Job
Count what's there
People through a door, vehicles in a yard, stock on a shelf — a running count fed into the roster, PSA or ERP that already depends on the number.
Eastbourne · UK
/ AI Solutions / Vision AIImages that notice what actually matters — and tell the right system.
Live video, still images, scans, uploads — turned into a signal your other systems can act on. Count, classify, flag, route. Built around the decision you need made, not the model we want to show off.
01/ What this actually is
An image, a decision, a system that acts.
Vision AI, stripped of the hype: an image — live from a feed, a phone upload, a scan at reception, a stored frame — picks up something you care about, a model turns that into a structured event, and the system that already runs that workflow picks it up and does the thing. The model is the easy part now — the value is in what happens next.
02/ Jobs it does
Six shapes of real-time vision work.
Real-time led — live feeds, live alerts, live counts — though the same stack handles batch image review when the job calls for it.
Job
Spot what shouldn't be
PPE missing, unauthorised access, equipment left running — the workflow fires, not just a notification.
Job
Read without typing
Number plates, serials, batch codes, handwritten forms — lifted off the image and handed to the system that needs the value.
Job
Route on sight
Event lands as a ticket, a page, a door unlock, a shift note — the right thing, triggered automatically, no human relay.
Job
Inspect at speed
Defect flagging on a line, damage check on arrival, quality sampling — the exception goes into the QA flow you already run.
Job
Measure the room
Dwell, queue, occupancy, flow — the operational numbers land in the dashboard your team already looks at.
03/ What's actually possible
Things most people don't realise vision AI can do.
Live video, a still off an uploaded phone photo, a scan at point of booking, a frame from a dashcam or a drone — all the same job once the pixels are in. A short list, not an exhaustive one. The ones that make people sit up.
Can I…?
Counting things a till can't count.
- The queue at the counter — length, and how long it's been there.
- Dwell time in front of a display versus the one next to it.
- How empty a shelf is, in real time, not at stocktake.
- People through the door vs till transactions — your actual conversion rate.
Can I…?
Spotting what changes.
- A fire door propped open. A walk-in freezer left open for ninety seconds.
- A tool that was on the board this morning and isn't now.
- A vehicle that entered a zone and didn't leave.
- A damage photo on arrival vs the photo at collection — delta flagged, claim routed.
Can I…?
Present, absent, present — object journeys.
- A tool on the board, gone, and back — normal. Gone too long — missing.
- A trailer in slot 7, gone, reappears in slot 11 — tracked without a beacon.
- A fire extinguisher that was in its bracket last week and isn't now.
- A part on the hook, picked up, tried, put back, not bought.
Can I…?
Reading images without typing.
- A fill-level sight glass on a tank with no sensor and no PLC port.
- The colour of a stack lamp — amber, red, green — on a machine from 1987.
- A handwritten collection form, photographed at reception and lifted straight into the ERP.
- A meter or gauge, snapped on a phone during inspection and logged as a reading.
04/ How we'd approach it
Pilot first, rollout second.
Vision work lives or dies on the footage. Every engagement starts with a short proving pass on your actual footage before anyone commits to a wider rollout.
Frame the decision
Start with what should happen when the model sees X. The model itself is a detail — the decision and the downstream action are the job.
Prove it on your footage
A short proving pass on your actual footage or sample images — real lighting, real angles, real edge cases. If it doesn't hold up here, nothing downstream will.
Wire it into the system that acts
The output of the model lands in the tool that already runs that workflow — your PSA, ERP, CRM, ticketing, paging, access control — not a new dashboard no-one logs into.
Tune in place, humans in the loop
Live with a human confirming the edge calls until the model earns trust. Thresholds, exceptions and escalation stay yours, not the vendor's.
05/ How we scope it
Honest about the footage before anyone commits.
Lighting, angles, edge cases — vision projects succeed or fail on the raw feed, not the demo. We size the commitment accordingly.
Every vision engagement starts with a short proving pass on your actual footage — real lighting, real angles, real edge cases — before anyone commits to a rollout. If it doesn't hold up there, you know inside a fortnight rather than a quarter. The bill for that phase is small and the decision it produces is a clear yes, no, or “yes, but only for these conditions.”
06/ Where it lands
The signal goes where the work already happens.
A vision event is only useful if it fires the thing that should happen next. That's where the rest of what we do earns its keep.
/ Into your systems
Events land in the systems your teams already use — PSA, ERP, CRM, ticketing, paging, access control. That integration work is our day job, documented on the systems integration page.
/ Into your workflows
Events trigger the downstream automation — tickets open, rosters update, alerts fire, approvals queue — handled the way the rest of our workflow automation work is handled.
/ From your footage
Custom models trained on your actual footage, deployed to an edge device on-site or a cloud endpoint we stand up for you. The training data, the weights and the integration logic are yours — not a locked vendor silo you rent forever.
07/ When it's not a Labs job
Cameras that just need to alert you.
Off-the-shelf person, vehicle, plate or face detection on a managed VMS is a different product to a bespoke vision build — and the right answer for plenty of real problems.
That work lives with our sister company M-Tech Systems — UniFi Enterprise partner, Avigilon Alta partner, installs-and-runs the physical estate end-to-end. If that’s closer to what you need, we’ll make the introduction. Faster, cheaper, and the right tool for that job.
Talk to M-Tech Systems/ Backed by
Delivered by M-Tech Labs with the compliance and security discipline of M-Tech Systems — Cyber Essentials certified, aligned to NCSC CAF 4.0 and progressing through the Assurix trustmark programme. Camera data, retention and consent handled as first-class concerns, with code continuously scanned for quality and security via Aikido and hosting kept on continuously pen-tested, current-version infrastructure.
Back to AI Solutions/ Start a conversation
Point a camera at the problem. We'll tell you if it's a vision job.
A short discovery call on the decision you want made, the footage you've got and whether vision AI is actually the right tool — before anyone commits to a build.