The jam
Stock is counted in basements and loading bays where the signal dies. Entries get lost, or written on paper “for later”.
What we’d build
An offline-first mobile app. Every scan and every count is saved on the device, queued, and synced automatically when the van is back in range.
The threshold
Zero entries lost after 48 hours without network.
The jam
The same file travels by email. Nobody knows which copy is right, and Friday’s figures contradict Monday’s.
What we’d build
A web app with one database, clear roles and a full history of who changed what.
The threshold
Any figure found in under 10 seconds, by anyone allowed to see it.
The jam
Technicians fill in paper forms on site. Someone retypes them at night. Errors creep in, and invoices leave late.
What we’d build
A mobile form with photos, signature and GPS timestamp. The PDF report generates itself and reaches the office before the technician leaves. Or the technician simply dictates, and AI turns the voice note into a structured report.
The threshold
Report delivered before leaving the site. No retyping.
The jam
Half the phone calls are customers asking for a status that already exists somewhere in your system.
What we’d build
A customer portal, with automatic notifications at each step.
The threshold
Status calls cut by half within three months.
The jam
An app built years ago, by someone who left. It still works, until the next iOS update.
What we’d build
An audit, a stabilisation plan, then updates in small, safe steps. We document everything as we go.
The threshold
Crash-free sessions above 99.5%, and a codebase someone else could pick up.
The jam
When the compressor stops, only one person knows why. He gets called at night, on holiday, from other sites. The day he retires, the knowledge leaves with him.
What we’d build
A diagnostic tool. Scan the machine’s QR code, see its history, follow a guided troubleshooting path from symptom to cause. An AI assistant searches manuals and past reports in plain language, and cites its sources.
The threshold
The right procedure on screen in under 30 seconds. The expert called only for what’s truly new.
The jam
Machines log everything. Maintenance reports pile up. But the pattern only becomes visible after the breakdown.
What we’d build
A pipeline that gathers logs, reports and spreadsheets into one database, and a dashboard on top: failures ranked by cost, downtime, drifts worth a look. AI models learn from past breakdowns, flag the unusual and warn before the next failure.
The threshold
Your five costliest failure causes, ranked, on one screen, within the first month.
The jam
Delivery notes, supplier invoices, certificates, test reports. All retyped by hand, all a source of errors.
What we’d build
AI document reading: a photo or a PDF goes in, structured data comes out. When the AI isn’t sure, a human validates in one click.
The threshold
No field saved unless the AI is confident or a human has checked it. No blind automation.
The jam
Customers write by email, form and chat, at all hours. By morning the backlog is waiting, and the urgent request is buried at the bottom.
What we’d build
An AI agent connected to your tools. It answers common questions, qualifies each request, creates the order or the ticket, and alerts a human for anything sensitive.
The threshold
Every request picked up in under a minute, day and night. Your team handles only what truly needs them.
The jam
“Which customers are ordering less?” The answer exists, but it takes two days of spreadsheets to find it.
What we’d build
An AI copilot connected to your data. You ask in plain language; it answers with the figure, the chart and the query behind it.
The threshold
An answer in under 30 seconds, with its source shown so anyone can check it.
Releases
Alongside client work, we’re developing our own apps. They’ll ship when they meet our own thresholds, not before.
One email signs you up: one message per release, nothing else.
