A prospective client reaches a consultant’s website, reads three service descriptions, and sends a message: “Which one should I choose?”
The website may look professional. The client still needs help making a decision.
For this fourth instalment of the Product Builder’s AI Arsenal, we’ll explore Design and prototype through a hypothetical consulting business. The example, service packages, and test observations below are fictional teaching material.
Earlier editions covered thinking and research, understanding customers, and communicating findings. Now we move into making an idea tangible. The practical output is a small interactive prototype, a record of what people did, and a reasoned next change.
1. Decide what the prototype needs to teach you
Our consultant offers three services: a focused advice session, a review of an existing offer, and support developing a new service. Prospective clients ask about the differences, scope, and booking process.
That could reflect unclear descriptions. It could also reflect overlapping packages or a need for personalised advice. A redesigned website is one option to investigate.
Start with a question: Can a prospective client identify a suitable service and understand how to request a consultation?
This gives the prototype a job. It needs service information, a way to compare options, and a consultation request. Accounts, payments, and a client dashboard can wait because they don’t answer this question.
A prototype represents an experience well enough to investigate it. It can be a paper sketch, linked screens, or a small working interface. GOV.UK’s prototyping guidance describes choosing the form that suits what you need to learn.
Before opening a tool, write a short brief:
Person: A small-business owner considering external advice.
Task: Choose a service and request an introductory consultation.
Uncertainty: Whether the service descriptions explain who each option suits.
Constraint: The prototype must work on a phone and use fictional details.
If you can’t explain the packages yourself, settle their scope first. AI-generated copy will otherwise conceal a business decision you haven’t made.
2. Explore designs that make different assumptions
AI can help you produce alternative layouts and wording. Give it your checked service details, including exclusions and pricing where available. Ask it to flag missing information rather than fill gaps with invented promises.
For our example, compare two approaches. The first presents three service cards side by side. The second begins with the client’s situation: seeking advice, reviewing an existing offer, or developing a new service.
The cards make comparison direct. The situation-led approach may help someone unfamiliar with consulting terminology, but could guide them towards the wrong service. These are assumptions to examine.
Try this prompt:
Using the service details below, propose two layouts for helping a prospective client choose. One should compare the services directly. The other should start with the client’s situation. Explain the assumption behind each layout. Preserve scope and exclusions. Mark missing information as unknown. Do not invent prices, testimonials, or results.
Review the wording before building. “Offer Review” needs a concrete explanation, such as “Review your existing service package and identify unclear scope or positioning.” Keep that description only if it accurately reflects what the consultant provides.
For a Kenyan version of this scenario, suppose the consultant normally shares service information through WhatsApp. Test how the prototype opens from that conversation on a phone. Check whether the page explains prices in the intended currency and whether the next step fits the consultant’s actual process. These are conditions of this example, not claims about every Kenyan customer.
3. Build one journey people can attempt
Choose the layout you want to investigate first. You don’t need three complete websites to test one service-selection task.
Use a tool that matches the experience you need to represent. Figma Make supports prompt-driven, code-backed prototypes that you can edit visually. Lovable and Replit Agent offer ways to create web applications from natural-language instructions. For this exercise, keep their output to the small interface you need. A paper sketch remains useful if you’re still deciding what information belongs on the page.
For an interactive version, make three connected screens:
ScreenWhat someone can doWhat you want to learnServicesCompare the three options.Can they distinguish the packages?Service detailsReview scope, exclusions, and the next step.Do they understand what they would receive?Consultation requestComplete a demonstration form and see a confirmation.Do they understand what happens afterwards?
Use this build prompt:
Create a mobile-friendly clickable prototype for a consulting business using the supplied service descriptions. Include a comparison page, service details, and a consultation request form. Allow users to return and choose another service. Show a missing-field message when appropriate. Use fictional contact details and label the experience as a demonstration. The form must simulate submission without sending or storing data. Do not add payments, accounts, testimonials, or guarantees. List anything you have assumed.
Check the result yourself. Follow every button, go back, change the selected service, and attempt an incomplete request. Confirm that the demonstration form behaves as instructed. Inspect the phone layout and ensure the text remains readable.
The generated result is a draft you need to inspect. A button that looks clickable may do nothing. A confirmation might imply a booking is confirmed when the business only accepts requests. Those details change what a participant thinks happened.
4. Watch prospective clients attempt the task
Recruit people who resemble likely clients. Friends can check broken links, but their familiarity with your work may hide confusing service descriptions.
GOV.UK’s moderated usability testing guide recommends realistic tasks, neutral instructions, and observing participants as they attempt them. Agree how you’ll take notes or record the session before starting.
For our example, give this task:
“You run a small business and want advice on improving an existing service offer. Find an option you would consider, then show how you would request an introductory consultation. Use the fictional contact details provided.”
Avoid directing them to a particular card or button. Ask them to explain what they’re thinking, then leave room for them to act. If you help, record the intervention; completion with guidance differs from completion alone.
Keep observations separate from explanations:
Fictional observationPossible interpretationNext checkA participant moves between two packages and asks how they differ.The distinction may be unclear.Ask what difference they expected.A participant completes the request and expects an immediate appointment.The confirmation may imply a confirmed booking.Ask what they believe happens next.A participant understands the packages but wants to discuss their situation first.Choosing independently may be unnecessary for this client.Explore what they need from an introductory call.
AI can organise anonymised notes after the session. Ask it to preserve source IDs, distinguish observed behaviour from interpretation, and include conflicting observations. Recheck the output against your notes.
5. Change what the evidence supports
Suppose observation reveals confusion between two packages. Revise the distinction and test again. If people understand both but need advice choosing, consider a general consultation route. If they misread the confirmation, clarify the next step and check their understanding.
Target the observed obstacle. Changing the layout, wording, and booking process together makes it harder to understand what helped.
A small usability test can expose problems and guide revision. It cannot establish a sales conversion rate. Someone completing a demonstration request has not committed to pay. Demand, pricing, and the consultant’s ability to deliver remain separate questions.
AI also creates checking work. For a simple page, editing a sketch may take less effort than correcting a generated interface. Choose the method that answers your question with manageable effort.
Take one task from your own business or product. Write what you need to learn, build enough for someone to attempt it, and record what happened before deciding on the next change.
Further reading and references
Earlier in the series
Prototyping and testing
Tools to explore
The walkthrough is a teaching example. We have not conducted a comparative test of these tools for this article.