This design is for one partner, or several partners in one network, testing their own messages. It changes nothing on SourceX's side, and no company record passes to the partner at any stage.
Length is a choice about context, not about effort
A short note respects a busy CEO and leaves SourceX's own pages to do the explaining. A long note answers the questions a CEO would otherwise have to ask: why this company, what happens after an application, what nobody can know yet. Either could produce better-informed decisions. The short one may bring faster replies and more applications that fail an obvious fit fact; the long one may bring fewer, better-prepared applications, or simply go unread.
Neither belief is evidence. Randomly assigning comparable readers to two versions is the usual way to find out which one does better than the other; the UK Cabinet Office's paper Test, Learn, Adapt sets out nine steps for such trials in public policy and treats the randomly allocated comparison group as the feature that shows whether a change beats doing nothing new (our paraphrase of a search excerpt). A partner's test is far smaller, but the logic is the same.
Hypothesis, written before the first send. The long version will produce at least as many qualified applications per assigned company as the short version, with fewer applications that fail a published fit fact and fewer clarifying questions.
The required core that neither version may drop
Both versions open with the same core, word for word. The short version is the core plus one line; the long version is the core plus added context.
- Why you are writing: one sentence naming the CEO's earlier yes to hearing about SourceX.
- Who decides: "You decide whether your company talks to SourceX."
- Who could be paid: "SourceX may pay me a referral reward if your company licenses data; your company's payout is the same either way." This rests on section 4 of the terms, which says the reward "is paid separately by SourceX and is never deducted from the supplier’s agreed compensation."
- No records: "I don't need any files, samples or exports from you."
- Fit is SourceX's call: a link to SourceX's who qualifies page and the words "SourceX decides after its own review."
- Two exits of equal weight: where the company applies itself, and that a one-word no ends the matter.
The core must also stay inside what the sample Channel Partner Agreement allows: no line may tell a company that an estimate, positive interest, a signature or a data upload assures a purchase or a reward (section 4.6), and nothing may suggest the partner can commit SourceX to anything (section 7.1). Disclosure practice for client relationships is discussed in do I have to disclose a referral fee.
Two truthful versions, side by side (Illustrative)
Illustrative lengths; your own versions will differ, but record their actual word counts.
| Part | Short version, about 110 words | Long version, about 280 words |
|---|---|---|
| Required core | The six lines above, about 80 words | The same six lines, unchanged |
| Why this company | One line from public facts | A short paragraph from public facts and what the CEO told you |
| What happens next | Left to SourceX's pages | SourceX's published sequence: the company applies itself, SourceX reviews, the company decides what scope to offer |
| What nobody knows yet | Not stated | Fit, rights and any offer are decided only in SourceX's review |
| Claims made | Only the core | Only the core; more explanation, no further promise |
The last row is the truthfulness rule. If the long version says anything the short one does not, it must be an explanation of something SourceX has published, never a new claim. Otherwise the test measures persuasion, not context.
The context-controlled email experiment
Relationship type is the context most likely to swamp a length effect: a long note from a trusted adviser reads differently from a long note passed on by a sponsor. So assignment happens inside each relationship type.
| Element | Specification |
|---|---|
| Unit | A company whose CEO has said in writing that they would like to hear about SourceX |
| Context groups | Current client; former colleague or peer; second-degree, reached through a sponsor who asked the CEO |
| Assignment | Inside each group, take CEOs in the order their yes arrived and alternate versions, the first one chosen by a coin toss written into your log before anything is sent |
| Held constant | Sender, subject line, send within two working days of the yes, and one follow-up rule for both (one identical reminder after ten working days, or none) |
| Taken out before assignment | Any company you know already has contact with SourceX; after assignment, no company leaves the count |
| Entry window | Two quarters, fixed in advance |
| Reading date | 120 days after the final note goes out, late enough for every 90-day attribution window to close with time left for SourceX's review |
| Analysis | Each company stays in the version it was assigned, even if the CEO never read the note |
Primary measure and four guard measures
Primary: qualified applications per assigned company. Counts as qualified: the company applied, SourceX verified the application and credited it to your introduction, and on the reading date your dashboard does not show it as closed. Use the stage wording the dashboard displays, unedited; how it works lists the reward statuses it shows as Estimated, Eligible, Approved and Paid. Opens, visits and clicks are not outcomes here, and section 2 of the terms makes the same point about credit, which no click decides on its own.
| Guard measure | What it catches | Where it comes from |
|---|---|---|
| Clarifying questions per company | A short note that left out context the CEO needed | Your inbox |
| Days from message to the CEO's answer | Whether length slows a decision | Dates you already hold |
| Misreadings: "we qualify", "SourceX approached us", "the reward comes out of our payout" | Wording that misleads, in either version | The CEO's own words, if offered |
| Records, samples or exports sent to you | A protocol failure; the target is zero | Your inbox |
Declines are counted per version, and recorded without the private reasons behind them, as a seven-code decline taxonomy shows.
Sample floor, entry window and stop rules
Fix the floor before the first send. Percentages wait until both versions have at least 20 assigned companies and the entry window is shut; with fewer, the report shows raw counts for each version. Several partners in one network can combine their numbers, provided every one of them sends the same two texts, uses the same three context groups and reads results on the same date. What small counts can and cannot settle is discussed in the ten-introduction experiment.
Stop rules.
- A version gets a program fact wrong: both are withdrawn, corrected and sent again in a new entry window.
- A company sends files, samples or exports: that version is paused while its no-records line is made plainer, and whatever arrived is not kept.
- A CEO takes either version to mean the company qualifies, or that SourceX wrote it: rewrite before the next send.
- A new terms version or a revised screening page ends the current window; both texts are re-dated and a fresh window opens.
- Anyone picks a version for a company because it feels right: the assignment is broken, so start again from the logged order.
- An early lead for either version is no reason to stop; nothing is read before the reading date.
Reading the comparison when it closes
The report covers every assigned company, broken down by version and by context group.
| Count on the reading date | Short | Long |
|---|---|---|
| Companies assigned | To be counted | To be counted |
| CEO declined | To be counted | To be counted |
| No answer by the reading date | To be counted | To be counted |
| Applied | To be counted | To be counted |
| Qualified applications (primary) | To be counted | To be counted |
| Closed by SourceX | To be counted | To be counted |
| Clarifying questions | To be counted | To be counted |
| Misreadings | To be counted | To be counted |
| Records sent to the partner | To be counted | To be counted |
Decision rule, agreed in advance. Before the first send, write down how big a gap would change what you do (say, ten percentage points in qualified applications per assigned company), and commit to reporting the gap as an interval rather than one number.
| What the interval shows | What you do |
|---|---|
| Long ahead by more than your threshold, misreadings no higher | Adopt the long version |
| Short ahead by more than your threshold, misreadings no higher | Adopt the short version |
| The interval includes zero or straddles your threshold | No verdict: keep the version with fewer misreadings and protocol failures, and record that the test did not settle it |
| Either version shows more misreadings | Do not adopt it, whatever the primary count says |
| Records sent to the partner above zero in either version | Fix the wording before any comparison is read |
Splitting the result by context group helps you describe it; a gap inside one group is too small a base to crown a version. For a test of what makes the sender and the message credible, see the five-factor trust study; for a test that adds a self-check to a message rather than context, see the company fit checklist experiment.
Rival explanations the design has to rule out
- The sender. If one person sends most long notes, the test compares people. Keep one sender per context group.
- Timing. A version sent mostly near year-end meets busier CEOs. Alternating assignment spreads both versions across the window.
- Relationship strength. Context groups control the type of tie, not its warmth. Record the relationship holder's own rating of warmth before assignment and compare it across versions afterwards.
- Changing terms. A new terms version mid-window changes what both notes say. That is why it triggers a stop.
- Duplicates. A company SourceX already knew is credited to nobody, whichever version it received; checking before assignment keeps it out of both counts.
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.
Before sending either version, register as a partner and sign the Channel Partner Agreement, build the required core from the current who qualifies page and log the program terms version beside each note you send. Other partner experiments are listed in the guides hub.