Let people lead the listing
The first flow leaned heavily on AI. Users preferred entering the details themselves.
I replaced it with a simple form followed by an editable AI draft. Users kept control of what they published.
AI-driven product building · Marketplace · UX design
Čistý kout
The right AI interaction depended on the task.
I turned my own search for cleaning help into a live marketplace—and changed the design when users showed me where AI helped.
At a glance
User feedback changed listing creation. Guided onboarding helped providers add profile details.
Live marketplace; shown redesign deployment not confirmed.
- Contribution
- Solo creator · Product, UX & AI-driven development
- Period
- Recent product work
- Status
- In production · Real users
- Live
- marketplace with real users
- Solo
- product, design and implementation
- 2 tasks
- different lessons about AI interaction
The challenge
I needed cleaning help and started a Facebook group that grew to 70,000 members. Čistý kout became the next step: helping households find providers and agree on the work.
Households struggled to judge reliability. Providers wanted leads without repeated calls and heavy admin. The product had to build trust while keeping coordination simple.
My role
I owned discovery, product direction, UX and implementation. I built the marketplace solo with AI assistance, using my OpenClaw agent setup for the first time on a live product.
Users, needs and an example journey
The starting point: separate finding cleaning work, managing a small team and hiring staff. On the client side, map what makes a cleaning arrangement dependable in each context.
Provider personas jobs to be done
- Individual cleaner
- Find suitable local jobs, establish trust and agree on scope without repeated calls.
- Small agency owner · 1–3 cleaners
- Keep the team booked, match people to jobs and coordinate availability with little admin.
- Larger agency owner
- Usually has internal systems; needs temporary cover or permanent staff to meet demand.
Client personas jobs to be done
- Working professional
- Arrange reliable cleaning around work and reclaim personal time.
- Family
- Find someone trusted who can maintain a clean home around family routines.
- House owner
- Agree on the scope, price and frequency needed to care for a larger property.
- Short-term accommodation / Airbnb host
- Arrange dependable turnovers between guests, with clear timing and backup cover.
- Organisation / company
- Hire a reliable cleaner for the premises, with agreed hours, responsibilities and continuity.
Design focus: make scope, availability and trust clear before contact. Keep small-team coordination lightweight; treat larger agencies primarily as recruiters with existing tools.
Sample journey · a working professional
Illustrative scenario: finding a cleaner for a recurring weekly visit.
- Define the need — Cleaning falls behind during a busy week. Set the tasks, frequency and preferred time.
- Find a match — Compare local providers. Need: clear services, availability and reasons to trust them.
- Agree on the visit — Clarify scope, price and access. Need: one shared understanding before committing.
- Try the service — Complete a first clean. Check reliability and quality against the agreed expectations.
- Make it recurring — Agree on a regular slot if the fit is right, with a clear way to handle changes.
The decisions
Use conversation to get people started
Only about 5% of providers completed their profiles before AI onboarding. I introduced a guided conversation to help them get started.
After AI onboarding, photo uploads reached 27% and completion of additional fields, such as location, reached 72%. These field-level rates are distinct from full-profile completion.
Explore the redesigned experience.
Captures from the working local application, showing discovery, the provider experience and AI-assisted requests.
Introducing Čistý kout.
An overview of the provider experience, from presenting services to receiving cleaning requests.
What changed
The live marketplace provided a place to learn from real use. Feedback changed listing creation to a form with an editable AI draft; conversational onboarding helped providers add profile details. The redesign shown here is separate from the confirmed live product.
Provider onboarding: the reported figures
- Before AI onboarding~5%
- completed their profile
- After AI onboarding27%
- uploaded a profile photo
- After AI onboarding72%
- completed additional fields, such as location
Founder-reported figures from comparable groups. Full-profile and individual-field completion are different measures. Detailed denominators, sample sizes and measurement periods are unavailable, so no overall completion uplift is calculated.
Adoption context
Founder-reported adoption: 1,000+ active app users after a few months, with no paid promotion. The activity definition and dates remain unconfirmed. The related 70,000-member Facebook community is a separate audience.
What I’d measure next
- Measure
- Full-profile completion, photo upload and required-field completion, each measured separately.
- How
- For future onboarding cohorts, record the denominator, period and provider count for each measure. Use a consistent full-profile definition and equal follow-up time, and track photo uploads and required fields separately. Connect completion to time to first useful enquiry.
- Check alongside it
- Comparable groups do not make different completion measures interchangeable: do not compare the earlier ~5% full-profile rate directly with the later 27% photo or 72% additional-field rates. Keep the activity definition for 1,000+ users unconfirmed until it can be established.