How a pay-on-uplift retention pilot works

In my pilot you pay $1,500 to set it up and 20% of the measured uplift over 90 days. Uplift is net: revenue times margin, minus what the offers cost. It is measured against a holdout group that gets nothing from me. If there is no uplift, there is no percentage. Here is how each part works, and what could go wrong.

What is the pilot?

A 90-day test of the fixes that the Revenue Leak Report points to: better failed-payment recovery, a reason-first cancel flow and a trial-end sequence. I connect Stripe, set up the campaigns, and send from your own brand and domain. A weekly report gives you the result in dollars.

It comes after the report, not before. The report tells us where the money is and how many events each group would get at your volume. The pilot tests whether the fixes recover it.

How is it priced?

Two parts. The setup is $1,500, for connecting Stripe, building the campaigns and getting them out under your name. The variable part is 20% of measured uplift. I think of the setup as covering my work and the percentage as the reason for me to care about the result.

There is no percentage without uplift. If the treated group does not beat the holdout, you owe only the setup.

What does "net uplift" mean?

It is the difference between the two groups in net revenue per customer: revenue times gross margin, minus the cost of the offers I gave. I define it that way so that I cannot make the number look good by giving discounts. If a discount wins back a customer and costs more than it earns, it counts against me.

The test used for rates, such as recovery or save rate, is a two-proportion test at 95%. A group with fewer than 30 events is reported as an early signal, not a result. I would rather report "not yet clear" than a number that will not hold.

Who is in the holdout?

20% of your customers, chosen by a stable hash of the customer id, so nobody switches between groups. They are not contacted by me at all. The recovered dollars are the gap between the groups. I wrote about why a holdout beats a before-and-after chart in what a holdout group is.

Holding 20% back has a cost: those customers do not get the fixes during the test. If the fixes work, you gave up some recovery to be able to prove it. I think that trade is fair, because the payment depends on the proof.

What safeguards are there?

• Every change can be rolled back. Changes are recorded as requests, and you can undo any of them.
• A unit-economics check on every offer. An offer that costs more than it can win back does not go out. See are discount save offers worth it.
• A weekly report in dollars, treated group against holdout.

You can also run it with your own AI agent over MCP. In that arrangement the agent proposes a change, you approve it, it applies it, and you can undo it.

Which channels does it use?

Email is live. WhatsApp, SMS, Viber and Telegram are available on request and are not part of the standard pilot. If you need a channel other than email, ask before you start, because it changes what I can set up and what a message to a customer is allowed to say.

How many events will each group get?

The report tells you, at your volume. In my synthetic sample, for a fictional company, the pilot plan expected about 87 treated and 22 holdout invoices for failed-payment recovery, 46 against 11 cancellations for the save rate, and 84 against 21 trials for conversion. Those are not real. They show why the holdout arm sets the pace: a group of 11 is far below 30 events, so that result would be an early signal at most.

If your account is smaller than that, I tell you before you pay. A pilot below roughly 2,000 paying customers rarely produces a clear result.

What could go wrong?

• Too few events. The test is inconclusive. You pay the setup and no percentage.
• Contamination. If holdout customers get similar messages from another tool, the gap shrinks. Tell me what you already send.
• Invisible saves. If your team saves customers by hand, it can blur the comparison. Both groups get that, so it is usually fine, but I want to know.
• A change in your price or product mid-test. It hits both groups, but it can swamp a small effect.

Why not charge a flat fee only?

Because a flat fee asks you to trust that the work pays, and the setup fee covers my time whatever happens. The percentage ties the rest of what I earn to a measured result that you can check against the holdout. It also means I should tell you early when a pilot is unlikely to show a clear result, which is why I answer size questions before you pay.

How do I start?

With the report. You send a restricted read-only Stripe key, and within 48 hours of my having it you get the leaks in dollars, the top three fixes and the pilot plan for your volume. The report is $490. You then decide whether to run the pilot. Read is the $490 report worth it for how to weigh that, or look at the synthetic sample.