# SaaS referral pilot lessons: a failure review | SourceX

Canonical: https://refer.sourcex.si/guides/what-a-software-ecosystem-can-learn-from-a-failed-referral-pilot
Last updated: 2026-11-04

Look for the cause of a stalled referral pilot in decisions made before the first email: who was chosen, by what data, who sent the message and what it implied. Review those decisions without blame, keeping every contacted customer in the record, including the ones who said nothing or complained. Then score product trust before trying again, because a customer who now wonders whether you sell their data is a cost the next pilot cannot repay. Relaunch only with opt-in customers and CEOs who asked for the conversation.

This page is about a software vendor's own referral pilot: introducing its customers' companies to SourceX, which licenses business records directly from the companies that created them. A failed AI pilot inside a client company is a different event, one that can point to valuable records, and the [failed AI pilot playbook](/guides/failed-ai-pilot-data-referral-opportunity) covers it.

## Review the pilot without looking for someone to blame

Google's site reliability engineering book describes a blameless postmortem as one that identifies contributing causes without indicting any individual or team for bad or inappropriate behavior, and assumes everyone acted with good intentions on the information they had. A referral pilot deserves the same treatment. The customer success manager who sent an awkward email followed a plan; the plan is what failed, and only a plan can be fixed.

Four rules keep the review honest:

1. Build the timeline from records (sent emails, tickets, replies, submissions), not memory.
2. Keep every contacted customer in the denominator, including those who never answered.
3. Name decisions, not people.
4. End each cause with a change to the plan, never a reminder to try harder.

## The pilot, week by week (Illustrative)

Illustrative. A fictional 400-person field-service software vendor and its customers; the counts are invented to show the method.

| Week | What happened | What it signaled |
|---|---|---|
| 1 | The partnerships team picked 60 customers with the most work orders stored in the product | Targeting by product usage data |
| 2 | Customer success managers emailed account admins: "Turn your data into USD 50,000 or more" | A payout headline sent to people who cannot authorize |
| 3 | Four support tickets asked whether the vendor was selling customer data | The first trust cost |
| 4 | One customer's counsel asked for written confirmation that no data had been shared | Trust cost reaching legal |
| 5 | Six replies, five of them from admins or IT staff | Interest without authority |
| 6 | The team submitted two introductions before either company's CEO knew | Introductions without the decision maker's knowledge |
| 8 | One company had 40 full-time employees at peak; the other's CEO declined a call | No screen before submission, no CEO agreement |
| 10 | Pilot paused with no verified applications | No qualified outcome, several trust costs |

## The failure review

Each row starts from a decision the team made, not an outcome it suffered.

| Decision | What happened | Contributing cause | Change for the next pilot |
|---|---|---|---|
| Selection | 60 customers ranked by work orders stored in the product | Product telemetry became a targeting list | Invite a segment chosen by customer type and marketing consent, never by usage |
| Message | A payout figure led the email | The reward is the vendor's, so the headline sold the vendor's incentive as the customer's gain | Lead with what the customer could license and who decides; disclose the reward plainly |
| Recipient | Account admins received it | The pilot wrote to people who run the product, not the company | Write to an executive who asked to hear more; the CEO authorizes |
| Sender | Customer success managers with renewal targets sent it | A note from a renewal owner reads as pressure | The partnerships lead sends, from a monitored address |
| Submission | Two introductions went in before the CEOs knew | No written confirmation step existed | Two confirmations first: the customer asks, the CEO agrees |
| Screening | A 40-person company was submitted | Nobody checked the published threshold | Check full-time headcount at peak (50 or more), years of records and who owns them before submitting |
| Measurement | The dashboard counted opens and replies | Activity metrics hid the trust costs | Count verified applications and trust signals side by side |

The submission row is the costliest. SourceX's published screening lists introductions made without the decision maker's knowledge among the cases that are usually not a fit, and week 6 produced two of them. Each one also told a CEO, after the fact, that their vendor had put the company's name forward.

## The product-trust assessment

Trust is the asset a software ecosystem spends when it runs a referral pilot, so measure it as carefully as introductions. Score these signals before the first send, at the end of the pilot and again 60 days later. Write the red-line rules down before launch; there is no published benchmark for them, so they are your team's own.

| Signal | How to count it | Where the data sits | Example red-line rule (set by your team) |
|---|---|---|---|
| Data-sharing questions | Tickets, emails or calls asking whether you share or sell customer data | Helpdesk search on agreed phrases | Any rise on the four weeks before the send |
| Legal or procurement requests | Requests for written confirmation about data use | Legal inbox and account notes | Any request that needs investigation to answer |
| Unsubscribes and complaints | From the pilot send, against your usual rate for comparable sends | Email platform | Above the usual rate |
| Renewal and QBR mentions | Account notes that mention the pilot | CRM | Any mention attached to a risk flag |
| Team discomfort | Customer-facing staff reporting awkward conversations | A two-question internal pulse | Any escalation to a manager |
| Decision-maker opt-ins | Customers whose CEO asked for a conversation | Reply inbox | None after a full pilot, read together with the rows above |

Customers have reasons to be alert. In a February 2024 post, FTC technology staff warned that a company adopting more permissive data practices, such as using data for AI training, and telling people only through a quiet, retroactive change to its terms or privacy policy may be acting unfairly or deceptively. A referral message that pairs "your data" with "AI" lands in that climate even when the vendor shares nothing at all; the four tickets in week 3 were that climate arriving. Have the answer ready before the send: the page on [whether a vendor trains AI on customer data](/questions/is-my-saas-vendor-training-ai-on-my-data), the [reply to a customer who asks if you sell their data](/resources/customer-asks-if-you-sell-their-data-to-ai) and the [customer data objection response](/questions/customer-data-objection-response) cover the common questions.

## Keep hosting and licensing apart in pilot two

Hosting customer records gives a vendor access under the subscription. It gives the vendor no say in whether those records are licensed, by whom or to whom. The customer's company decides, exports from its own systems and deals with SourceX under its own agreement. Pilot two writes that into its plan:

- No customer appears on a list because of what the product knows about them.
- No customer's name goes to SourceX without that customer's request. Microsoft's Partner Center applies the same discipline to co-selling: it tells partners to obtain the customer's consent before sharing contact details with Microsoft, because Microsoft can use them to contact the customer directly.
- Nobody at the vendor touches, samples or exports a customer's records for licensing.
- Records that belong to the customer's own clients stay the customer's question; SourceX's published screening treats them as usually not a fit, and the guide to [company data versus customer-owned data](/guides/how-to-distinguish-company-data-from-data-owned-by-its-customers) helps the customer's team sort them.

A different question, whether licensing its own records would surprise a company's customers, belongs to the company itself; SourceX's [customer trust assessment](https://sourcex.si/insights/data-licensing-customer-trust-assessment) covers that side.

## The gate before pilot two

Run the second pilot only when every line holds:

1. A written failure review, shared with the people who ran the first pilot, with each cause tied to a plan change.
2. A trust scorecard back at its pre-pilot level, read by someone outside the partnerships team.
3. An opt-in cohort: customers who asked to hear about licensing their own records, whether through a customer advisory board, a webinar or a reply.
4. Fixed message lines: what SourceX does, the referral reward disclosed, no customer data shared, the product unchanged.
5. A signed SourceX Channel Partner Agreement before any introduction is submitted; a version signed when the document bore its earlier title, Channel Partner Non-Circumvention and Non-Disclosure Agreement, keeps its force as signed.
6. For each company, two written confirmations: the customer asks for the introduction and the CEO wants the conversation.
7. A metadata-level screen before submission: full-time headcount at peak, years of records, the systems that hold them and who owns them.
8. A success measure built on verified applications. Attribution follows a verified application from the company inside the 90-day window, not a click or a reply.

If pilot two leads to a license, the reward follows the published terms. Illustrative: on a company's first qualifying transaction, an Actual Supplier Payout of USD 400,000 would give a USD 70,000 reward (the greater of USD 50,000 plus 5% and 15%), and a USD 40,000 payout would fall below the USD 100,000 Minimum Transaction and earn nothing unless SourceX agrees in writing. Nothing is payable until the conditions in the terms are met, and SourceX then has 60 calendar days to pay, starting from whichever is later: the Acquisition Partner's final acceptance, in writing, of the complete, validated dataset, or completed handoff of that dataset. Progress on each submitted company is visible as described on the page about [seeing the current stage of an introduction](/questions/where-can-a-partner-see-the-current-stage-of-an-introduction), and the [guides hub](/guides) holds the related playbooks.

## 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.

When the review is written, the scorecard is back at its baseline and a customer's CEO has asked for the conversation, [sign up for your partner link](/register), hold that one company up to the criteria on [who qualifies](/who-qualifies) and go through the [program terms](/terms) before anything is submitted.

## Sources

- https://refer.sourcex.si/terms
- https://refer.sourcex.si/who-qualifies
- https://sre.google/sre-book/postmortem-culture/
- https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive
- https://learn.microsoft.com/en-us/partner-center/referrals/manage-co-sell-opportunities
