“Design thinking sounds lovely,” says the owner of a small accounting practice, “but we don’t have a research budget, an innovation team or two spare days for a workshop. We’ve got clients to look after.”
It’s a fair point. Most of what’s written about design thinking assumes the resources of a large company: dedicated researchers, facilitators, specialist software and weeks of time. For a business with five people and a full diary, that version is about as useful as a recipe that begins “first, build a professional kitchen”.
But strip away the workshops and the jargon, and the core of design thinking is surprisingly cheap. It comes down to three habits: understand the problem by watching and listening to real people, test the cheapest possible version of your idea, and let what you learn change what you do. None of those require a budget. They require a little time, a bit of nerve and a willingness to be wrong.
And now there’s a new temptation. AI tools promise to simulate customer interviews in seconds. Could they replace the conversations altogether?
One Like insight: The most valuable research a small business can do usually costs nothing but an hour and a bit of courage.
You need fewer people than you think
One reason small businesses skip research is the assumption that it needs large numbers to mean anything. For some questions, such as estimating how many people will buy something, that’s true. But for the most common design question, “Where are people getting stuck?”, you need surprisingly few.
In a famous article from 2000, the usability researcher Jakob Nielsen of the Nielsen Norman Group summarised earlier work he had done with Tom Landauer on how many problems each additional test user reveals. Testing with a single person typically uncovered almost a third of the usability problems in a design. By the fifth person, you had found around 85%, and each additional person mostly repeated what you’d already seen. His blunt summary was that “zero users give zero insights”, and the jump from nothing to a little is enormous.
Nielsen’s advice wasn’t to stop at five, though. It was to run several small rounds: test with five people, fix what you find, then test the improved version with another five. He also cautioned that if you serve very different groups, such as parents and children, you need a few people from each.
The research was about websites and software, but the principle travels well. Watch five customers try to book an appointment through your website, fill in your quote request form or find their way around your shop, and you’ll spot most of the obvious problems. You won’t learn how many people will buy, but you’ll learn where they’re getting stuck, and that’s often worth more.
Highlight: Five people won’t tell you how many customers you’ll win. They’ll tell you where you’re losing them.
Prototype the promise, not the product
The word “prototype” conjures images of 3D printers and expensive mock-ups. For most small businesses, a prototype can be far humbler. It’s anything that lets a customer react to an idea before you’ve fully built it.
A few examples show the range. An accountant considering a fixed-fee monthly package could write a one-page description, show it to five existing clients and ask which they’d choose over their current arrangement. A café thinking about a lunchtime delivery service could run it manually for two weeks, with the owner taking orders by WhatsApp, before investing in an app. A boutique planning a new layout could move a single display and watch how people walk through the shop for a few days. None of these cost much, and each one tests the most important question: will people actually choose this?
The principle is to test the promise before you build the product. Sketches, a mock-up of a web page, a hand-written menu insert or a manual pilot can all reveal whether the idea resonates, long before the expensive part begins. And because each test is cheap, it becomes easier to try two or three options at once, which helps protect against falling in love with your first idea.
[Infographic placement] The prototype ladder. A staircase rising from left to right, with cost and commitment increasing at each step. Step 1: A conversation about the problem (cost: an hour). Step 2: A sketch or one-page description (cost: an afternoon). Step 3: A mock-up or simple web page (cost: a day). Step 4: A manual pilot with a handful of real customers (cost: a few weeks). Step 5: Full launch (cost: significant). Caption: Climb only as high as you need to answer your next question.
There’s one discipline that makes these tests worth doing. Decide before you start what result would make you change course. If fewer than two of your five clients prefer the new package, perhaps you rethink the price or the offer. Writing it down protects you from the very human temptation to explain away results you don’t like.
Highlight: Climb the prototype ladder only as high as you need to answer your next question.
Can AI do the talking for you?
The newest shortcut in this space is the “synthetic user”: an AI-generated stand-in for a customer that you can interview in seconds, at any hour, without the awkwardness of asking real people for their time. For a busy business owner, the appeal is obvious.
The Nielsen Norman Group has been testing these tools. In a study published in October 2026, its researchers took a question they had previously asked 604 real design and technology professionals and put it to 120 synthetic respondents. The headline result was reassuring: both groups pointed to the same overall conclusion. But the detail was revealing. The humans used more than 50 different labels in their answers. The synthetic respondents collapsed onto about a dozen, and none of the unusual answers real people gave appeared at all. The AI also pulled different groups towards the same response, so differences between types of people were understated.
NN/g’s conclusion was clear: “Research with synthetic users is still not user research,” because no actual users are being studied. Its researchers suggest synthetic respondents can help generate hypotheses or check broad patterns, but not replace people when the less common views matter.
For a small business, that distinction is crucial. The most valuable things customers tell you are usually the ones you didn’t expect: the reason they nearly didn’t book, the competitor they compared you with, the odd way they actually use your product. An AI trained on what most people typically say is, almost by definition, least likely to give you those surprises. It can help you prepare better questions. It can’t answer them for your customers.
Highlight: AI can tell you what customers usually say. Only real customers can tell you what you didn’t know to ask.
What this means for a small business
You don’t need a design thinking programme. You need a rhythm. Once a month, pick one thing in your business you’re unsure about, perhaps a confusing step in your booking process, a service that isn’t selling or an idea you’re excited about. Then spend a few hours learning about it from real people before you spend money on it.
That might mean watching five customers use your website, calling three clients to ask about the last time they nearly went elsewhere, or running a small manual version of a new offer for a fortnight. Use AI to help you prepare questions, summarise your notes or draft a rough mock-up. Keep the conversations themselves human.
A few questions can help:
- What’s the one thing in our business we’re least sure customers actually want?
- Who are five real customers we could watch or talk to this month?
- What’s the cheapest possible way to test our next idea?
- What result would make us change our plans, and have we written it down?
Which brings us back to the accountant who said they didn’t have time for design thinking. They were right about the workshops. But an hour with five clients, a one-page description of a new package and a willingness to hear “not for me” is design thinking. It just doesn’t need the sticky notes.
Sources
- Jakob Nielsen, “Why You Only Need to Test with 5 Users”, Nielsen Norman Group, 18 March 2000. Based on Nielsen and Landauer, “A mathematical model of the finding of usability problems”, Proceedings of ACM INTERCHI’93.
- Rachel Banawa, “Synthetic Users vs. Human Respondents: Same Conclusion, Less Variation”, Nielsen Norman Group, 16 October 2026.
