Your referral app will tell you how many customers came through a link or a code. It can't tell you how many would have bought anyway. The friend who was already on your email list, the customer's partner who shares the same card, the shopper who found a code in a coupon forum: all counted as the loop's work.

Look again at how the numbers in part two were stated. Houston didn't say the referral program drove 35% of signups; he said it permanently increased signups by 60%, measured on all signups. PayPal's S-1 counted accounts, not bonuses paid. The honest measure of a loop is total inflow.

A referral code is a receipt, not proof.

Hold out the ask, not the product

The Second Order introduced the Holdout Digit: a random number written once to every profile, used to keep a fixed group out of marketing. For a loop, use it to withhold the prompt.

1. Pick the groupA random tenth of new customers, by digit, never sees a referral ask: no post-purchase block, no in-box card, no referral email.
2. Leave the door openThe referral page stays reachable for everyone. You're measuring what the ask adds, not whether sharing is possible.
3. Count referred first orders per 100 customersIn both groups, over the same weeks, completed and unrefunded. The gap is what the prompt produced.
4. Fix the read date firstWrite the verdict date before launch. On a small file, the holdout can confirm a big effect, not measure a modest one.

OVRLND's plan wrote a random digit from 0 to 9 to every profile at ingestion and opened every rebuilt flow with an exit for digit 0: a 10% universal holdout, read over the launch month plus 30 days so both groups rode the same cycle.

Check who the referred customers were

A cheaper test runs on the data you already have. Take last quarter's referred customers and check how many were already known to you before the referral: on your email list, with a prior visit, a past order under another address. That share is a floor on "would have come anyway." It's often a third or more, and it goes straight into the calculator in Pick the Currency.

Always show the denominator

One rule from OVRLND's scorecard applies everywhere: no channel revenue number is shown without total store revenue beside it. In mid-2026, flows earned about 4% of a store doing about $136,000 a month, and the target was 8% to 10%. A referral program reporting "$40K referred revenue" means one thing next to $400K of store sales and another next to $4M.

Read the holdout

Do this

• Exclude a random tenth of new customers from every referral ask, by digit, and set the read date.
• Check last quarter's referred customers against your list and write down how many you already knew.