Home/What we do/AI & automation
Only if it fitsAI & Automation Opportunity
Practical uses of AI and automation — identified against the work, not against the hype.
Overview
AI can provide information quickly. It cannot tell you whether a process should exist, whether your data can support the idea, or whether the organisation can absorb another system. Those questions need context and judgement.
We look for work that is repetitive, costly, or error-prone enough to change — then test whether AI is required, or whether a simpler operational change would do. The useful outcome is a yes, a no, or a not yet, with reasons.
Four stages before a path is treated as settled.
Each stage leaves an artefact you can challenge. Nothing is carried forward on confidence alone.
- 01
Understand
Study the work, the systems, the people, and the constraints as they are.
- 02
Define
Name the actual problem. Write the non-goals.
- 03
Explore
Hold more than one credible path long enough to compare them.
- 04
Structure
Turn the choice into architecture, owners, and sequence.
Inside ai & automation.
Opportunity scan
Find work that is repetitive, costly, or error-prone enough to change.
Necessity test
Ask whether AI is required, or whether a simpler change would do.
Data reality
Check whether the organisation can actually feed a useful system.
Risk reading
Name the failure modes before a pilot is funded.
Pilot design
Shape a first step that can prove or disprove the case.
Build vs buy
Decide what to adopt, what to assemble, and what to leave alone.
The terms of the work.
| Signal | Likely move | Unlikely move |
|---|---|---|
| Repetition with stable rules | Automation | A generative layer |
| Missing information | Process and data | A summariser |
| Judgement under exception | Better owners and joins | Full automation |
| High volume, clear intent | AI with guardrails | Another dashboard |
A working note.
The kind of thing this work produces — inspectable, arguable, and short enough to be read.
PROPOSAL A copilot to summarise tickets. TEST Is the delay in writing, or in missing facts at intake? FINDING Intake is incomplete. Summaries would be fluent and wrong. MOVE Restructure intake. Do not fund a model for a data problem.
When this usually appears
- Automation that stalled after a pilot
- AI proposals without a problem statement
- Manual work hidden in spreadsheets
- Customer operations under volume pressure
AI & automation, asked directly.
Frequently asked questions
Will you help us pick a model?
If a model is warranted. Model choice is a later, smaller decision than problem choice.
What if leadership already announced an AI programme?
Then the work is to give that programme a real problem, a boundary, and a first test.
Do you build models?
We design whether, where, and how they should be used. Implementation can sit with your team or a partner.
Adjacent work.
Bring the situation as it is.
You do not need a polished brief to start. A conversation is usually enough to name the problem and decide whether there is work here at all.