Artificial intelligence
AI, grounded in operational excellence
If operational excellence is the engine, AI is the new fuel. But fuel does nothing for a broken engine.
Why take AI advice from an operations consultancy?
You cannot automate a broken process — and AI on bad data scales bad decisions.
Most AI initiatives in industry fail on the process, not the technology: pointed at something nobody had stabilised, using data nobody had validated.
We are not an AI shop that discovered manufacturing. We came the other way round — from Lean, Six Sigma and change management into AI.
01
What AI really is
We train people who do not program. A working mental model, and no hype.
A language model is not a search engine
A very well-read pattern machine. Superb at drafting, hopeless as the authoritative record of anything.
RAG: teaching a model your own data
It answers from your documents instead of its training — your standards, your specifications, your procedures.
Public tools versus in-house
A consumer chatbot and a deployment inside your data boundary are not the same thing. We cover how to tell.
How to judge a tool
Not by the demo. By where your data goes, whether the output can be checked, and whether the plant can run it unaided.
02
Where to start — and where not to
The honest version: the best uses are unglamorous, and the most tempting ones are a bad idea today.
Good places to start
Drafting and summarising
Shift reports, incident write-ups, minutes. You stay the editor.
Getting to information faster
Ask procedures, specifications and history a question instead of hunting for the folder.
Preliminary document checks
A fast first pass, with the final check kept firmly with a person.
Not here — not yet
Quality records and critical decisions
Batch release, deviation closure, anything an auditor will read: validated systems and human sign-off, not a language model.
AI is a great drafter and sparring partner. Not, yet, a system of record on the shop floor.
03
What changes in practice
The real shift is who gets to try an idea, and how long they wait.
| The old default | The new default |
|---|---|
| Idea → ticket → roadmap → maybe next year | Idea → try it this afternoon |
| Improvement is an IT project | Improvement is a daily reflex |
| Months of waiting per use case | Hours, not months, to first value |
| Power sits with the few | Power in the hands of the doer |
| Many ideas die in the queue | Most ideas actually get tested |
04
Responsible use is part of the training
In a regulated plant, an AI rollout with no rules is a liability. We teach the guardrails alongside the tools.
- Know where your data goes before you paste anything.
- No AI output becomes an official record without human sign-off.
- Every answer is a draft to verify, not a fact to forward.
- A person stays accountable for every decision.
- Augment your thinking — don’t replace your judgment.
05
From chatbot to AI agent
An agent does not just answer. It takes a goal, uses tools and acts — more value, and more to supervise.
An agent can be given a job rather than a question, and reaches your systems through tools instead of living in a chat window.
Participants build a working agent during the course rather than watching a slide about one.
The Andon agent
When a line stops, most of the useful work is gathering — and it can happen while people deal with the machine.
- Captures the stoppage in a consistent structure, not a free-text note
- Pulls the relevant history, so nobody rediscovers a known failure
- Notifies the right people and logs it where it can be analysed
- Leaves the diagnosis and the decision to the humans
06
How we engage
Always from a process and a number, never from a tool looking for a home.
- 01
AI readiness assessment
Are your processes stable and your data clean enough for AI to help? You get a shortlist ranked by feasibility and value.
- 02
Hands-on training
From structured problem solving through AI tools and Python in Excel to building your own agents, on your own processes.
- 03
Proof of concept, fast
We work alongside your people on a first use case, so the capability stays with them.
- 04
Governance a regulator would accept
Data boundaries, human sign-off, and which decisions AI may never own.
Where would AI actually help you?
Tell us the process and the number you want to move. We will tell you honestly whether AI is the answer.