# Referral message controlled test: one variable | SourceX

Canonical: https://refer.sourcex.si/guides/how-to-compare-two-outreach-angles-without-changing-everything
Last updated: 2026-11-04

Change one thing and freeze everything else. Write two truthful versions of the opening paragraph, each giving the CEO a different reason to talk, and keep the sender, the list rule, the subject line, the disclosure, the ask, the links, the timing and the follow-up word-for-word identical. Assign CEOs to the two versions from a list drawn before you write a line, then count verified applications per arm at a fixed cutoff. Opens, clicks and reply totals decide nothing. No partner or SourceX results for this comparison exist yet; this page is the design.

Most "tests" of referral outreach compare two different emails, sent by two people in two different months to two kinds of contact. When the second one does better, nobody can say whether the angle, the sender, the season or the list did the work. The worksheet below lists every element that can drift between arms and pins each one down, so the only difference left is the one you meant to test.

## What an angle is, and the truthfulness bar both versions must clear

An angle is the reason your message gives a CEO for a conversation now. Two examples, written for this page and labeled Illustrative:

- **Angle A, the documented event.** "When we spoke in June you mentioned the old service desk is being retired next quarter. Before that archive moves, it may be worth asking whether years of your own case history could be licensed."
- **Angle B, the program.** "I have started introducing established companies to SourceX, which licenses business records directly from the companies that created them, for AI training and evaluation. Your firm's long project history made me think of you."

Both versions must clear the same bar before either is sent:

1. **Every statement is true of this CEO's company.** Angle A needs an event the CEO actually told you about or the company announced; a migration you guessed at is invented urgency. If the event is not real for a given company, that company cannot receive angle A, and it comes off the eligible list before the draw, not after.
2. **Neither version implies the company qualifies.** SourceX describes its fit page as screening guidance and applies the criteria of each specific program; fit is decided in its review, never in your email.
3. **Equal promise strength.** If one version says "may be worth asking", the other cannot say "will be worth it".
4. **No amounts in the angle.** Money belongs in the frozen disclosure sentence, which both arms share word for word.
5. **The CEO decides.** Both versions are addressed to the CEO, who opens a licensing discussion and signs for the company under the program; an angle aimed at a deputy changes the recipient, which is a second variable.

A [message review for false urgency](/guides/the-message-review-that-removes-false-urgency) is a useful pass over both drafts before the freeze.

## The single-variable test worksheet

Fill in every row before allocation. Exactly one row may read "varied"; if two do, you are running a different experiment.

| Element | Arm A | Arm B | Status | How you hold it |
|---|---|---|---|---|
| Opening paragraph | Event text, version 1 | Program text, version 1 | Varied: the only one | Both dated and frozen before the draw; lengths within 15 words of each other |
| Sender and signature | You | You | Fixed | One sender and one account for both arms; no colleague sends a share |
| Who is eligible | CEOs you know in your own right | Same list | Fixed | One eligible list, written before either angle is drafted |
| Subject line | "A question for you" | "A question for you" | Fixed | Neutral wording that carries neither angle |
| Disclosure | Identical sentence | Identical sentence | Fixed | States plainly that SourceX may pay you if the company goes ahead |
| The ask | Reply yes, no or not now | Same | Fixed | One question, the same three reply words |
| Links | Who-qualifies, then the terms | Same | Fixed | Same pages, same order, no tracking redirects |
| Rest of the body | Identical text | Identical text | Fixed | Copied from one master file |
| Send window | Interleaved | Interleaved | Fixed | Alternate arms within each sending day, same hour band |
| Follow-up | One identical note after ten days | Same | Fixed | Or none for both; decide before the first send |
| After a yes | Same next step | Same | Fixed | You submit one introduction; the company applies itself |
| Channel | Email | Email | Fixed | A call or a chat message is a different treatment |

**Eligible list rules.** Include only CEOs you can contact in your own right, whose companies you have no reason to think already deal with SourceX, and whom you have never asked about SourceX before. Under section 3 of the terms, a company with an earlier application, relationship or valid introduction is a duplicate and "not credited to a later partner", so remove known duplicates before the draw. Anything removed afterwards is reported as removed, with the reason, and stays counted in its arm.

**The draw.** Give every name on the list a random number, sort, and assign alternately down the sorted list. If your list mixes CEOs you speak to monthly with CEOs you have not seen in years, draw within each of those two groups separately so neither arm collects all the warm relationships. NIST's handbook describes this shape of test, one factor with its levels assigned to units at random, as a completely randomized design, and notes that equal numbers per level make the comparison most sensitive (paraphrase). Keep the arms the same size.

## The count that decides: verified applications per allocated CEO

**Primary measure.** For each arm: the number of allocated companies that, by the cutoff, have submitted a verified application credited to your introduction and are not closed as duplicates, divided by the number of CEOs allocated to that arm. Read the stage from your partner dashboard and copy its words exactly; SourceX's [how it works](/how-it-works) page says the dashboard shows each referral's stage and its reward status.

Two rules from the terms shape this measure. Section 2 gives credit to whichever valid introduction of a company new to SourceX came first inside the 90-day attribution window, and only "when that company submits a verified application"; the same section adds that "an earlier link click alone does not decide credit". So a click is not an outcome, and the books close only after the window has had time to run: set the cutoff at least 90 days after your last send, plus a margin for review that you name in advance.

**Supporting measures**, recorded per arm and never used on their own to pick a winner:

| Measure | What it tells you | Source of the count |
|---|---|---|
| Written yes from the CEO | Early interest; a yes that never becomes an application is still not an outcome | The CEO's reply |
| Not now | Timing, not rejection | The reply word |
| Please stop, or a complaint | The angle is costing goodwill; this is a veto | The reply |
| Misreadings | The CEO took it as an endorsement, an approval or a promise of fit | What the CEO wrote |
| Questions about your payment | Whether the shared disclosure landed | Your inbox |
| Files or records sent to you | A protocol failure; the target is zero | Your inbox |
| Duplicates found | Should be similar in both arms; a gap points to a flawed draw | The dashboard |

You never ask for records and never handle them. SourceX's process page says the partner does not need to export, upload or describe confidential records, and the company controls whether it proceeds. If a CEO sends files anyway, do not open or keep them, say you do not need them, and log the event. Declines are recorded without the private reasons behind them, as the [seven-code decline taxonomy](/guides/how-to-record-a-company-s-decline-without-keeping-its-private-reasons) shows.

## Minimum sample, duration and the rules for stopping early

Set two limits before the first send: a total number of verified applications across both arms at which you will read the result, and a calendar end date. Compare the arms once, when the first of the two is reached. Twenty verified applications in total is a reasonable floor; fewer cannot separate two angles unless one is dramatically better, and [the five-replies page](/guides/how-to-avoid-declaring-a-winner-from-five-replies) shows the arithmetic and a threshold table to adopt with it. SourceX publishes no conversion rates, so you cannot calculate in advance how many CEOs that takes. If the end date comes first, report counts and no winner.

A partner whose network will never produce twenty verified applications in a reasonable window has two honest options: pool the test with other partners who use exactly the same frozen text and draw rule, or accept that a test cannot settle the question at this scale and choose the angle on truthfulness and relevance alone, saying so.

**Stop or pause when:**

- An angle proves untrue for a company it reached (the migration was cancelled, the CEO had left). Correct it with that CEO, keep the company in its assigned arm for the analysis, and fix the text if the error was in the template.
- One arm draws two requests to stop or complaints before the floor is reached. Pause that arm and review its wording; the paused count is reported.
- SourceX publishes a new version of the terms or changes who-qualifies mid-test. Stop, note the versions, and start a fresh comparison; mixed versions muddy the comparison.
- A frozen element drifts: a colleague edits the master file, a send goes out from another account. Log every affected send; if more than a handful are affected, restart.
- Never stop because one arm looks ahead. The stopping rule watches the combined total, never which arm is ahead.

## Reading the result, including the result that names no winner

Analyze every allocated CEO in the arm the list assigned, whatever happened afterwards. One table, at the cutoff:

| Count at cutoff | Arm A | Arm B |
|---|---|---|
| CEOs allocated | Not yet collected | Not yet collected |
| Removed after allocation, with reasons | Not yet collected | Not yet collected |
| Written yes | Not yet collected | Not yet collected |
| Please stop or complaint | Not yet collected | Not yet collected |
| Verified applications credited to the introduction | Not yet collected | Not yet collected |
| Closed as duplicates | Not yet collected | Not yet collected |
| Files or records sent to the partner | Not yet collected | Not yet collected |

| What the counts show | Decision |
|---|---|
| The difference in verified applications passes the decision rule you adopted before the first send, and the angle ahead has no more complaints or misreadings | Use that angle for the same kind of CEO, and retest when your network or SourceX's pages change |
| The difference does not pass that rule | No winner. Keep your current angle; the test could not tell the two apart, which is not proof they are equal |
| The angle ahead also draws more complaints or misreadings | Veto: rewrite it before anyone uses it |
| Any files or records reached you | Fix the wording first; the comparison waits |

What the comparison can never tell you is whether a company will be accepted or whether any reward follows. SourceX decides fit in its review, under the program the company is assessed against, and the terms say a lead, meeting or signed agreement alone does not earn a payable reward. Cuts by sector or company size are descriptive; with small counts, some cut will always look striking by chance.

## Where a one-variable comparison still misleads

- **An angle is a bundle.** The opening paragraph differs in many words at once, so a result is about that paragraph, not about any single phrase in it.
- **You are the strongest variable.** The draw spreads your relationships across arms, but a short list stays lumpy; the within-group draw helps and does not cure it.
- **CEOs compare notes.** Members of one peer group may see both versions. Note any you know of; do not drop them.
- **The world moves during a long window.** Interleaving sends protects against season and news, but not against a change to SourceX's own pages, which is why a new version restarts the test.
- **Your network is not someone else's.** A result describes your CEOs in your period. Another partner who wants it must run it.

For the next step up in complexity, several message factors tested together, see the [five-factor trust study](/guides/what-makes-a-company-introduction-feel-trustworthy); for the wording of the shared disclosure, see [how to disclose a reward naturally](/guides/how-to-disclose-a-reward-naturally-in-an-introductory-email).

## 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 the draw, [register as a partner](/register) and sign the Channel Partner Agreement, freeze both angles against the current [who qualifies](/who-qualifies) page, and write the [program terms](/terms) version into your test log; related designs sit in the [guides hub](/guides).

## Sources

- https://refer.sourcex.si/terms
- https://refer.sourcex.si/who-qualifies
- https://refer.sourcex.si/how-it-works
- https://www.itl.nist.gov/div898/handbook/pri/section3/pri331.htm