# New referral partner learning research: plan | SourceX

Canonical: https://refer.sourcex.si/guides/the-referral-learning-curve-for-a-new-partner-cohort
Last updated: 2026-11-04

No one has measured whether SourceX partners get better at introductions with practice, and this page publishes no curve. It sets out how to measure it: follow every partner approved in one quarter for twelve months, grade each introduction on a ladder of dated outcomes, and compare each partner's later introductions with their own first, so that strong partners who simply stay active do not pass for learning. Partners who never introduce anyone stay in the count. Results wait until the cohort closes.

The plan needs SourceX's own records of introductions and applications, and the consent of the partners whose records are used. No company is named anywhere in the analysis, and nothing in it changes how an introduction is reviewed or credited.

## What "learning" would have to mean

A program can look as if partners are learning when nothing of the kind is happening. Three different things get called a learning curve:

| What someone might measure | What it really shows | Used here? |
|---|---|---|
| More introductions per month as the cohort ages | Activity, which may fall or rise for reasons unrelated to skill | No, reported only as context |
| The cohort's average outcome rising over time | Partly who stayed: partners who stopped after one introduction no longer count in later months | No |
| A partner's later introductions reaching further than that same partner's first | Change within the same person, which is what learning would look like | Yes, as the primary comparison |

The third reading still is not proof. It is consistent with learning, and the rival explanations further down have to be ruled out before anyone says so. Whether a particular training changes results is a different question that needs assignment, as in [the onboarding experiment](/guides/the-partner-onboarding-experiment-that-measures-understanding) or [the enablement pilot](/guides/the-partner-enablement-test-before-a-large-network-rollout).

Four research questions:

1. Within partners, does the share of introductions reaching a verified application differ between a partner's first introduction and their later ones?
2. Do harm events, such as an introduction reaching SourceX before the CEO knew of it, become rarer with practice?
3. How long does a new partner take to make a first introduction, counting partners who never make one?
4. How many partners reach a second, third and fourth introduction at all?

## The cohort, the clock and who is left out

- **Cohort.** Every partner whose account is approved, and who has signed the SourceX Channel Partner Agreement, within one calendar quarter.
- **Clock.** Each partner's clock starts on their approval date and runs for twelve months.
- **Reading each introduction.** Every introduction is read once, 120 days after it was made: the 90-day attribution window plus a month for records to settle. Introductions made late in the year are still read at day 120, so the study closes 120 days after the last partner's twelve months end.
- **Consent.** Partners agree to their dashboard records and their own introduction logs being analyzed under coded identifiers. Partners who decline appear in the ledger as a count and nothing more.
- **Excluded:** SourceX staff, and partners who held an earlier account, since they are not new.

## The outcome ladder for every introduction

Each rung is a dated event with a named source. A missing record is coded "not recorded", never "no".

| Rung | Event | Where the dated record comes from |
|---|---|---|
| R0 | Introduction made | The dashboard submission or a dated written introduction |
| R1 | The CEO agreed in writing before the introduction went in | The partner's own log, shared with consent |
| R2 | The company was new to SourceX | SourceX's duplicate review under section 3 of the terms |
| R3 | The company submitted a verified application within the attribution window | SourceX's records under section 2 of the terms |
| R4 | The company passed SourceX's fit review | SourceX's review record; SourceX decides, the study only counts |
| R5 | Later dashboard statuses | The partner dashboard's Estimated, Eligible, Approved and Paid statuses |

R3 is the primary rung: it arrives within months, it depends on the partner's choices as well as the company's, and the terms define it. R4 also reflects what SourceX needs at the time; its screening page notes that each program applies its own criteria and that a company missing one program may fit another later, so a change in R4 can mean demand moved, not that partners improved. R5 is too slow and too sparse for a first-year cohort and is reported as counts only.

Three harm events are coded alongside, from SourceX's review notes and any complaints, by coders who do not see the introduction's order number: H1, the partner offered or collected company records; H2, the partner told the company it would fit; H3, the introduction reached SourceX before the CEO knew of it, which SourceX's screening page says is usually not a fit. A harm event is handled by the program as it happens; the study only records it.

## The cohort learning measurement plan

| Element | Setting |
|---|---|
| Unit of analysis | One introduction, indexed by its order within the partner: first, second, third, fourth or later |
| Primary comparison | Each partner's later introductions against the same partner's first, among partners with two or more |
| Primary measure | Share of introductions reaching R3 |
| Secondary measures | R1, R2 and R4 shares; harm events H1 to H3; time to first introduction |
| Time to first introduction | A survival estimate in which partners with no introduction at the cutoff stay in as censored |
| Survivorship table | The number of partners reaching each order, printed beside every estimate |
| Recorded adjustments | Terms version in force, calendar month, practice type, any training attended (recorded, never assigned) |
| Floor before any within-partner estimate | 40 partners with two or more read introductions; below that, counts only |
| Preregistration | This plan, the ladder and the coding rules dated before the first partner is approved |

The survival estimate matters because silence is the commonest outcome to drop by accident. A survival tutorial in PubMed Central describes the method as using both participants who reach the event and those who are censored, and defines the median as the time by which the event has happened to half of them (Lira and colleagues, 2020). Here the event is a first introduction. If fewer than half of the cohort ever make one inside twelve months, the median is simply not reached, and the report says so rather than quoting a time.

The terms version matters because the program itself moves. Section 11 of the terms lists dated changes across five versions in October 2026 alone; every introduction carries the version in force on the day it was made.

## Rival explanations a rising line must survive

1. **Survivorship.** Partners whose first introduction went nowhere may stop, leaving only the more capable in later columns. The within-partner comparison handles this for partners with two or more introductions; the survivorship table shows how many that is.
2. **Order of companies.** Partners may introduce their best-fit company first and weaker ones later, which would make later introductions look worse with no loss of skill. Partners report, at R0, whether this was the company they had in mind first; it is recorded, not judged.
3. **Changing demand.** A rise or fall in R4 can follow a change in SourceX's programs rather than in partners. That is why R3, not R4, is primary.
4. **Training given along the way.** Some partners will attend sessions or read guides. Their attendance is recorded; any difference by attendance is described, never credited to the training.
5. **Small numbers.** A cohort may produce few second introductions. Below the floor, the report gives counts and stops there.

## The ledger, and the table that stays empty

| Ledger line | Count |
|---|---|
| Partners approved in the quarter | Not yet counted |
| Declined consent | Not yet counted |
| Analyzed | Not yet counted |
| No introduction by the end of their twelve months | Not yet counted |
| Exactly one introduction | Not yet counted |
| Two or more introductions | Not yet counted |

| Introduction order | Introductions read | Reached R3 | Harm events |
|---|---|---|---|
| First | Not yet counted | Not yet counted | Not yet counted |
| Second | Not yet counted | Not yet counted | Not yet counted |
| Third | Not yet counted | Not yet counted | Not yet counted |
| Fourth or later | Not yet counted | Not yet counted | Not yet counted |

## What the cohort report will say, and what it will leave out

As of 2026-10-27 no cohort has been followed. When one closes, the report will give counts with their denominators, the within-partner differences with intervals, the survivorship table beside every figure, and the time-to-first-introduction estimate with censored partners shown. It will name no partner and no company, rank no partner, quote no earnings and project nothing forward. Partners receive aggregate results only. The plan, the ladder definitions, the coding rules for harm events and the analysis script are released with the results. What can and cannot be cited from SourceX's records is set out in [the partner results research guide](/guides/partner-results-research).

## Limits of following a single cohort

- **One cohort is one cohort.** A quarter's partners may differ from the next quarter's in network, practice and season.
- **Partner logs go missing.** R1 depends on partners keeping a dated record; "not recorded" will be common and is reported as such.
- **Lag.** An introduction made in month eleven is read four months later; impatience to publish early would bias the result toward early introductions.
- **The measure is not the relationship.** A partner who makes two well-judged introductions a year may serve CEOs better than one who makes ten.
- **Stalls are another question.** Where introductions stop between stages is studied with stage clocks in [the stall study design](/guides/the-most-common-reasons-company-introductions-stall), and the effect of screening on volume in [the quantity-versus-quality analysis](/guides/when-better-screening-lowers-applications-but-improves-outcomes).

## Habits worth starting from a partner's first day

Keep your own dated log from the first day: the company, when its CEO agreed in writing, and what you heard back. Make your first introduction the company whose CEO you know and whose facts you can confirm, not the one with the biggest name. Submit from the dashboard so a dated record exists; [how referrals work](/how-it-works) explains the statuses it shows. More research designs sit in the [guides hub](/guides).

## 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.

New partners who want every introduction to count in a study like this should [register](/register), sign the Channel Partner Agreement before submitting, test each company against [who qualifies](/who-qualifies) and read the [program terms](/terms) before the first introduction.

## Sources

- https://refer.sourcex.si/terms
- https://refer.sourcex.si/how-it-works
- https://refer.sourcex.si/who-qualifies
- https://pmc.ncbi.nlm.nih.gov/articles/PMC11826739/
