Product design AI-assisted build Retail operations Live testing Supabase

Clinch · Retail operations · 2026

Five hours to a shift planner the store corrected 224 times

My store failed a big-retailer audit in July. I built the missing piece in five hours, put it on the floor the next morning, and let a month of real shifts rewrite it.

Client My own store, not a client project Role I chose what kind of memory each job gets Timeline 2–30 August 2026 Tools AI-assisted build · Supabase · Vercel

What shipped

Version one on the shop floor in five hours, 2 August 2026.
55 build jobs and 224 logged changes by 30 August, each one traceable to a shift.
Handover, training and audit actions now survive the date change. Every tick carries a name and a time.
Store result: not measured yet.

Context

The audit named six failures. They had one cause.

Two highs. Four mediums. Safe systems of work, cash reconciliation, CCTV refunds, database accuracy, price checks.

Nobody had skipped the work. The work had worked loose.

Our procedures lived in company systems, laminated sheets, emails, calendars and the heads of five experienced managers.

No single place could say what mattered now, what was already done, or what the next manager was about to inherit.

The Clinch audit view listing the six open audit findings, graded high and medium — The six findings, later loaded into the app itself. Two highs, four mediums.

Problem

Three ways a busy shift loses work

No plan for the shift. A 7am opening holds fire doors, cash, red orders, stock, weekly checks and the team brief. The customer at the counter wins. A safe count sits quietly.
A tick proved nothing. It could not say who did the job, when, how long it took, or whether someone hit Stop by mistake.
Unfinished work vanished at midnight. Jobs built up all morning, then landed on the late manager with no owner and no context.

Version one of the planner, with three pins marking the flat task list, the anonymous tick and the work that reset overnight — Version one, with the three problems pinned on it.

So the design question got simple. How do I take the week out of managers' heads and put it where the whole shift can see it?

The build

Five hours to version one. Then the store took over.

I shipped version one on 2 August after roughly five hours of AI-assisted build. A weekly calendar, timed blocks, checklists.

Testing started the next morning. It was wrong in useful ways.

Four weeks of live correction, compressed.

Decision 1

The people using it rewrote the plan

Morning Paperwork needed until nine, not half eight. The trading check belonged on Monday. Two task lists became one.

Before and after of the Morning Paperwork pane, from one flat list to four labelled bundles with a PI count tile — Fifteen flat jobs became four timed bundles with a weekly stock-count pace.

Timing and wording turned out to be the product. A planner can look tidy and still tell the store to do the wrong thing.

I deleted as much as I added. The weekly focus pill competed with the calendar, so it went. So did a notes box and every duplicate control.

Before and after of the header, showing the weekly focus pill removed and role columns added — The weekly focus pill went. The calendar is the focus.

Decision 2

A tick became a data card

Deliveries now record the delivery number, the invoice count, missing SKUs and elapsed time. The timer asks before it saves, so a bad tap can be undone.

Before and after of the TCND delivery block, from a four-item checklist to a card with number, invoice and timer fields — A checklist became a data card: delivery number, invoices, timer, issues.

Stock Monster does the same for the clear-down. Managers enter the stock left through the day.

The Stock Monster pane showing a stock count of 45, down 265 from the previous count, with a daily pace line — It works out the pace, and the rate the team needs to clear by seven.

Decision 3

Each job got the right kind of memory

Daily checks reset tomorrow. Training stays done. A compliance module runs for weeks, carrying its own people list and its own deadline.

The Training pane, showing a module with one person outstanding and ten completions stamped with dates and initials — A module keeps its own people and stays done overnight.

Handover became a shared list that survives the date change. A job stays open until someone closes it, stamped with the date and the initials.

The Clinch handover: a shared action list with importance and urgency flags, an overdue counter on each job, owner initials and a four-tile shift summary — Real use, 17 September. Two of these jobs are 20 and 18 days old. Under the old roll-up they would have gone at midnight.

The next manager sees the shape of the day without opening a single block.

Clinch looks like a calendar. Its real job is memory.

To clinch a nail is to bend the point over on the far side, so it can never work back out.

Impact

What moved, and what has not

5 h

Idea to first live version

Build jobs, 2–30 August

224

Logged changes, each traceable

The product moved. The store number has not been read yet.

The reading will cover checklist completion, training cleared before deadline, and open audit actions.

Until then this page proves I can build it. It does not prove the next audit gets easier.

The Clinch work log: 55 numbered jobs across August 2026, each with its date and the range of change numbers it produced — The work log. 55 jobs, 224 numbered changes, every one traceable to a shift.

Learned

What the store taught the product

Task data is part of the design

A clean screen cannot rescue the wrong timing, vague wording or a duplicate job.

Different jobs need different memory

A fire-door check, a month-long training module and an audit action each need their own reset rule.

Delete your own ideas

The focus pill felt useful on my laptop. On the floor it stole attention from the calendar.

Next up · Wellnest →