OT and IT do not mix easily
Operational technology has different priorities, different lifespans and, correctly, different attitudes to risk than the IT department.
Manufacturing & industrial
The line stops for nobody, including your deployment.
Industrial software lives next to machinery that is older than the software industry, speaks protocols designed before the internet, and cannot be restarted because someone wants to ship a feature. The engineering constraint is not the model or the interface. It is that the physical process is authoritative and your system is a guest in its building.
What buyers here open with
“Can you integrate without touching what is already running?”
24/7
The only operating window that exists on a production line
The pressure
Four realities of building next to a production line.
Downtime is the whole cost
An hour of stopped line is measured in real money, which sets the risk appetite for every change you propose.
Data exists but is stranded
Machines have produced telemetry for years. Most of it goes nowhere and is retained for days.
Skills are retiring
The people who know why the process works the way it does are leaving, and it is not written down anywhere.
The work
What we build for manufacturers
Four systems, none of which is permitted to stop the line.
Shop-floor visibility
Real production state, not the state the planning system believes, surfaced where supervisors can act on it.
Maintenance systems
Moving from fixed-interval servicing to condition-based, using the telemetry the machines already produce.
Quality and traceability
Batch, lot and component traceability that answers a recall question in minutes rather than weeks.
Planning integration
Connecting what the ERP believes to what is physically happening, which is usually where the discrepancy lives.
Read-only first
How we earn the right to change anything.
Software that can affect a physical process starts by not affecting it. Every phase below has an exit condition that the plant agrees to, and we do not move on without it.
01
Phase 0, weeks 1 to 3
Observe and change nothing
We tap existing telemetry through a one-way boundary. No writes, no control, no new hardware on the control network. Exit condition: our numbers match what the plant already believes.
02
Phase 1, weeks 4 to 8
Make it visible
Dashboards for supervisors showing real production state rather than planned state. Exit condition: the plant uses our screen instead of the whiteboard, unprompted.
03
Phase 2, weeks 9 to 14
Predict, but only advise
Condition models raise maintenance recommendations to a person who accepts or rejects them. Exit condition: acceptance rate high enough that the plant trusts the queue.
04
Phase 3, from week 15
Write, with interlocks
Only now does anything write back, only to systems agreed in writing, and always behind the existing safety interlocks. Exit condition: a documented rollback that the plant has executed once as a drill.
Non-negotiable
Four rules the plant sets, not us
- Read before you write
- The first version observes and changes nothing. Trust is earned before any system is allowed to affect the process.
- Network segregation
- The control network is not the office network, and integration happens through a defined boundary, not by flattening the two.
- Fail safe, not fail open
- If the software is unavailable the process continues as it did before. Software must never be able to stop the line.
- Long lifecycles
- Equipment lasts decades. Anything built here has to survive vendors going out of business.
Where AI actually lands
Four places it earns its cost, and one line we do not cross.
Worth doing
- Predictive maintenance on real telemetry, which is the clearest return in the sector when the data is actually retained.
- Visual quality inspection, consistent at a volume and duration no human can match.
- Yield and process optimisation, finding the parameter combinations that a manual search would never reach.
- Capturing retiring expertise, turning decades of judgement into something the next shift can query.
Where we stop
Closed-loop control of a physical process without hard safety interlocks. The engineering discipline that governs machinery predates this technology and is not going to relax for it.
Evidence
We are early in this sector and would rather say so than dress up an adjacent project as a manufacturing reference.
How we deliver it
The services this work draws on.
Enterprise Software Development
Platforms that hold up under real load, real compliance, and real headcount.
AI Modelling & Development
Production ML pipelines, retrieval systems, and evaluation harnesses built to survive audit.
DevOps
Pipelines, infrastructure as code, and deploys that stop being an event.
Other sectors we work in
Next step
Talk to the engineer who would run your build.
No discovery call with a salesperson, no deck. A senior engineer reads what you send and replies with a real assessment, including when we think you shouldn’t build it.
