How can referral quality be compared across partner roles without confusing quality with volume?
Divide each outcome by the number of introductions the role made, at the same age, and report the uncertainty. Counts answer "who is busiest"; rates at equal age answer "whose introductions tend to fit".
Three things blur a role comparison. Volume: a large group produces more of everything, good and bad. Age: an introduction made last month has had no time to reach review, so a role that joined recently looks weak. Mix: if one role mostly introduces larger companies, its rate may reflect company size rather than the partner's judgment. A sound study controls for all three before it says one role "introduces the most suitable companies". The existing guide on measuring quality instead of volume sets out what a good introduction looks like; this page sets out how to compare groups of them fairly.
Which milestones should the study measure?
Use milestones that a company reaches regardless of who introduced it, defined in the program terms or in SourceX's review, and never a partner's own activity.
| Measure | Numerator | Denominator | Why it is not a volume number |
|---|---|---|---|
| Application rate | Introductions that became a verified application | All introductions in the cohort | Attribution is credited at verified application, so this is the first fixed point |
| Fit rate | Verified applications that passed SourceX's eligibility review | All introductions in the cohort | Measures whether the company met the baseline, not how many were sent |
| Agreement rate | Companies that signed a licensing agreement | All introductions in the cohort | Reached only after rights and authority checks |
| Paid rate | Companies whose handoff led to a paid reward | All introductions in the cohort | The slowest milestone; report only for old cohorts |
| Open share | Introductions still in progress at the cutoff | All introductions in the cohort | Shows how much of the answer is still unknown |
Keep the denominator the same across rows. Switching to "of those reviewed" or "of those closed" makes a role with many pending cases look better or worse depending on which cases happen to have finished. Sibling studies on conversion denominators and stage durations treat those two problems in detail.
A worked comparison: four roles, one cohort
Illustrative. These are fictional numbers chosen to show the method; they are not SourceX results. Cohort: introductions made in one calendar half-year, each observed for 180 days. Measure: fit rate.
| Partner role | Introductions | Verified applications | Passed eligibility review | Fit rate | 95% interval (Wilson) |
|---|---|---|---|---|---|
| MSPs | 48 | 12 | 5 | 10% | 5% to 22% |
| Management consultants | 22 | 8 | 4 | 18% | 7% to 39% |
| Accountants | 14 | 7 | 5 | 36% | 16% to 61% |
| Lawyers | 5 | 3 | 2 | Not reported | Fewer than 10 introductions |
Read by count, MSPs and accountants tie at five companies each, and MSPs look like the stronger channel because they sent more. Read by rate, accountants come first and MSPs last. Read with the intervals, the honest conclusion is narrower: accountants may be ahead, but every interval overlaps, so this cohort cannot prove a difference. The lawyers' row is withheld rather than shown as 40%, because two companies out of five says almost nothing.
The Wilson interval is a standard way to put bounds on a proportion from a small sample; any equivalent method works if it is named. What matters is that the reader sees how wide the uncertainty is before seeing a ranking.
Three traps, and the control for each
Age. If one role's introductions are mostly a month old and another's a year old, the young cohort has not had time to reach review. Statisticians call this censoring: the observation window closed before every case reached the outcome. The control is a fixed observation age for every introduction, and an open-share column so readers can see how much is still pending.
Mix. Simpson's paradox describes how a relationship can reverse when a population is split into groups. If accountants mostly introduce 60-person firms and consultants mostly introduce 400-person firms, the role comparison may simply be a company-size comparison. The control is to report the role rates within strata, such as peak headcount band or record type, and only then the combined figure.
Selection. Partners who joined early may be more engaged than those who joined later. The control is to compare cohorts by join date as well as by role, and to state which partners were included.
What to record for each introduction
Everything needed for the study is metadata the program already holds or the partner already knows. Partners never send raw records.
| Field | Example value | Source |
|---|---|---|
| Partner role | Accountant | Partner's registration |
| Introduction date | 2026-03-14 | Submission |
| Company peak headcount band | 50 to 99 full-time | Owner's answer, verified later |
| Main record type | Project histories | Owner's answer |
| Milestone dates | Application, review, agreement, payment | SourceX status |
| Status at cutoff | Open, declined, withdrawn, paid | SourceX status |
A partner who wants to run this on their own book can keep the same columns, as the guide on tracking introductions by network segment suggests, and wait until each row is at least 180 days old before computing a rate.
Which companies belong in the study at all?
Only introductions of companies that plausibly met the baseline, so the comparison measures judgment rather than misunderstanding. The introduced company needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license its records and an authorized contact who can discuss a license. 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.
Before introducing, a partner can verify four things at metadata level: the peak headcount band, the main systems and how many years they cover, whether customer contracts restrict reuse, and which executive can authorize a licensing discussion. What stays unknown until SourceX reviews the company: whether its records will be selected, the rights review result, any offer and any reward.
Quality and the reward are separate questions
A better rate does not change the reward formula. Company compensation is what SourceX pays the referred company under its licensing agreement; the referral reward is funded separately and is never deducted from it. Every role earns on the same default terms, described on the referral partner explainer, so a quality study is a tool for deciding where to spend introduction effort, not a pay grade. An accountant planning a first month can use the accountants' referral plan alongside it.
Limits and unanswered questions
SourceX has not published role-level quality figures. A credible first release would state the cohort dates, the observation age, every denominator, the interval method, the strata used and the number of partners per role, and would withhold any cell with fewer than ten introductions. Until then, treat any claim that one profession "refers best" as a hypothesis.
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 introductions would make a strong row in this table, register as a partner, screen each company against who qualifies, and read the program terms for how attribution is credited.
Why can one documented company matter more than a giant contact list?
Because an introduction only counts when the company passes the fit tests and its CEO wants the conversation, and a contact list records neither. A partner who has worked inside one company's operations can usually answer the size, records, rights and authority questions before asking; a list of 1,200 names answers none of them. Whether that context raises the share of introductions that qualify is what the comparison below tests. SourceX has not published results, and nothing here shows that small networks always win.
- Known: SourceX screens for roughly 50 or more full-time employees at peak, several years in operation, records the company itself may license, and a decision maker who knows about the introduction. The CEO authorizes a licensing discussion.
- Conditional: whether an introduced company is accepted, licenses anything or leads to a reward is decided only in SourceX's review and a signed agreement.
- Thesis under test: partners who know a company's operating history reach its CEO more often and waste fewer asks, even when they introduce fewer companies.
| Term | Meaning in this study |
|---|---|
| Documented company | One whose headcount history, main systems, record span and CEO the partner can describe from direct work with it |
| Owner access | A CEO the partner can ask directly, or through one named person who will ask |
| Upstream rejection | An ask that never became an introduction: the CEO declined or never replied, or the partner screened the company out |
| All-outcome ledger | One row per ask, from first conversation to last recorded status, with no row deleted |
| Fit rate | Introductions that passed SourceX's eligibility review, divided by all introductions in the cohort |
The small-specialist versus broad-network cohort comparison
Illustrative. No partner or company interviews are approved for publication, so these cohorts are fictional. The specialist cohort is 18 partners who introduce only companies they have worked inside, such as fractional CFOs and sector accountants; the broad cohort is 22 partners working from large contact lists. Every ask made in one half-year is followed until each introduction is 180 days old. Partner hours are self-reported: 170 for specialists, 540 for the broad cohort.
| Measure | Specialist cohort | Broad cohort | What it shows |
|---|---|---|---|
| CEOs asked | 52 | 230 | Reach |
| CEO declined or never replied | 11 | 134 | Upstream rejections the program never sees |
| Introductions submitted | 41 | 96 | The volume the program sees |
| Verified applications | 26 | 41 | Where attribution is credited |
| Passed eligibility review | 15 | 19 | The broad cohort wins on count |
| Fit rate per introduction, 95% Wilson interval | 37% (24% to 52%) | 20% (13% to 29%) | Intervals overlap: no proven difference |
| Passed review per CEO asked | 29% (18% to 42%) | 8% (5% to 13%) | The gap opens before the introduction |
| Partner hours per company passing review | 11 | 28 | Effort |
By count, the broad cohort leads, 19 eligible companies to 15. Per introduction, the specialists look better, but the intervals overlap. Per CEO asked, the gap is large, and almost all of it sits in the row no program records: 11 declines against 134. Were these figures real, the finding would be that context buys owner access and saves asks and hours, not that specialists' companies fare better once introduced.
Check mix first: if most specialists served one industry where demand happened to be strong, the comparison measures the industry. Report rates within headcount band and record type, as the mix control above sets out.
One relationship, from discovery to a conditional outcome
Illustrative. A sector accountant has closed the books for a 130-person specialty distributor for eight years. She knows its ERP has held orders, returns and pricing exceptions since 2016, that its customer contracts do not cover its own operating records, and that the founder is still CEO. She asks him; he wants the conversation; she submits the introduction, and the company completes a verified application. The outcome stays conditional: rights review, current demand and any agreement are SourceX's to decide.
Her other large client, a 210-person software company, is a no-fit she never introduces. Most of its useful records sit in customer tenants, and SourceX's screening guidance treats material that belongs to clients as usually not a fit. In her ledger it is one row coded "screened out: client-owned records", a rejection a contact-list approach usually discovers only after submitting.
When scale wins and when context wins
| Condition | Favors | Why |
|---|---|---|
| Most of your clients are below 50 full-time employees at peak | Scale | Deep knowledge of companies that fail the size test produces nothing |
| Your list is CEOs, not functional managers | Scale | Owner access is already high, so reach converts |
| Each declined ask spends a client's goodwill | Context | The broad cohort spent 134 declined asks to find 19 companies |
| Rights hinge on contract detail | Context | Only someone who has seen the contracts can screen early |
| You have a few hours a month | Context | Hours per eligible company is the binding limit |
The recommendation changes if a real cohort shows the specialist advantage vanishing within headcount and record-type strata, or if broad-list partners who confirm each contact is the CEO match the specialists' upstream rate. The comparison with B2B lead generation explains why a good sales contact can still be a poor data introduction.
What should a VC platform research director know about referral quality benchmark methodology?
A referral benchmark is only as honest as its denominator. Keep every ask and every introduction in it, including CEO declines, SourceX declines, withdrawals, duplicates and cases still open; publish the protocol before the first cohort closes; and report rates as descriptions, never as causes. SourceX has published no benchmark results. What follows is a protocol a platform team can adopt for its own network's introductions, and a standard to hold any program's published numbers to.
Much partnership research surveys program leaders: PartnerStack's State of Partnerships in GTM 2026 draws on 100 senior leaders at B2B SaaS companies with $50M+ revenue, surveyed by Wynter, and reports that only 42% of companies use multi-touch attribution for partner revenue. A platform team that logs each introduction to its outcome holds evidence a survey cannot, provided the failures stay in.
The all-outcome benchmark protocol
| Element | Fixed in advance | Why rejections stay in |
|---|---|---|
| Registration | A dated plan posted before the first cohort closes, with every later change dated | Results cannot be reshaped after declines are seen |
| Sampling frame | Every ask and introduction by consenting partners in a fixed entry window | No case leaves because its outcome was poor |
| Consent | Partners opt in before their window opens | Nobody joins after a good quarter or leaves after a bad one |
| Outcome codes | Paid, agreement signed, in review, declined with one primary code, withdrawn, duplicate or outside the window, open at cutoff, plus the upstream codes | Every case has exactly one status at the cutoff |
| Observation window | Long enough for payment, due 60 calendar days at most after both the Acquisition Partner's final written acceptance and completed handoff | Slow successes are not counted as failures |
| Analysis | One denominator, Wilson intervals, strata named in the plan, cells under 10 introductions withheld | Small groups cannot produce headline rates |
| Anonymization | Companies under codes; owners who declined never named outside the partner's own ledger | Partners can report rejections without exposing a client |
Decline codes come from the rejection reasons study, censoring rules from the stage duration benchmark. The Center for Open Science describes preregistration as specifying a research plan in advance and submitting it to a registry, separating planned from unplanned analysis.
What could make the benchmark wrong.
- Self-selection. Partners who opt in may be more careful; report how many declined to take part.
- Recall. Upstream asks are self-reported; log them as they happen.
- Gaming. A published fit rate rises if partners stop introducing borderline companies; show counts beside rates.
- Program change. SourceX notes that a company declined for one program can fit a later program, so one cohort's decline is not a permanent verdict.
- Attribution gaps. A company with an earlier valid introduction is reviewed as a duplicate, so a partner's ledger and program records can disagree; reconcile them first.
Do larger professional networks produce better referrals?
Nobody has shown it for company data introductions, and SourceX has not published data that could. A larger network will almost certainly produce more introductions; whether a larger share fit is open, and the illustration above suggests the answer may differ before and after the introduction. The study treats size as one more stratum in the role comparison: define it, fix bands before outcomes exist, and compare within role and company profile.
Total contacts are easy to count and say little. The study records three numbers at entry and uses the third as its main measure: total professional contacts; companies worked with in the past three years; and reachable CEOs, meaning chief executives the partner can ask directly or through one named person. Bands are set in the plan: fewer than 25 reachable CEOs, 25 to 99, and 100 or more.
| Question | Measure | Compared within | What would answer it |
|---|---|---|---|
| Do bigger networks introduce better-fitting companies? | Fit rate per introduction | Role and headcount band | Rates rising with band in every stratum |
| Do bigger networks waste more asks? | CEO agreement rate per ask | Role | Agreement falling as band rises |
| Does reach add eligible companies? | Companies passing review per partner | Role and program tenure | Count rising faster than hours spent |
| Is any gap really company mix? | All three measures | Record type and sector | A gap that vanishes within strata |
Two traps are specific to size. Successful partners may grow their networks, so measure size at entry, not after outcomes. And size can travel with role: if bankers in the study know more owners than accountants do, a size effect may be a role effect. Groups like those on the professional referral networks page should record which member made each ask, since one partner is credited per company. Even a clean difference would be descriptive: partners choose their networks and companies, so no study of this kind shows that adding contacts would change a rate.
What do successful introductions have in common, and how would anyone know?
Only a comparison that keeps the failures can say. Studying paid introductions alone finds traits that declined ones shared too. The plan records candidate factors at submission, before anyone knows the outcome, then compares how often each appears among paid, open, declined and withdrawn cases. SourceX has not published results, and every factor below is a question, not a finding.
| Factor recorded at submission | How it is recorded | Would support it | Would undercut it |
|---|---|---|---|
| CEO knew and agreed before submission | Yes or no, with date | Far more common among companies passing review | Equally common among declines |
| Peak headcount documented | Source named, such as a payroll report or the owner's statement | Fewer size declines | No change in size declines |
| Record span and system named | Years covered and system, metadata only | Fewer data-fit declines | Data-fit declines unchanged |
| Rights question asked | Whether client contracts restrict reuse of the company's own records | Fewer rights declines | Rights declines unchanged |
| Partner has worked inside operations | Yes or no | Higher fit rate within role and band | Advantage gone within strata |
| A dated event | A documented migration, archive deletion or retention change | Shorter time to review | No change in time to review |
Three rules keep it honest: exclude factors entered after the outcome, since hindsight fills them in; report every outcome group, open cases included; and report how common each factor is, because one present in nearly every introduction explains nothing.
The counterexample to expect, Illustrative: a 300-person freight broker introduced with all six factors present is declined because there is no current dataset demand for its dispatch records. Nothing the partner could check would have changed that; why a well-documented company can still be declined covers the case. The reverse belongs in the data too: an introduction missing three factors that passes review anyway. The hypothetical approved referral shows the steps of one accepted case; this plan asks how often such cases differ from the ones that failed.
A measurable next step. Add the six factor columns and an upstream-ask log to your tracker now. When your first ten introductions are 180 days old, compare factors across outcomes and count your CEO declines. With fewer than ten, keep recording and conclude nothing.
Taking part in the study
Taking part is optional, and partners share metadata only: role, network size at entry, dates, factor answers and coded outcomes, never client records, samples or credentials. A partner who publishes results from their own book, or recommends SourceX publicly while earning rewards, should say so plainly: the FTC's endorsement guidance lists monetary payment among the material connections to disclose. Program rules still apply: sign the SourceX Channel Partner Agreement before submitting; credit requires the first valid introduction in the 90-day attribution window plus the company's verified application. Other study designs sit in the guides library.
If your book holds a company you know from the inside, led by a CEO open to the conversation, request a partner link, compare the company with SourceX's fit criteria and study the program terms before you ask.
Evaluating SourceX as a company is a separate decision from taking part in referral research; anyone making it can ask for investor materials via 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.