How to Test a Business Idea Before You Spend Too Much

Validate and test a business idea by isolating its riskiest assumption, collecting evidence from real customer behaviour, and deciding in advance what result means continue, change or stop. A useful test reduces uncertainty; it does not merely create encouraging activity.

Turn the idea into assumptions

Write one sentence for each uncertainty: customer, problem, urgency, current alternative, access, willingness to pay and ability to deliver. Then make each falsifiable. The numbers below are illustrative test thresholds, not observed market evidence.

Run evidence-seeking customer interviews

I would start with people who recently faced the problem you want to solve, including people who chose another solution. Friends can help you rehearse, but encouragement from friends is weak evidence of demand. These are customer discovery conversations, not job interviews or a sales script.

Open the conversation clearly

Explain who you are, what you are researching, how long you expect the conversation to take and how you will use the notes. Make participation optional and let the person skip a question or stop. Ask separately before recording; written notes are often enough for a small first test. Do not put confidential client work, account details or sensitive personal information into your research log.

A neutral customer discovery script

1. Context: “Tell me about the last time you needed to [complete the task].”
2. Sequence: “What happened first, and what did you do next?”
3. Difficulty: “Which part, if any, was difficult?” Let “none” be a valid answer.
4. Current alternative: “What did you use or who did you ask for help?”
5. Actual cost: “What time or money did that take?” Accept an estimate and label it as one.
6. Priority: “What happened when it was delayed or left undone?”
7. Buying process: “How did you decide whether to pay for help, and who was involved?”
8. Missing context: “What have I missed about how you handle this?”

Follow a concrete answer with “Can you give me an example?” Avoid questions that supply the desired answer, such as “Would an affordable automated tool save you hours?” A hypothetical promise to buy is weaker evidence than a recent purchase or a clearly agreed next step. If you show an offer afterwards, record that separately so a response to your pitch does not become evidence from an unprompted interview.

Keep an evidence log, not a list of compliments

Download the editable customer discovery script and evidence log (plain text). Open it in a text editor and duplicate the blank entry for each conversation. No paid research or CRM subscription is needed.

• ✓Use a participant code and record why this person fits the proposed audience.
• ✓Separate what happened, an exact quote or labelled paraphrase, and your interpretation.
• ✓Record the current alternative, reported cost and evidence that contradicts your idea.
• ✓Mark unknowns as unknown, and distinguish an offered next step from one actually completed.
• ✓Set a decision and review date before collecting the next round of evidence.

Keep contact details separately and restrict access. A code does not make notes anonymous if you can reconnect them to a person. Collect only what the decision needs, explain your retention approach and delete identifiable research notes when they are no longer needed. Agreement to a research conversation is not permission to add someone to a marketing list; follow the email permission guidance if you invite them to subscribe separately.

A fictional result that changes the next test

Suppose five freelance designers discuss their most recent invoice. Three describe chasing missing purchase-order details; one reports no difficulty and one cannot recall a recent example. Two of the three already use invoicing software. I would test a clearer client-information checklist before building another invoicing app. That is an interpretation of five fictional conversations, not proof of market size or a conversion forecast. The next test could ask suitable participants to use a checklist on a real invoice, with their agreement, and report what still goes wrong.

Research and privacy guidance checked on 28 September 2026: GOV.UK on in-depth interviews, participation consent and the ICO data-minimisation principle. The script is my suggested structure, not a validated survey or a claim that five interviews are sufficient.

Choose the smallest honest test

A manual pilot, paid workshop, clickable prototype, sample deliverable or focused landing page can test different risks. A smoke test must not imply that a finished product exists when it does not. If you accept pre-orders, state the delivery date, refund terms, product status and trader details clearly.

A seven-day validation plan

1. Day 1: list assumptions and choose the riskiest one.
2. Day 2: identify ten plausible customers and write neutral questions.
3. Days 3 and 4: run five interviews; update the problem language.
4. Day 5: create the smallest honest offer and delivery outline.
5. Day 6: invite a relevant group individually; follow direct-marketing rules.
6. Day 7: compare results with the pre-set threshold and decide.

The experiment sheet records one risky assumption. When the evidence is strong enough for a broader operating decision, use my one-page business plan template to connect the customer and offer to sales activity, delivery capacity, money, risks and the next 90 days. Do not copy untested assumptions across as facts.

Interviews, a landing page or a paid pilot?

I would use interviews first when the problem itself is uncertain, a landing page when the uncertainty is whether a clear message attracts a relevant next step, and a paid pilot when willingness to pay and delivery effort are the decisive risks. Interviews offer detail but can reward polite answers. A landing page tests a message and route to attention but cannot establish delivery quality. A pilot asks for a stronger commitment but takes real work and must have clear terms. For a small service with an understood problem, my preferred next test is a tightly scoped paid pilot.

Read a small result without overstating it

Hypothetical result: ten suitable people receive the same offer, four discuss it and two buy a £40 session. That is 40% reaching a conversation and 20% buying among those ten, with £80 gross sales. It is not a reliable population conversion rate. If delivery takes three hours in total, gross sales per delivery hour are £26.67, before research, sales time, costs and tax. The buying signal could pass while the delivery economics fail.

The example counts and seven-day schedule are planning choices, not validated minimum samples or a promise that research finishes in a week. Record refusals, non-responses and what changed between offers. Repeat with another relevant group before treating an early result as repeatable demand.

Spot something that needs correcting? See the corrections process.