Build-Measure-Learn: Test Your Business Idea with an MVP Before Spending Real Money, with a Prompt Guideline for AI
However good an idea is, it is still a guess until customers actually pay. Build-Measure-Learn is a way to test ideas bit by bit at low cost, then use data to decide whether to keep going or change direction. This article explains what data to prepare for AI and gives you prompts to copy straight into ChatGPT, Gemini or Claude.
What is Build-Measure-Learn
Build-Measure-Learn is the core loop of the Lean Startup approach. Eric Ries began writing about it on his blog in 2008 and explained it fully in his book The Lean Startup (2011), building on Steve Blank's Customer Development. The idea is to turn an idea into something customers can try (Build), measure what customers actually do (Measure), then learn so you can decide whether to keep going or change direction (Learn), repeating as fast as possible so you do not waste time and money on things customers do not want.
What the Build-Measure-Learn loop looks like
The loop has 3 actions (coloured boxes) and 3 outputs (circles). Ideas are built into a product, the product is measured to produce data, data leads to learning, and learning feeds new ideas for the next round. Tap any point to jump to its explanation and prompt.
Prompt Guideline: Data to Prepare for AI to Design Your Test
If you only tell AI "make me an MVP", you get broad ideas that do not answer the questions that matter to your business. Prepare these 8 kinds of data first. The numbers in the last column show which step each item feeds (0 Ideas · 1 Build · 2 Measure · 3 Learn).
| Data to prepare | Examples of what to include | Where to find it | Used for stage |
|---|---|---|---|
| The idea you want to test | The product, service or change you want to try, and why you think it will work | Where to find itYourself, your existing Lean Canvas or BMC | 0 |
| Your riskiest assumption | 1 to 3 sentences in the form "We believe [customer] will [action] because [reason]" | Where to find itThe Problem and Revenue boxes in your Lean Canvas | 02 |
| Customers to test with | Who, how many, how to reach them | Where to find itExisting customer list, LINE OA chats, people you know | 1 |
| Resources available | Hours per week, budget you can afford to lose, what you can do yourself right away | Where to find itYou and your team | 1 |
| Numbers before the test | Orders, new customers and repeat customers from the previous period, for comparison | Where to find itSales log, LINE OA Manager, accounting software | 2 |
| Pass / fail line | The number you set before starting for how much is enough to keep going | Where to find itSet it yourself from your costs and goals | 23 |
| Actual results | Numbers after the test, with dates, plus what customers said or typed | Where to find itSales log, chats, short surveys | 3 |
| Customers' own words | Why they bought, did not buy or bought again, in their own words | Where to find itChats, phone calls, interviews after delivery | 3 |
6 rules before you paste data into AI
- 01Always set the line before you startWrite down the pass / fail number before testing. If you set it afterwards, you tend to find reasons that favour your own idea.
- 02Test one assumption at a timeChange several things at once and you will not know what caused the result. Ask AI to split tests into separate items.
- 03Give numbers with their time periodTell AI the dates your numbers cover so it can compare before and after the test correctly.
- 04Do not let AI guess resultsAI can help design tests and read results, but do not let it predict sales or purchase rates in place of real data.
- 05Separate vanity numbers from decision numbersLikes and views do not show customers will pay. Have AI focus on orders, deposits or repeat purchases.
- 06Remove personal dataRemove customer names, phone numbers and addresses before pasting chats or survey answers into AI.
No data yet? Let AI interview you first
AI asks one question at a time, from the idea and your riskiest assumption to how you will decide, then sums it up as a one-page experiment card.
Act as a lean startup coach using Eric Ries's Build-Measure-Learn loop. Interview me ONE question at a time, in Thai, to collect what you need to design my first experiment: the idea, my riskiest assumption, who I can test with, my time and budget, baseline numbers, and how I will decide. If I do not know an answer, tell me how to find it cheaply and move on. When done, summarise everything as a one-page experiment card and list what is still missing. Do not give recommendations yet.
Have the data? Design a full test round
Get a hypothesis, an MVP, the numbers to measure, a pass line, and what to do next whether it passes or fails.
Act as a lean startup coach. Design one full Build-Measure-Learn experiment using ONLY the data below. Business today: [PASTE] Idea to test: [PASTE] Riskiest assumption: [PASTE] Who I can test with: [PASTE] Time and budget: [PASTE] Baseline numbers (with dates): [PASTE] Give me an experiment card with: 1) hypothesis in the form "We believe [customer] will [action] because [reason]", 2) the MVP to build and what NOT to build, 3) 2 to 3 actionable metrics and how to collect them, 4) a pass / fail line (explain the logic, leave the number for me), 5) test length, 6) what I will do if it passes and if it fails. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
Build-Measure-Learn step by step
The example is a homemade Thai dessert shop selling online through LINE OA, the same fictional shop used in other articles in this knowledge hub. This time it wants to test a new product, meeting snack sets for companies, before investing in equipment and new packaging.
0Ideas and hypothesesIdeas → Hypothesis
Every round starts with an idea, but an idea cannot be tested until it becomes a hypothesis: a sentence stating what you believe customers will do, which you can measure as true or false. Test first the one that would stop the business if it were wrong, not the one that is easiest to test.
- If this idea fails, which assumption is it most likely to fail on?
- Is the hypothesis measurable: who does what, and by when?
- What do you need to know before you dare to invest more?
Help me turn my idea into testable hypotheses (the IDEAS step of Build-Measure-Learn). Business and what I sell today: [SHORT DESCRIPTION] New idea I want to test: [IDEA] Target customer: [WHO] Why I think it will work: [PASTE] Write 3 to 5 hypotheses in the form "We believe [customer] will [action] because [reason]". Rank them by risk (which one kills the idea if wrong) and recommend the one to test first. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
1Build an MVPBuild
Build the smallest thing that can answer your hypothesis, called an MVP (Minimum Viable Product). It does not have to be a finished product. It can be a pre-order post, a sample menu or a service done by hand at first. Ask yourself: if I cut this part, do I still get the answer? If yes, cut it.
- Which MVP answers this hypothesis fastest and cheapest?
- Which parts can be done by hand first, without investing in a system?
- Which customers will try it, and how many?
Help me design an MVP (the BUILD step of Build-Measure-Learn). Business and what I sell today: [SHORT DESCRIPTION] New idea I want to test: [IDEA] Target customer: [WHO] Hypothesis to test: [PASTE] Time and budget I can spend: [PASTE] Who I can test with: [PASTE] Suggest 3 MVP options from simplest to more complete (for example a pre-order post, a concierge service done by hand, a small limited batch). For each: what to make, how long it takes, what it can and cannot prove. Recommend one and list what I should NOT build yet. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
2MeasureMeasure
Measure what customers actually do, not whether they say they like it. Most people say they like it to be polite, but orders, deposits and repeat orders are evidence you can trust. Choose numbers you can decide with, and set the pass line before the test begins.
- Which number shows directly whether the hypothesis is true?
- What is the pass line, and what is it based on?
- Where will you collect the numbers, daily or weekly?
Help me plan what to measure (the MEASURE step of Build-Measure-Learn). Business and what I sell today: [SHORT DESCRIPTION] New idea I want to test: [IDEA] Target customer: [WHO] Hypothesis: [PASTE] MVP I will run: [PASTE] Data I can collect (LINE OA, sales records, ads): [PASTE] Baseline numbers before the test (with dates): [PASTE] Suggest 2 to 3 actionable metrics that show what customers actually do, how to collect each, and a simple tracking table. Help me reason about a pass / fail line from my own costs and goals, but leave the final number for me to decide. Point out any vanity metric in my plan. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
3Learn and decideLearn
Compare your numbers with the line you set, then choose one of two paths: persevere by improving, or pivot by changing a core assumption, such as the customer group, the problem you solve or how you make money, and start a new round. A result that does not pass is not a failure but learning you gained at low cost.
- Did the results pass the line you set before starting?
- What did customers say that you did not expect?
- If you pivot, what changes: the customer, the problem or the revenue model?
Help me learn from my test (the LEARN step of Build-Measure-Learn). Business and what I sell today: [SHORT DESCRIPTION] New idea I want to test: [IDEA] Target customer: [WHO] Hypothesis: [PASTE] Pass / fail line I set BEFORE the test: [PASTE] Actual results with dates: [PASTE] What customers said (anonymised): [PASTE] Compare results with the pass line, summarise what we learned, and recommend persevere or pivot with reasons. If pivot, suggest 2 to 3 pivot options (customer, problem, channel, revenue model) and the next hypothesis to test. Be honest if the data is too thin to decide. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
Which MVP types suit small businesses
An MVP is the version that lets you learn the most about customers with the least effort. Frank Robinson first used the term in 2001, and Steve Blank and Eric Ries made it popular. Small businesses do not need to build an app or website. These formats work right away.
Say what you will sell and at what price, then take bookings or deposits before you produce anything. It tests whether people really want to buy.
Dessert shop: Post a menu image of the meeting snack sets on LINE OA with a booking form for next week
Deliver the whole service personally, one on one. Customers know a person is doing it, and you learn their needs in detail before building a system.
Dessert shop: The owner talks to admin staff directly, plans menus to fit their budget and headcount, and delivers in person
Customers see what looks like a ready system, but people do the work behind the scenes. The example Eric Ries tells is the founder of Zappos, who posted photos of shoes from local stores online, then bought and shipped them once someone ordered.
Dessert shop: An automatic order form in LINE OA, while receipts and delivery rounds are still handled by hand
Let customers try a small real version, then measure how many place a real order afterwards.
Dessert shop: Send small tasting boxes to admin teams at 10 companies, then track how many order a full set
Make a small quantity and sell it for real over a short period. You get real sales and cost data without much risk.
Dessert shop: Take a limited number of meeting snack set orders per week for 4 weeks
Let AI help you choose an MVP
Compare the 5 types against your hypothesis, then plan your first week.
Help me choose the right type of MVP. Business and what I sell today: [SHORT DESCRIPTION] New idea I want to test: [IDEA] Target customer: [WHO] Hypothesis: [PASTE] Time and budget: [PASTE] Compare these MVP types for my case: pre-order or landing page, concierge (done by hand), Wizard of Oz (looks automated, done manually), sample or demo, limited batch. For each: what it can prove, effort, risk. Recommend one and give me a step-by-step plan for the first week. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
Vanity numbers versus numbers you can decide with
In the Measure step, Eric Ries splits numbers into two kinds. Vanity Metrics make you feel things are going well but do not show whether the business really works. Actionable Metrics tell you what to do next.
- Likes and shares
- Video views
- Follower growth
- People who answer "interesting" in a survey
- People who order or pay a deposit
- Repeat order rate within a set period
- Cost per order from ads
- Customers who came through referrals
The same number can be either kind depending on the business. Views are a vanity number for a dessert shop, but a key number for media that earns from advertising.
Let AI sort the numbers you track
Paste the numbers you check every day and let AI show which ones you can really decide with.
Here are the numbers I track today: [PASTE WITH DATES] My current experiment: [PASTE] Sort them into vanity metrics and actionable metrics for this experiment, explain why, and suggest what I should track instead. Show a simple weekly tracking table I can keep in a spreadsheet. Answer in Thai, in short bullet points a beginner can act on. For every point, say which of my data it comes from, mark anything that is a guess, and never invent numbers, prices or results.
Pivot or persevere: how to decide
A pivot is a structured change of direction to test a new core hypothesis. It is not starting over from zero, because you keep what you learned. A commonly cited example is Groupon, which began as an online activism platform called The Point. Almost no one used it, so the founders tried a discount coupon for a pizzeria in their own building and built on that.
For a small business, signs to consider a pivot are numbers that do not move after several rounds even after improving the product, or customers repeatedly raising a different, bigger problem. A sign to persevere is a group of customers who clearly buy again, even if they are few. The prompt in the "Learn and decide" step above can help you think it through.
Use with Lean Canvas and BMC
Find your riskiest assumption in the Lean Canvas then test it with Build-Measure-Learn. Once you have results, update your Lean Canvas to a new version. If your business already has customers and wants to test a new product, the relevant boxes in the Business Model Canvas also work as a starting point.
Mistakes beginners often make
- Perfecting the product before customers see it, spending months only to find no one wants it
- Asking customers if they like it instead of checking whether they actually order or pay
- Not setting a pass line first, then reading the results in favour of your own idea
- Changing several things in one round, so you cannot tell what caused the result
- Using likes or views to decide an idea has passed
- Testing once and stopping, when the heart of Build-Measure-Learn is looping many times, fast
Checklist before you start testing
- The hypothesis is a measurable sentence, and the riskiest one goes first
- The MVP is the smallest version that still answers the hypothesis, with everything unnecessary cut
- Numbers measure what customers actually do, not likes
- Set the pass / fail line before starting, and fix the test length
- Customer personal data is removed before pasting into AI, and there are no numbers invented by AI
- Set a review date, decide to persevere or pivot, then start the next round
References: The Lean Startup: Principles, Wikipedia: Lean startup (Build-Measure-Learn, MVP, Pivot), Wikipedia: Minimum viable product, Wikipedia: The Lean Startup (book)Eric Ries, The Lean Startup, Crown Business (2011) · The business and numbers in the examples are illustrative. AI analysis is a suggestion to consider, so check it against real data before making investment decisions.
Share it with a friend who is starting a business. Share to Facebook Share on LINE


