A test answers which, an interview answers why
An experiment tells you which of two versions did better on a count you chose. It cannot tell you why people behave as they do, and it cannot suggest a version you never thought to try. Interviews do the reverse: they show the reasons people give and the steps where they stop, but they cannot say how common any reason is.
The usual trap looks like this. Partner registrations rise; introductions stay flat. The team proposes a monthly reminder, an FAQ and a workshop, all at once. Each is a fix for a different cause: forgetting, unanswered questions, missing skills. Picking one to test is picking a cause without evidence, and a null result will not say whether the fix was weak or aimed at the wrong problem.
Government service teams have a name for the step that comes first. The GOV.UK Service Manual describes discovery, in paraphrase, as the work done before a team commits to building anything, warns teams not to start building during it, and advises a team handed a solution to question it and restate it as a problem. In a partner program that means turning "we need a reminder email" into "registered partners make no introductions, and we do not know why."
The research-before-test decision guide
Two questions sort every conversion problem. Do you know its cause, from the people it happens to rather than from your own guess? And can you count the outcome a fix should move, inside the window the program allows?
| Outcome countable within 90 days | Outcome not yet countable | |
|---|---|---|
| Cause unknown | Interview first, then test the fix the interviews point to | Interview first, and build a codebook for the outcome while you do |
| Cause known | Test now, with failure criteria written before the first send | Define the measure and pilot it on past cases before testing anything |
Signals that put a problem in the top row:
- You can count the drop but cannot name the stage where people stop.
- Three or more fixes are on the table at once.
- The fixes are things the team would like to build, not things anyone asked for.
- The people who stop tell you nothing; silence is the main evidence.
- Your explanation came from your most active partners, not from the quiet ones.
Signals that the outcome is not yet countable: the goal is phrased as "better introductions" or "more trust" with no definition; two colleagues score the same five past cases differently; or the outcome only shows up after the 90-day window, so a test would end before it could be read. Under section 2 of the terms, credit goes to the first valid introduction of a new company inside that window and only once the company submits a verified application. That makes the verified application the natural countable outcome for most partner tests, and it explains why submissions alone are a weak stand-in.
For the bottom-right cell, the design rules in a referral experiment with clear failure criteria apply.
Worked case: registered partners who never introduce anyone
Everything below is a plan, published before any interview, so that what was intended can be told apart from what turned up.
| Element | Plan (Illustrative) |
|---|---|
| Sampling frame | Partners registered 90 days or more with no introduction submitted ("silent"), plus a small contrast group who have submitted at least one |
| Sample | Up to twelve silent partners spread across partner types (accountants, consultants, investors, association staff) and four contrast partners; stop sooner if three interviews in a row add no new reason |
| Invitation and consent | One invitation; a stated purpose ("to learn why the program is not useful to some partners"); recording only with permission; no thank-you tied to making introductions; withdrawal at any time, notes deleted on request |
| Interview guide | Eight questions about recent behavior: the last time a company came to mind, what happened next, what you would have needed to know, whom you would have asked, what made you hesitate, what you believed a reward depended on, what you would have said to a CEO, and anything that made you decide against it |
| Never asked | Company names, client details, anything about a company's records, or whether a client would fit |
| Recorded | Partner type, months since registration, the stage where they stopped, reason codes, and any misreading of a program rule, checked against the terms |
| Analysis | Two people code each interview with one shared codebook, compare, and settle differences in writing; each reason is reported with how many partners raised it |
| Reporting | Use COREQ as the reporting checklist; publish the guide, the codebook and every reason, including those raised once |
COREQ is listed by the EQUATOR Network as a 32-item checklist for reporting interview and focus-group studies (Tong, Sainsbury and Craig, 2007). Working through it before publishing keeps the method visible beside the findings.
The stages to code come from the partner's side of the program, not from SourceX's: no company in mind; a company in mind but no path to its CEO; a path, but hesitation; a start that stopped at the form. A reason for silence that lives in SourceX's review (a declined company, for instance) belongs to a different study; where and why company introductions stall is that study.
Who is missing, who said no, and what interviews cannot say
- Partners who decline the interview. Count them and publish the count. The silent partners least likely to answer are the most silent of all, so the twelve who talk may differ from the ones who do not.
- The contrast group. It shows what differs between silent and active partners, not what causes the difference.
- Descriptive, never causal. "Five of twelve said no company came to mind" describes twelve people. It is not a rate for the program, and it is never turned into a percentage.
- What people say about themselves. Partners may give the answer they think the interviewer wants. Questions about recent, specific events reduce this; they do not remove it.
From reasons to the test each one justifies
| Reason heard | What it suggests | Next step |
|---|---|---|
| No company came to mind that fits | Fit knowledge, or a network with few fitting companies | A fit-tool test, or accept a smaller active group |
| A company, but no path to its CEO | A relationship gap, not a message problem | No message test; map who holds each CEO relationship |
| Unsure when a reward would be paid | A misread payment trigger | Measure the misreading first with a payment-trigger survey |
| Worried the ask would strain the relationship | Ask wording and trust | Compare two ask openings with trust measures |
| Not sure their professional rules allow it | Not a testing question | The partner's own counsel or ethics adviser answers it; nothing in the program can |
| Simply forgot | Attention | A reminder test with written failure criteria |
| Lost at the form | Usability | The form walkthrough |
| The same question kept coming up | Missing answers | The FAQ experiment |
Misreadings deserve a direct check against the program's own words. Section 5 of the terms says, "A lead, meeting or signed agreement alone does not earn a payable reward." A partner who believed a meeting was enough has a different problem from one who never thought of a company.
Interpreting what you heard, and deciding what to change
Interviews show which reasons exist and how they connect; they do not rank them by size. If choosing between two fixes depends on which reason is more common, follow the interviews with a short survey of every silent partner, using the reason codes as answer options. Then pick one change, decide in writing which reason it addresses and what count would show it worked, and test it.
Some problems need neither interviews nor a test:
- A wrong statement in your own materials. If a partner page says companies "qualify" when only SourceX's review can decide, correct it. Testing a falsehood against the truth is not an experiment anyone should run.
- A program change. SourceX says its criteria follow each program and can move; a drop that lines up with a dated change may need new wording, not research.
- A cause already in the logs. Error messages or abandoned steps with timestamps can answer the question without asking anyone.
What partners must hear correctly whatever you learn
Whatever the interviews turn up, the facts partners pass on stay fixed. A company needs 50 or more full-time employees at peak, contractors not counted; documented records of its own; rights to license them; and a CEO who agrees to the discussion and authorizes it. 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.
Partners never collect or forward records; at most they describe what kind exist. Whether those records match a current program and whether contracts restrict them are settled in SourceX's review, and a company that misses one program may fit another later.
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.
Sort your problem in the grid before choosing a test. Partners who are ready to introduce a company can register, compare it with who qualifies and read the program terms first.