The deck has forty slides. The pipeline has three names who already knew you. The usual diagnosis is “we need more top of funnel.” Often the first customers never became evidence. They became a story you tell about the deck.
Ten is a working set, not a law
The Founder Sales System uses the first ten instinct-closed deals as the moment luck stops being a method. The live back copy is careful: conviction and warm introductions may get you to ten customers. They will not tell you which opportunity deserves time, why a discovery call worked, or what next quarter is likely to produce.
That “ten” is a working set. It is not a sample-size law, and it is not a surveyed success rate. If you have four customers you can still describe, audit four. If you have fourteen, take the last ten you can still describe without opening the CRM for the name.
The book is drafting English first. It is not for sale. Its public catalogue and its manuscript currently disagree on the acronym that organizes the motion. This page does not pick that fight. The shared claim is enough: a deal you cannot repeat is not a system.
Discovery before the demo is part of that claim. So is a forecast that can name its evidence, and a weekly review that turns a lost deal into a change you can point at. You do not need the acronym to write those three things down.
One sourced claim per customer
SENSE is the live protocol in AI Agents for Startup Strategy: Scan, Evaluate, Navigate, Simulate, Execute. The 90-day install of a standing layer is a different page. This page is smaller. Each early customer is one claim that should change a watchlist, an assumption register, or a Monday decision.
Scan. What did they actually buy, and what did they ignore?
Evaluate. What evidence counts: signed, used, expanded, left. A compliment in a call is not evidence.
Navigate. Who owns the GTM bet, and on what cadence does a killed claim die?
Simulate. If the next ten look like these ten, what breaks: price, segment, channel, or the promise?
Execute. One change with an owner. A slide that “we should revisit positioning” is not a change.
If you cannot write the claim in one line, you do not have a customer in the system. You have a name in a deck.
An agent can help gather the raw material. It cannot own the claim. The unit of work is still one sentence with a confidence grade, not another unread summary.
A test bed, not a vendor number
The live book is explicit about this surface. Go-to-market can be a strong first test bed because the work repeats, feedback arrives, and you already measure something. That is why it is a useful place to instrument a loop. It is not a promise that customer-acquisition cost will fall.
Writing-task studies do not validate go-to-market outcomes. Vendor case studies without controls are hypotheses. Measure against your own baseline, never against a number you were given in a demo. The chapter that says this also refuses the “highest-return first deployment” headline. The test-bed argument is enough.
If you put a model on outreach before you can state your own cost per acquired customer and your own reply rate, you did not install a loop. You installed a faster way to chase a proxy.
Keep the prior visible
The Predictive Founder treats judgment as something you can inspect. PREDICT is Prior, Reality-test, Evidence, Decide, Iterate, Calibrate, Track. The book is in active development. It is not for sale. The bible is explicit: the seven-letter bundle is an authorial practice. It has not been validated end to end as a training effect. Use it anyway as a hygiene rule.
Write what you believed before the tenth customer paid. Segment, price, channel, the sentence that was supposed to close. If you only write the story after the invoice, you are calibrating on a memory. The prior is gone.
Do not confuse ten customers with ten forecasts. That book uses “first ten eligible forecasts” as a different working set, under a predeclared inclusion rule. This page is the customer set.
A GTM deck is a prior. The customers are the reality-test. If you cannot match a name to the deck without forcing it, the deck is not describing that customer. Count the misses. That count is the prior’s score. Kill the claim the misses falsify. Do not hire a slide to defend it.
A fifteen-minute audit
Pick the customers you would least like a journalist to sit with.
- List the last ten you can still name, or fewer if that is all you have.
- For each, write why they bought, in their words. One line.
- Circle the ones that match the deck’s ICP. If you need the deck to force the match, it is not a match.
- Write the claim you would kill if the next three look like the mismatches.
- Write whose name owns that kill, and the date it dies if the evidence does not arrive.
If step 2 is blank, you do not have a go-to-market system. You have a deck that has not met a customer. Write the claims. Then the next ten have somewhere to land.