How should a company-data referral conversion study define its denominator and observation window?
Use verified applications in a fixed entry window as the denominator, follow every one to a single stated cutoff date, and report open cases separately. Anything else makes the rate move with the method rather than with the program.
The denominator
Referral funnels have several plausible starting points, and each answers a different question:
| Starting count | What it measures | Why it is not the headline denominator |
|---|---|---|
| Introductions submitted | Partner activity | Includes duplicates and introductions outside the 90-day attribution window |
| Introductions credited | Attribution outcomes | Credit only lands once the company submits a verified application |
| Verified applications | Companies that entered review | Recommended headline denominator |
| Companies that met eligibility | Review pass rate | Hides everything screened out at eligibility |
The SourceX terms credit the first valid introduction within the 90-day window when the company submits a verified application, so verified applications are the first point where a case is unambiguously in the program. Report the upstream counts too. CONSORT 2025 guidance for clinical trials makes the same point about the number assessed for eligibility and the reasons for exclusion: the count before the denominator tells readers how selective the entry point is.
The endpoint
Define conversion as a completed data handoff for which SourceX has paid the company its Actual Supplier Payout. In reward terms, that is the point at which a reward can move to Eligible on the rewards pageA 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. A meeting, a signature or an estimated status is an intermediate stage, not a conversion. The terms say so directly: a lead, meeting or signature alone does not trigger payment.
The observation window
Two dates define the window: the entry window (for example, verified applications dated within one calendar quarter) and the cutoff (one date on which every case's status is read). The NIST statistics handbook calls this a fixed-time design: cases without the event by the cutoff are right censored, meaning their final outcome is unknown, not negative.
A worked cohort: the same 30 cases, three different rates
Illustrative only. The cohort below is fictional, built to show the arithmetic; it is not SourceX data, a forecast or a benchmark.
Upstream, 40 introductions were submitted in the entry quarter; 6 were duplicates or fell outside the attribution window, leaving 34 attributable introductions; 30 of those companies completed a verified application, which is when credit lands. At the cutoff, the 30 split like this:
| Status at cutoff | Count | Share of 30 |
|---|---|---|
| Paid handoff completed | 3 | 10% |
| Declined in review | 17 | 57% |
| Withdrawn by the company | 4 | 13% |
| Still open | 6 | 20% |
| Total | 30 | 100% |
Now three ways to report it:
| Method | Calculation | Rate | Verdict |
|---|---|---|---|
| Observed conversion | 3 of 30 | 10% | Correct, if reported with the open count |
| Open cases dropped | 3 of 24 | 12.5% | Inflated: assumes open cases behave like closed ones |
| Bounded range | 3 to 9 of 30 | 10% to 30% | Correct: shows what the open cases allow |
The 12.5% figure is the common error. Dropping six open cases from the denominator silently assumes they will fail at the same rate as the closed ones, which nobody knows. Publishing the 10% to 30% range, plus the date of the cutoff, is the honest version. As cases close, rerun the same cohort at a later cutoff; never top up the cohort with newer applications. The survival-analysis literature makes the general point: time-to-event data analyzed without allowing for censoring can be misinterpreted. The companion guide on measuring stage duration applies the Kaplan-Meier method to the same problem.
What a published study should state
A partner reading a conversion figure from any program, including SourceX, should find these fields. If one is missing, treat the figure as unverified.
- Entry window dates and the event that admits a case (here, a verified application).
- Cutoff date on which statuses were read.
- Denominator count and every upstream count, with reasons for exclusion.
- Endpoint definition, in words a partner can check against the terms.
- Counts for each status at the cutoff, including still open.
- The range implied by open cases.
- Whether any signed Schedule A or addendum terms differ for part of the cohort.
Decline reasons belong in their own report with its own denominator, set out in the rejection reasons study design. The stages themselves are described in the referral funnel overview. Status tracking for a single introduction is covered in tracking a referral from submission to qualification, and teams who need an audit trail of who changed a status can borrow the logging ideas in MCP audit logging.
Which introductions belong in the cohort at all?
Only companies that could qualify, so screen at metadata level before you introduce. 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.
The company also needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license its records and an authorized contact. As a partner you can confirm which systems hold the records, roughly how much and how many years, who owns them and which executive can authorize a licensing discussion. You do not collect exports, samples or credentials. Dataset demand, the rights review result and any offer stay unknown until SourceX reviews the company, which is exactly why the cohort's still-open count matters.
Company compensation and the reward are counted separately
A conversion study counts companies, not dollars, but partners will ask what a conversion is worth. Keep the two amounts apart. The company receives its Actual Supplier Payout under its own agreement; SourceX funds the partner's reward separately and never deducts it. Illustrative only: an Actual Supplier Payout of USD 120,000 on a company's first qualifying transaction would mean a reward of USD 56,000; a payout of USD 400,000 would mean USD 70,000. Neither is a quote.
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 you want your own introductions to count cleanly in a future cohort, register as a partner before introducing, screen each company against who qualifies, and use the program terms to define the endpoint you track. Related studies sit in the guides hub.
What should a partner research lead know about data referral value drivers?
Value in this program appears at one point only: a completed handoff that SourceX pays the company for. A value-driver study therefore divides the value that reached paid status by every introduction that tried, zeros included, and asks which conditions at submission the paid cases shared. Code five candidate drivers (access, eligibility, timing, context and follow-through) before outcomes exist, compare them within strata in a cohort where every case has resolved, and report concentration beside averages. SourceX has published no results; this section is the method.
- Known: a reward accrues only as SourceX pays the referred company, and a lead, meeting or signed agreement alone earns nothing.
- Conditional: whether an introduction produces any payout depends on SourceX's review, the Acquisition Partner's final written acceptance of the dataset and a completed handoff.
- Thesis to test: most value comes from a few introductions in which the relationship owner reached the CEO directly and the fit facts were known at submission, while timing and follow-through matter less than partners assume. It is a hypothesis, not a finding.
| Term | Meaning in this study |
|---|---|
| Resolved cohort | Every verified application from a closed entry window, each with a final status at a fixed horizon |
| Zero-outcome introduction | A case that ends declined, withdrawn or unpaid at the horizon; it stays in the denominator at zero |
| Value per application | Total value in the cohort divided by every verified application in it |
| Driver | A condition coded at submission that might explain where value came from |
| Concentration | The share of the cohort's total value carried by its largest one and three cases |
Value means three things, and they do not move together
| Whose value | What is counted | Unit |
|---|---|---|
| The company | Cumulative Actual Supplier Payout within the horizon | Dollars paid to the company |
| The partner | the greater of $50,000 plus 5% and 15% of Actual Supplier Payout on the first qualifying transaction, then 15% on each further transaction first agreed within 12 months, with no cap | Dollars, accruing as installments are paid |
| SourceX | Handoffs the Acquisition Partner accepted in writing | A count; SourceX's commercial figures are not published and are outside the study |
The USD 50,000 floor and the USD 100,000 Minimum Transaction make the first two diverge. Illustrative: a cohort in which three companies are paid.
| Paid company (Illustrative) | Actual Supplier Payout | Partner reward |
|---|---|---|
| First | $40,000 | None: below the USD 100,000 Minimum Transaction |
| Second | $250,000 | $62,500 |
| Third | $800,000 | $120,000 |
The largest case carries 73% of the company payout ($800,000 of $1,090,000) but 66% of the partner rewards ($120,000 of $182,500), because the floor lifts the mid-sized first transaction and the Minimum Transaction leaves the smallest one unrewarded. A driver that predicts the size of the biggest handoffs therefore matters more to company value than to partner value, and a driver that predicts how many companies get paid matters relatively more to partners. Every finding has to say whose value it describes. Installments pay the reward proportionately as they arrive, and later transactions first agreed within the 12-month Repeat Window pay 15% without a second floor, which the renewal guide works through.
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.
The figures above are arithmetic on the published formula, not a forecast, a quote or a typical result.
The resolved cohort, and why the zeros stay
A cohort is resolved when its entry window has closed and every case has a final status at a fixed horizon. The unit is the verified application, the denominator the benchmark above already uses; introductions that never reached one are logged separately as the partner's cost. Fix the horizon in advance, for example 18 months from each verified application, and count cumulative payout inside it. Installments that arrive later go to a dated follow-up readout and never top up the original.
The trap is the resolved part of a young cohort. A decline can arrive at any stage, but paid status can arrive only after review, a signed agreement, a handoff and the Acquisition Partner's final written acceptance, and the partner's reward falls due up to 60 calendar days after the later of that acceptance and the completed handoff. Cases that have resolved early are therefore mostly declines, and a study run on them understates value and flatters drivers that merely speed up a no. Analyze a cohort only when all of it has resolved, or report the bounds the open cases allow, as the conversion benchmark does.
Zeros stay at zero. In the illustrative cohort, if those three paid companies came from 40 verified applications, dividing by 40 instead of 3 shrinks the per-case figure more than thirteenfold, and the smaller number is the one that tells a partner where effort pays. The average size of a paid deal is worth reporting too, but it answers a different question.
Coding the five candidate drivers before anyone knows the outcome
| Driver | Coded as | Coded when | Trap to report |
|---|---|---|---|
| Access | Who asked the CEO: the partner directly, one named intermediary, a longer chain, or the CEO unaware at submission | At submission | Large companies are often reached through chains, so access can stand in for size |
| Eligibility | How many of four fit facts were known and passing: peak full-time headcount, operating history, the rights question asked, the authorizer | At submission | Known fit facts can prevent declines but cannot create demand |
| Timing | A documented, dated event such as a migration, retention change or wind-down, and the days until it | At submission | Dated events cluster in distressed companies, where authority and rights are harder |
| Context | What the partner knew: worked inside the operation, knows systems and record span at metadata level, or name and title only | At submission | Partners with context may choose different companies, so context can stand in for selection |
| Follow-through | Partner actions in the first 30 days: CEO kept informed, scheduling questions answered, loop closed | Day 30, before review results | Later, promising cases attract more effort, and effort would look like a cause |
Follow-through never includes handling data; partners stay at metadata level throughout. The partner quality study asks which partner traits raise the share of introductions that pass review. This study asks where the dollars came from, so each driver is read against value per application and concentration, not against a pass rate alone.
Comparing outcomes without claiming causes
Report each driver level as a row: verified applications, paid cases, value per application with zeros, the median, the largest case's share, and value per application with the largest case removed. Five rules apply:
- Counts beside every rate and value; withhold any cell under 10 applications.
- The median beside the mean. With mostly zeros the median will usually be zero, and saying so is the point.
- Leave one out. If removing the largest case changes a driver's rank, report no stable finding.
- Stratify by peak headcount band and record type before comparing drivers.
- Describe, never attribute. Partners choose their drivers, so no comparison here shows that changing one would change value.
Rule 4 exists because pooled comparisons can mislead. The Stanford Encyclopedia of Philosophy describes Simpson's paradox: an association in a population can appear, vanish or reverse when the population is split into subpopulations. Illustrative and fictional, not SourceX data or an expected rate:
| Stratum | With a dated event | Without a dated event |
|---|---|---|
| Peak 200 or more full-time | 6 paid of 30 (20%) | 3 paid of 10 (30%) |
| Peak 50 to 199 full-time | 0 paid of 10 (0%) | 2 paid of 50 (4%) |
| Pooled | 6 paid of 40 (15%) | 5 paid of 60 (8.3%) |
Pooled, a dated event looks like it nearly doubles the paid rate. Within each band, cases without one did better. The pooled gap comes from where dated events fell, mostly in larger companies, not from timing. The ten-case cells in this example sit at the reporting floor, which is one more reason to read it as arithmetic only.
The cost and concentration of the strongest paths
A path is a combination of drivers, such as direct CEO access with four fit facts known. For each path, report:
| Measure | Why it matters |
|---|---|
| Partner hours, self-logged, including asks that never became introductions | Value per hour, not per introduction, decides where a partner spends a month |
| CEOs asked who declined | The relationship cost the program never sees; log it as in the rejection reasons study |
| Share of the cohort's total value | Whether the path matters at all |
| Largest-case and top-three shares | Whether the path is a pattern or one handoff |
| Distinct relationship owners behind its value | If one person produced most of it, the path is a person, not a method |
Report concentration in both company-payout and partner-reward terms; the floor and the Minimum Transaction mean they will differ. Stage timings come from the stage duration benchmark, so a slow path is not mistaken for a failing one.
Limitations and the next test
- Heavy tails. One large handoff can decide every comparison in a small cohort, which is why the leave-one-out check is mandatory.
- Long resolution. Paid status arrives last, so horizons must be long and young cohorts must wait.
- Program change. SourceX applies each program's own criteria, so a company with no fit in this cohort's program could still fit a later one. A zero is a zero at the horizon, not a permanent verdict.
- Reductions. If a supplier payout is later reduced or reversed, the reward is reduced too; record value as of the readout date and log adjustments.
- Attribution. Where a company already had a valid introduction, duplicate review leaves any later partner uncredited, so reconcile partner ledgers with program records first.
- Selection. Drivers are chosen, not assigned, so even a clean, stable difference is a description.
The next test: code the five drivers at submission for every verified application in two entry quarters, fix the horizon, strata and reporting floor in a dated plan before the first case resolves, and publish the first readout only once that cohort has resolved, with full counts even if no driver stands out. Until then the figures are not yet published, and any value-driver claim, from SourceX or anyone else, should be read against the rules above. Partners who want their introductions to count cleanly can start logging the five drivers now, register for a partner link first, and screen each company against who qualifies.