Connect Prototype Testing From Recruitment to the Next Build

What got faster is screen creation

A screen prototype is a model where people can press buttons and move to the next screen, even though real payments or recommendations do not work. AI-assisted tools can create these models, but production time still depends on scope, and you can only learn whether people understand them by having people use them directly.

GENIE AI in Korea combines screen design and web prototype creation with testing by real users. For consumer services, it describes a target-customer group of 30 people: five join video interviews and 25 complete online surveys. It gives a total timeline of one to two months. That is roughly the input of one small classroom, but it does not disclose client improvement results or pricing.

Wave Rider lists AI prototype production costs as KRW 2 million for basic setup, KRW 3 million for one feature hypothesis, and KRW 1.5 million for one AI model. Choosing one of each totals KRW 6.5 million, and the work usually takes one to two months. Even if screens can be drawn quickly, assigning real functionality and connections to external information can still make the idea-validation stage costly.

In research on AiKINO, a generative-AI short-form video production service, nine non-specialists took part in usability testing.

These examples point to a bottleneck that is not screen-making time. It is deciding who receives the prototype link, explaining what they should try, recording where they get stuck, and moving that evidence into the next revision and production work. As screen creation gets faster, this downstream manual work repeats more often.

Where the work breaks for a video-service operator

Consider someone running a paid course-video service with one-to-one consultation booking and a team of three. They want to change the flow in which a new member chooses a course, pays, and then books a consultation. Before handing the work directly to a development agency, they create a clickable screen prototype in one evening with an AI-assisted tool.

Once the prototype is ready, the first break appears. To recruit participants, the operator exports a customer list, creates an application form, sends links through KakaoTalk, and transfers available times into a calendar. Names and contact details are entered four times: in the application form, calendar, video-meeting invitation, and reward list.

The second break comes when tasks are sent to participants. The operator messages, “Choose a course, make a payment, and then try booking a consultation.” But it is hard to tell whether a participant explored the prototype first or started the task immediately. If the link does not open or the payment button does not respond, the operator has to ask again whether the problem is the screen or the prototype connection.

The third break comes when organizing observations. The operator finds the moment where a participant stopped in a video-meeting recording, notes the time, captures the screen, and adds an explanation in a separate document. Problems with different causes—such as “could not find the payment amount” and “mistook consultation booking as included in the payment”—can be flattened into a one-line revision request when time is short.

A service that connects this process would first ask the operator for just three things: a prototype link, participant criteria, and tasks to try. Participants read the consent language and start the task from one link. The service collects the screens where they paused for a long time or went back. The operator tags each moment with labels such as “button naming problem,” “sequence problem,” or “prototype error.”

When testing ends, the result should be a revision bundle rather than a long report. In one place, it should show which screen needs to change, how many people got stuck in the same place, what they actually said, and what they should be able to do after the change. Instead of sending a development agency “revise the payment screen,” the operator can send a completion state: “Participants should be able to find the total amount and choose whether consultation is included without extra explanation.”

Some work still belongs to people. The operator must judge whether participants are actual target customers, decide whether to trust words or behavior when they differ, and fix repeated confusion across several people before one person’s strong complaint. A service should not replace that judgment. It should keep evidence from recruitment through revision handoff from disappearing.

Validation operations have already become a product

Goosechase, a Canadian participatory-event service, used Lyssna to decide where to place a new AI feature for users such as teachers, HR staff, and tourism operators. It created two layouts in Figma, a screen-design tool, and gave the same tasks to external participants and existing customers. Contrary to the team’s expectation, users clearly preferred the other layout, so the team changed it before development.

Lyssna can recruit a service’s own customers for testing, and charges by participant count and test length when external participants are needed. Its Growth plan costs about KRW 230,000 per month when billed annually, while external participants start at about KRW 1,400 per test minute per person. The claim that results arrived within minutes comes from a customer story published by the provider, so speed can vary with participant conditions.

Zigzag, a UK dog-training service, created Ziggy, an AI conversational assistant for handling lighter questions that had been handled by people. It first found through customer interviews that users did not want to interrupt human coaches with minor questions. It then showed screen prototypes to prospective and existing customers, revising the assistant’s placement and the way it asked questions.

UserTesting said that in the six months after launch, Ziggy had 80,000 interactions, received 96% positive feedback, and reduced support requests reaching human coaches by 27%. These are all figures published by the provider, so they do not establish that other services will get the same results. UserTesting sells participant recruitment and test operations through annual custom quotes and does not publish fixed prices.

Things you can build now

1. A prototype-validation operations board

• What it does: Manages prototype-link registration, participant applications, scheduling, task guidance, reward status, and discovered issues in one place.
• Who uses it: An operator of a small video service that sells paid courses and consultation bookings together.
• Why now: Screens can produce several versions in a day, but the people and schedule for each test still need to be organized again each time.
• First screen: For each active prototype, show the target recruitment count, completed bookings, completed tests, and screens where people repeatedly got stuck side by side.

2. A friction-moment collector

• What it does: Collects where participants paused for a long time, went back, or clicked incorrectly while completing tasks, together with the relevant screen moment.
• Who uses it: A person at an independent video-subscription service redesigning sign-up and recurring payments.
• Why now: Prototype versions multiply quickly, while rewatching every recording from the beginning slows revision work.
• First screen: Show the number of people who paused and a representative moment for each prototype screen, then let the operator add a cause tag immediately.

3. A human-operated testing ground for AI features

• What it does: Tests question style, response speed, and the point for handing off to a person before building the AI feature, with an operator answering behind the scenes.
• Who uses it: An education-service operator considering an assistant for video-content recommendations or course consultation.
• Why now: Before building the actual feature, they can collect what users ask, how they phrase it, and which answers they do not trust.
• First screen: Put the participant’s conversation screen on the left, and the operator’s response field, examples of previous answers, and a handoff-to-human button on the right.

4. A service that turns user evidence into a production brief

• What it does: Groups test notes and friction moments into screen-by-screen revision lists and completion criteria for the production team.
• Who uses it: A content-business operator assigning reservation, sign-up, and payment screens to an external development agency.
• Why now: When observations are scattered across messengers, recordings, and documents, the reasons disappear and only instructions such as “make the button bigger” remain.
• First screen: Show the screen to revise, the repeated problem, evidence moments, the intended change, and what the user should be able to do afterward as one bundle.

Why this matters where you are

AI can speed up prototype creation, but recruitment, task delivery, observation, and revision handoff can still be fragmented across tools. Check how many tools your own testing process requires and whether the same participant information is copied between them. The pricing, participant sources, and privacy requirements may differ in your market, but the workflow breaks described here are concrete places to investigate.

What to check in 30 minutes today

Send a prototype link to one current customer and, while sharing the screen, ask them to complete one task from sign-up through payment or booking. Do not explain anything. Mark on paper every time you switch tools or move participant information or a friction moment somewhere else.

If completing the test and recording it requires three or more tools, or if you move the same information more than once, connecting validation operations may be a problem worth working on before improving screen creation. If the participant cannot finish the task without help, that single moment is core material your first product should collect.