate consent, requested after the case reaches its disposition | Anything about the company's records |
Two rules apply to every population. Nothing is published that would let a reader identify a company or a partner, and no partner ever sees another partner's records, which the program terms already rule out. A partner may be quoted only with written permission, and because a partner can earn a reward, the quote carries that relationship beside it. The FTC's endorsement guidance treats a payment as a material connection to disclose when a significant minority of the audience would not expect it, and it warns that vague labels do not explain the relationship.
The annual research methods blueprint
The blueprint has two parts: a calendar that fixes when each decision is made, and a question register that fixes what the edition will answer.
Part A, the calendar. Each edition reports two cohorts: last year's introductions at a six-month cutoff, and the year before at eighteen months, because payment can come long after an introduction and a single short window would undercount every slow outcome.
| Window | Step | Published output |
|---|---|---|
| January | Pre-register the twelve questions, variables, denominators, suppression rule and analysis | Dated pre-registration |
| January to December | Cohort entry: every introduction submitted joins the census | Running count of introductions only |
| 31 December | Entry closes | Final entry count |
| 30 June, following year | Cutoff: every case receives a disposition | Disposition flow |
| July to August | Analysis as pre-registered; anything extra is labeled exploratory | Filled tables |
| September | Outside review of method and calculations; owner and partner surveys close | Reviewer note |
| October | Publication | Report, codebook, flow, tables, limitations, change log |
Part B, the question register. Every question below is open. No study data has been collected and no answer exists yet; an edition fills a row only when its denominator clears the suppression rule.
| # | Question | Measure | Denominator |
|---|---|---|---|
| Q1 | How many introductions reach a verified application? | Application share | All introductions submitted |
| Q2 | Where do declined cases stop, and why? | Primary decline code by stage | Verified applications |
| Q3 | How many declines could a partner have caught from metadata alone? | Share of declines a partner could have checked before introducing | All declines |
| Q4 | How long does each stage take, open cases included? | Median time per stage with open cases treated as censored | Cases entering each stage |
| Q5 | How many introductions are duplicates or fall outside the window? | Attribution outcome share | All introductions submitted |
| Q6 | How many registered partners never introduce anyone? | Inactive share | All registered partners |
| Q7 | Which companies do partners screen out themselves, and for what reason? | Self-screen reason counts | Screening-log entries |
| Q8 | Did owners who declined find the introduction clear and respectful? | Owner rating, with non-response shown | Owners invited to the survey |
| Q9 | Do outcomes differ by partner role, network type or company size band? | Rates within strata, each with an interval | Introductions per stratum |
| Q10 | How often are rewards reduced, reversed or canceled, and for what recorded reason? | Reason counts | Rewards that reached Estimated status |
| Q11 | How much of the cohort is still open at the cutoff? | Open share | All introductions submitted |
| Q12 | What changed since the previous edition, and is it larger than noise? | Difference with an interval | Both cohorts |
Q3 uses the codes in the rejection reasons study, where headcount, geography and a missing authorized contact are the declines a partner could check. Q4 follows the stage-duration method. Q10 uses the reward statuses defined on the rewards page.
Every case gets a disposition, including the ones that went nowhere
The flow starts at "introduction submitted" and ends with every case in exactly one box. Nothing leaves the denominator because it is inconvenient.
| Disposition | Meaning | In the headline denominator? |
|---|---|---|
| Paid | The company was paid for a completed handoff | Yes |
| Declined | Stopped in review, with stage and one primary code | Yes |
| Withdrawn | The company chose to stop | Yes |
| No application | Introduced, but no verified application followed | Yes |
| Open | Still in progress at the cutoff | Yes |
| Duplicate or outside the window | An attribution outcome, not a judgment about the company | Shown in the flow and counted separately |
| Missing field | A case lacking a variable the analysis needs | Shown as its own row; never filled in by guesswork |
"No application" is the box most reports skip, and it often holds the largest group. A company that received a referral link and did nothing is a real outcome: the owner may have checked who qualifies and decided the company was too small, or simply not replied.
Descriptive, not causal. Partners choose whom to introduce, so a difference between partner roles may be a difference in the companies they know. Reporting rates within size bands and record types narrows that problem without removing it. The report therefore says "introductions from accountants reached review more often in this cohort", never "accountants produce better introductions". Causal wording is reserved for questions where something was randomly assigned, such as the message experiment in what makes a company introduction feel trustworthy.
What each edition publishes for reuse
- The pre-registration, with its date, and a list of every deviation from it.
- A codebook defining each variable, disposition and decline code, with the taxonomy version.
- The disposition flow, filled, with counts that reconcile from top to bottom.
- Every pre-registered table, including the unflattering ones and the ones with null results.
- The suppression log: which cells were withheld and why.
- The program terms version that applied to each case, because the terms changed four times in October 2026 alone (versions dated 12, 19, 21 and 22 October).
- A one-page method card for sharing: the question register, the cutoff date, the denominators and a link back to the full report.
Where the blueprint would break
A no-fit case it must record. Illustrative: a partner introduces a 35-person engineering consultancy whose founder is a longtime client. The founder applies, and review declines it on headcount, because the program needs 50 or more full-time employees at peak. In the flow it is one row in "Declined", stage eligibility, a headcount code. It looks trivial, yet it is exactly what Q3 counts: a decline the partner could have caught with one question. The cost of an unqualified introduction sets out how to measure what such a row cost the three people involved, and the poor-fit patterns page lists the patterns behind most of these declines.
A year too small to call a state report. If an entry cohort is small, most cells are suppressed and the intervals are too wide to say anything. The rule is written in advance: if more than half the pre-registered cells would be suppressed, the release is called a methods edition, publishes the flow and the counts, and makes no comparative claims.
The operator studies itself. SourceX runs the program and would run the study, which is a conflict of interest. The controls are the pre-registration, publishing every pre-registered table, an outside reviewer named in the report, and a change log.
Self-selection. Screening logs and surveys are opt-in, so Q7 and Q8 describe the partners and owners who chose to take part. The report states the invitation count beside each response count.
Criteria move. SourceX's needs differ by program, so a company outside one program's criteria today can fall inside another's later. A decline in one edition is not a permanent verdict on a company, and the report says so.
A measurable next step. Before the first edition exists, test the method on your own book. List every introduction you made in the last 12 months and assign each a disposition from the table above. Count three things: cases you cannot classify, cases in "No application", and declines you could have caught with a metadata question. If the first number is more than a handful, your own records are the first gap to close. Then follow a live case end to end with the submission-to-payment guide. Anyone submitting introductions signs the SourceX Channel Partner Agreement first; a copy signed under the name it carried before, Channel Partner Non-Circumvention and Non-Disclosure Agreement, stays in force on its signed terms.
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.
If your own disposition count shows introductions worth making, register for a partner link, screen each company against who qualifies, read the program terms, and introduce an authorized owner only when the company wants the conversation. More study designs sit in the guides hub.