Customer Research on a Startup Budget: How to Create Knowledge Efficiently in 2026

Every test you run costs time, money or both. A startup has little of either. So spend the least you can on a test that could still change your decision, and skip the tests that couldn’t.

This is the third guide in our customer development series. The second guide showed how to write and rank hypotheses. This one shows how to pick the cheapest test for each, what it costs in 2026, and how a one-page written brief saves you from paying for the wrong one.

Start with the decision, not the method

Before you choose a test, write two lines:

• Which decision does this feed? Build, change the plan or stop.
• What result would change that decision? If no result would, skip the test.

Most wasted research money goes on tests that couldn’t change anything. A survey that asks people whether they like your idea is the classic example. Everybody likes ideas.

Seven tests compared by time, cash and evidence

Test	What it answers	Time	Cash	Strength of evidence
Desk research	Is anyone already complaining about this problem?	1-2 days	$0	Weak: you see opinions, not behavior
Ten problem interviews	Is the problem real, and who has it worst?	1-2 weeks	$0, or small thank-you gifts	Medium: stories of what people did
Landing page with a sign-up	Does the message make anyone act?	1 week	A domain and a page builder, plus ads if you buy traffic	Medium to weak: a click is cheap
Clickable prototype	Do people understand the product and want it?	1-2 weeks	Figma has a free Starter plan; a Professional full seat is listed at $16 a month	Medium
Pre-order or deposit	Will people pay before it exists?	1-2 weeks	A Stripe payment link costs nothing extra on standard pricing; each card payment costs 2.9% plus a fixed fee	Strong: money moved
Concierge test	Will they use it and come back if you do it by hand?	2-4 weeks	Your time	Strong
A paid first version	Will they keep paying for a working product?	2-4 weeks to build	Our MVPs typically cost $5,000-15,000	Strongest

Read the table from top to bottom. Each step costs more and tells you more. Move down only when the cheaper test has passed.

Test material got cheaper to make. In the 2025 DORA report, 90% of respondents said they use AI at work and more than 80% believe it made them more productive. A landing page or the first clickable screens now take hours or days. That makes your time with customers the expensive part, so spend it on the tests that move money or behavior.

Sometimes a small paid product is the cheapest strong test. The founder of GhostMinutes, an AI meeting notes app, wanted a paid product in front of users fast and then to find out which audience and price worked. We shipped the iOS app 2 weeks after the start and a web app a week later. After launch, the founder’s marketing team tested features and pricing on the live product. It reached $1,500 in monthly recurring revenue.

Write it down before you spend

Talking through an idea out loud favors whoever pitches best. A confident story can hide holes in the logic, and people who need time to think get left out. Writing fixes both. Gaps show on paper, and everyone reads the same thing at their own pace.

Amazon writes before it builds. Its Working Backwards process starts with a written press release and an FAQ for a product that doesn’t exist yet. According to Amazon, the press release is a few paragraphs and always less than one page, the FAQ is five pages or less, and at the meeting everyone reads the document to themselves first.

A startup needs less. One page is enough:

1. Who the customer is, in one sentence.
2. The problem in their words, with one quote from an interview.
3. The solution in one paragraph, as a press release would describe it.
4. The three riskiest guesses behind it.
5. The test for the riskiest guess, with a pass and fail line and a date.
6. What the test costs in hours and money.

Have someone else try to kill it

It’s hard to see the holes in your own idea. Before you spend, give the page to someone whose job is to find them: a co-founder, an advisor or a founder friend. Ask them to argue for stopping.

Set a threshold so this doesn’t become a meeting for everything. A rule of thumb from the earlier version of this guide still works: anything that will take more than 16 hours of people’s time in total gets a written review first. Below that, just run the test.

A five-question checklist before any test

A short checklist stops a weak test before it costs anything. Run these five questions before you spend money or more than a day:

1. Which decision does this test feed?
2. What result would make us stop or change course?
3. Is there a cheaper test that gives the same signal?
4. Can we reach ten people from the right group this week?
5. What does it cost in hours and money, and is that below our threshold?

If you can’t answer the first two, don’t run the test.

One shortcut to avoid: asking an AI chatbot to play your customers. It feels cheap, but a 2026 study in the NIM Marketing Intelligence Review found AI-made respondents consistently too positive and less varied than real people. The problem interview checklist covers the study. Use AI to draft your questions and sum up your notes, not to replace the people.

How we would estimate this

Test costs only make sense next to the cost of the build. If you know a first version costs $10,000, two weeks of interviews and a $0 payment link look cheap.

Our project estimate turns your brief into user stories. You confirm what’s in scope, and every hour in the budget traces back to a story you kept. We check the hours the model suggests against what our team actually books on finished projects, then multiply by our published rate of $39-55 an hour.

For market size, the estimate suggests customer groups and gives each one a TAM, SAM and SOM with a note on what the number covers. Treat them as hypotheses to test, not verified demand, and put the riskiest one at the top of your one-page brief. It takes about four minutes to get a free project estimate.

What we would do

1. List your open guesses and the decision each one feeds.
2. For each, pick the cheapest test in the table that could change that decision.
3. Write the one-page brief and have one person try to kill it.
4. Run the checklist, then the test. Keep it to one or two weeks.
5. Move down the table only when a cheaper test passes. Interviews come first; our problem interview checklist has the script. A pre-order or a pilot comes after a solution interview.
6. When you need a build price and a technical plan rather than market evidence, a 1-week discovery sprint costs a fixed $3,000 and counts toward the build. When the cheapest strong test is a paid first version, our MVP development typically takes 2-4 weeks.

FAQ

How can a startup do customer research with no budget?

Start with desk research and ten problem interviews. Both cost only time. Then use a free prototype tool and a payment link for a pre-order, which costs nothing until someone pays.

What is the cheapest way to test a startup idea?

Talking to ten people who should have the problem, using questions about what they did, not what they might do. It costs a week or two of your time and tells you whether the problem is real.

What is a PR/FAQ?

Amazon’s Working Backwards document: a press release of less than one page for a product that doesn’t exist yet, plus an FAQ of five pages or less. Writing it forces you to explain the customer benefit before anyone builds.

Can AI replace customer interviews?

No. AI can draft questions and summarize notes, but a 2026 study found AI-made respondents too positive and less varied than real people. Decisions about what to build need real people.

How much should you spend on validation before building?

Spend what it takes to answer the questions that could stop the build, and no more. Compare each test with the cost of the build. If tests cost a small part of it and could save the whole thing, run them.