# Customers and hypothesis

URL: /business/record/customers-and-hypothesis · Id: `page:business/record/customers-and-hypothesis`

> The ICP a Startup serves, the Problem it solves, and the founding Hypothesis whose five slots Evidence grades one week at a time.

A [Startup](/business/nouns/startup) serves one [ICP](/business/nouns/icp) and the [Problem](/business/nouns/problem) that ICP already pays to solve. Its founding [Hypothesis](/business/nouns/hypothesis) says why that ICP switches, in five slots that outside [Evidence](/business/nouns/evidence) grades.

## The ICP and its Personas
An ICP names who the Startup is built to win. For platecost.app, it is an independent restaurant with one to three locations whose food cost runs above 30% of sales.

Inside one ICP sit several [Personas](/business/nouns/persona), each with its own job. The owner accepts the [Offer](/business/nouns/offer), the chef updates recipes, and the bookkeeper's [Agent](/business/nouns/agent) pulls the weekly cost export. The one who pays, the one who uses, and the one who benefits are often different, and one of them is software.

Before a party accepts an Offer, at any [Price](/business/nouns/price) including zero, it is a [Prospect](/business/nouns/prospect). After, it is a [Customer](/business/nouns/customer). A Startup keeps one ICP until it reaches [Repeatable](/business/lifecycle/states/repeatable); a second ICP gets a sibling Startup pursuing the same [Idea](/business/nouns/idea).

## The Problem in their words
A Problem records what the ICP struggles with, how they solve it today, and what that costs. The words are theirs, not the Startup's.

> Every time the produce prices move, the food-cost sheet gets rebuilt by hand. That is my whole weekend.

Those words name the work, its frequency, and its cost. Four hours a week of an owner's time, or the $400 a month a bookkeeper charges for the same sheet, anchors what platecost.app can charge.

The Problem also records three facts about the work:

- The moment that starts it: a new supplier invoice.
- The progress wanted: from guessing margins to knowing them per plate.
- What the ICP uses today: a spreadsheet, a bookkeeper, or nothing.

A Problem with no current cost has no market.

## The founding Hypothesis
The founding Hypothesis is the root of the definition: one sentence in a fixed shape. platecost.app writes this one:

```text
If we help independent restaurants with one to three locations
solve food-cost drift from changing supplier prices
with plate costs recomputed from every invoice they photograph,
they will choose it over the spreadsheet and the bookkeeper
because margins update the same day, per plate, for $79 a month.
```

Each of its five slots is a child Hypothesis with its own [Metric](/business/nouns/metric) and threshold. The sentence stays a draft until every slot names both.

| Slot | Claim | Settled when |
| --- | --- | --- |
| Customer | These restaurants exist and answer | 15 of 40 owners contacted take an [Interview](/business/nouns/interview) |
| Problem | They rebuild cost sheets by hand, weekly | 10 of 15 Interviews describe the rebuild unprompted |
| Approach | Photographed invoices give correct plate costs | 95% of invoice lines parse without correction |
| Switch | They pay to stop | 8 refundable $79 deposits within 14 days |
| Differentiation | Same-day margins beat a lower price | "Margins every morning" outsells "cheaper than a bookkeeper" on deposits |

Beside the sentence, the [Founder](/agents/founding/founder) sets one number: the minimum success criteria, a [Goal](/business/nouns/goal) of outside revenue by a date. platecost.app sets $4,000 a month within nine months. [Strategy, Goals, and economics](/business/record/strategy-goals-and-economics) turns that number into dated targets.

## Grading Evidence
Evidence is a recorded quote, reading, or artifact bound to the slot it supports or refutes. Each slot shows the strongest grade among its outside Evidence.

| Grade | platecost.app example | Weight |
| --- | --- | --- |
| Behavior | A paid month, then a second one | Highest: observed, not promised |
| Commitment | A refundable deposit or a booked onboarding call | Real: it costs the giver something |
| Opinion | "I would pay for that." | Zero: free to give |
| None | Nothing recorded | Zero |

Opinions stay on the record and weigh nothing. Only outside parties grade a slot, so a sibling Startup, the [Studio](/business/nouns/studio), or a friend of the Founder never moves one.

[Lifecycle](/business/nouns/lifecycle) gates read grades, never filled-in fields. platecost.app cannot leave [Idea](/business/lifecycle/states/idea) by typing into the switch slot; it leaves when deposits land.

## The weekly test
At the start of each week the Startup ranks its five slots by (1 − grade) × impact. Impact is how far a wrong slot would move the minimum success criteria.

The weakest slot gets the week's one [Experiment](/business/nouns/experiment), and that Experiment's Metric is the one metric that matters. At the end of the week, office hours reads the Evidence and writes one [Decision](/business/nouns/decision): keep the slot, revise it, or test again.

The first test is usually reserve-first. platecost.app publishes a reserve page the same day, taking refundable deposits on the $79 Offer, and sends 40 owners there. Talk is cheaper but settles less: once ten more Interviews would leave the Decision unchanged, the Startup stops interviewing.

## One Product, several stories
The same [Product](/business/nouns/product) needs a different story for each Persona. The owner hears the margin and the price; the chef hears "no more weekend spreadsheets" and sees the recipe screen.

An Agent buying for a restaurant group needs none of that prose. It needs the schema, the terms, the turnaround, and the Price, all machine-readable.

Every one of those stories compiles from the founding Hypothesis and the Interview quotes, and each sentence records the slot it came from. Revise the differentiation slot and every headline built on it recompiles, with the [Change](/business/nouns/change) previewed before it lands.

## Next
[Products, Offers and revenue](/products/offers-and-revenue): what the Startup sells, and how a sale becomes revenue.