AI Transformation
AI that reaches production and gets used.
Why it fails
Most AI programmes stall for the same reasons.
The tool was chosen before anyone mapped the workflow.
A broken process got automated and became faster at being broken.
Several pilots run at once and none connect.
Nobody owns the pilot once the demo is over.
The data or the permissions were never wired up.
There is no plan for getting people to use it.
Nobody measured the old way, so nothing can be compared.
An autonomous agent was used where a controlled workflow was safer.
It was run as an IT programme with no business owner.
Software was bought before the problem was understood.
My approach
From problem to a measured result.
Seven steps. Each one earns the next, and nothing ships that can’t be measured.
- 01
Diagnose
Understand the business result, current workflow, people, data, systems, constraints, risk, and baseline performance.
- 02
Prioritise
Rank opportunities by business value, frequency, time spent, error cost, data availability, integration difficulty, adoption difficulty, risk, and measurement speed.
- 03
Design
Choose the right mechanism: better process, automation, AI-assisted workflow, retrieval, copilot, controlled agent, a fully autonomous agent only when justified, or traditional software when AI is unnecessary.
- 04
Pilot
Build the smallest real system capable of proving or disproving the thesis.
- 05
Integrate
Connect the pilot to the actual tools, data, permissions, and people required for everyday use.
- 06
Drive adoption
Define owner, users, training, feedback loop, escalation, human review, and failure handling.
- 07
Measure
Compare against the baseline. No invented examples or results.
Related reading
- essayHow to get cited by ChatGPT and Google's AI OverviewsWhen Google shows an AI summary, users click a result half as often. What gets a page quoted instead, what the research measured, and what I got wrong on my own site.17 min read
- essayHow to become an AI-native software engineer without getting slowerA randomised trial found experienced developers were 19% slower with AI while believing they were 20% faster. This is what closes that gap, and the workflow that makes the speed real.14 min read
Serious AI work to get through?
Tell me the context and what needs to change. I will tell you honestly whether AI is the answer, and whether I am the right person to build it with you.