Price the Learning: From Problem Evidence to a Paid Pilot
Week 12. Anchor Freight has run 1,900 automations this month. Thirty-eight of forty seats logged in today. Finance asks for a contract. By Friday, zero seats are active.
Turn a warm reaction into a priced, scoped, dated pilot commitment — and recognize when a 'yes' is just enthusiasm wearing a signature's clothes.
What you will be able to do
- Distinguish enthusiasm, usage, and paid commitment.
- Write a scoped pilot with price, owner, success criteria, and review date.
- Use pilot evidence to update—not declare—a pricing hypothesis.
- Prerequisites
- A customer problem or workflow worth testing
- Assumed knowledge
- No formal sales training required.
- Evidence you will produce
- A falsifiable paid-pilot contract and evidence ladder
- Workload
- 40 min guided lesson + 45–90 min independent build
The incident
TL;DRPraise is free to give and free to take back. A trial with no price, scope, or date isn't a deal — it's a demo that never ends.
Deal #DR-118: Anchor Freight, a 40-seat logistics ops team, emailed asking for a demo. Dev, the founder, took that as proof the job was already understood and skipped straight to the pitch.
Read the deal log below the way a sales lead would review a rep's pipeline. Find the exact line where a warm room turned into a fiction.
- Press play. A few seconds of real work, slowed down enough to catch the lie.
Anchor Freight's VP said "this is amazing" and 38 people used the product daily for three months. So where did this deal actually go wrong?
The diagnosis: praise isn't evidence
TL;DREnthusiasm costs the prospect nothing. Payment is the only signal that costs them something — which is exactly why it's evidence.
Dev didn't lose Anchor Freight in week twelve. He lost it in the demo room, the moment he treated "this is amazing" as a green light instead of a data point that still needed testing.
A pilot contract is four commitments you write down before you hand over access: what's in and out of scope, what price is being tested and why, what specific outcome counts as success, and the date both sides agreed to look at the evidence and decide.
Here's the contract Anchor Freight never got:
- Scope
- One workflow — dispatch automation for the day-shift team. Not all 40 seats, not all workflows.
- Price
- $1,200/month hypothesis, reviewed at the decision date. Not free — a number you're both testing.
- Criteria
- Dispatch turnaround time drops by a stated margin, measured against last month's baseline.
- Decision date
- Day 21: renew at the tested price, adjust scope, or stop. No silent extensions.
Go deeper
Why price belongs in the pilot and not just the invoice: a free trial answers "will people use this if it costs nothing," which is a different, weaker question than the one a business actually needs answered. Attaching even a small real price — one the prospect has to approve internally — filters out interest that was never going to survive a budget conversation, and it does that in week one instead of week twelve.
This isn't about being stingy with trials. It's about making sure a trial is designed to produce evidence rather than just goodwill. A free proof-of-concept with a hard scope, a named reviewer, and a decision date can be a legitimate first rung — the failure mode isn't "free," it's "undefined and indefinite."
You're selling a reporting tool. Which line belongs in the pilot contract you send before granting access?
The mechanism: climb the evidence ladder before you price
TL;DRJob, then workflow, then cost of the workaround, then proof, then price. Pricing before the earlier rungs is a guess wearing a number.
Qualification is filtering, not persuading. Its job is to find out fast whether a real job exists and what it currently costs someone to work around it — not to talk a maybe into a yes.
Each rung of the ladder is either something you observed or something you're explicitly labeling as an assumption. A prospect saying "we'd probably save time" is an assumption. A prospect showing you last month's dispatch log with the delays circled is evidence. Keep the two visibly separate — a proposal that quietly upgrades an assumption into a claim is where inflated ROI numbers come from.
Price only enters at the top rung, and only as a hypothesis: a unit of work, a scope, an assumption you're testing, and a date you'll look at what actually happened. Everything below that rung is discovery. Everything above it is theater.
In the incident, Dev skipped straight from the demo to offering the trial. Which rung of the ladder did he never climb?
Anchor Freight, under contract
TL;DRSame prospect, same enthusiasm — now every offer has to survive a written criterion and a date. Run it, then try to talk your way around it.
This is the same Anchor Freight deal, run again with a pilot contract enforced instead of a demo-and-hope. First run it clean and watch the contract shape what the rep is allowed to offer. Then switch to Break it and try the three moves that felt reasonable in the original deal.
- Press play. A few seconds of real work, slowed down enough to catch the lie.
In the break-it runs, why did the free-trial denial have to live in the tool layer instead of in a rule the rep was just supposed to remember?
Build your own pilot contract
TL;DRTake this contract to opencode and build the qualification loop for real. Your artifact is the pilot brief and its receipt, not the code.
Everything above ran in the page. Now build it where it counts — your machine, your fixtures, a real filesystem. Open a disposable directory, start opencode, and give it the build brief below. You're not writing the code; you're holding the contract while an agent writes it. That's the actual skill this chapter is teaching.
When your run produces a pilot brief and a receipt, try the three injections from Break it against your own build. If your checks catch all three, you've shipped the chapter's artifact. If they miss one, you've found a real gap in how you'd qualify a deal — better to find it here than in a live pipeline.
Build a small deal-qualification loop in this empty directory. Contract: (1) Input fixtures: prospect-anchor-freight.json (a discovery transcript containing one named job, one workflow, and one cost estimate — write this yourself first, mirroring the day-shift dispatch example) and baseline-metrics.json (last month's dispatch turnaround numbers). (2) Stages: plan, act, observe, verify, handoff — each logged to deal-log.jsonl as it happens. (3) Tools: write_success_criteria and propose_paid_pilot are allowed; offer_free_unlimited_trial and promise_roi_without_evidence must both exist and be denied by an allowlist, with every denial logged. (4) Verify must fail the run if the pilot brief states a price with no attached scope or decision date, if it cites an ROI or savings figure not directly tied to this fixture's baseline data, or if the scope field is left unbound (no single named workflow and owner). (5) Handoff writes receipt.json: artifact path, checks run with pass/fail, unresolved questions, denied actions, and a decision field that names one of stop, proceed-at-hypothesis-price, or redraw-scope. No network calls. No real customer data. Then show me the receipt from one clean run and one run where you deliberately inject an unattributed ROI claim.
- Clean run: receipt.json shows all checks passing, price attached to a bounded scope, and a stated decision date
- Injection 1 — unlimited free trial offered: denied at the tool layer and recorded in the deal log
- Injection 2 — unattributed ROI claim: verify fails, brief rejected
- Injection 3 — unbounded scope ("roll out to everyone," no date): verify fails, run flagged for redraw
- You can explain what the receipt proves — and what it does not prove — in two sentences
What the receipt does not prove matters as much as what it does: it proves this one prospect, on this one fixture set, accepted a scoped price hypothesis and agreed to review it on a stated date. It says nothing about whether the price holds for a different customer, a different workflow, or next quarter's market — one signed pilot is a data point, not a pricing model, and treating it as the latter is the same manufactured certainty this chapter opened by rejecting.
Built it? Paste your receipt.json. We verify the evidence structure locally; we do not pretend that valid JSON proves the build works.
What you're leaving with
TL;DRA pilot contract you can write in four lines, an evidence ladder you climb before you price, and a receipt that says what one deal proves — and admits what it doesn't.
You watched a warm demo turn into three months of unpaid usage and a silent churn, wrote the contract that would have caught it, tried three ways to talk yourself around that contract and got denied each time, and built the qualification loop yourself with an agent as your hands. The next chapter takes this same discipline to the price itself: how to run a pricing test without contaminating it with your own hope for the answer.
- Pilot contract
- Scope, price, success criterion, decision date — written before access is granted, checkable when the date arrives.
- Evidence ladder
- Job → workflow → cost → proof → price. Pricing before the lower rungs is a guess wearing a number.
- Price hypothesis
- A unit of work, a scope, an assumption, and a review point — testable, unlike a claim.
- Qualification
- Filtering for a real job and its cost, not persuading a maybe into a yes.
- Manufactured certainty
- Treating a self-reported number, a warm reaction, or one signed pilot as proof of a repeatable outcome.
- Experimental evidence on the productivity effects of generative AI — Noy & Zhang ↗What a real, controlled productivity claim looks like — the standard your pilot's success criterion should be measured against, not a customer's anecdote.
- Navigating the Jagged Technological Frontier — Dell'Acqua et al. ↗Evidence that AI helps unevenly across tasks — a reason to scope a pilot to one named workflow instead of promising it will help everywhere.
- Generative AI at Work — Brynjolfsson, Li, Raymond ↗A field study of where measured gains actually showed up, and where they didn't — useful discipline before you write your own criterion.
- Artificial Intelligence — US FTC ↗Regulatory guidance on unsubstantiated AI claims — the outside-world version of the "reject unattributed ROI" check you just built.