Problem clarity
Can we describe the problem, who experiences it and the cost of leaving it unsolved—without mentioning AI?
BASTIONHOST / AI PROJECTSAI is a possible answer.
AI PROJECTS
AI is a possible answer.
It is not the starting assumption.
When capacity allows, I offer voluntary help to organisations deciding whether an AI idea deserves to exist, then help move a real problem towards a controlled, measurable and supportable service. I am not afraid to say AI is not the solution.
AI READINESS REVIEW
Six questions before an AI project receives a green light.
A compelling demo is not a business case. This AI readiness review tests whether the idea has a useful purpose, suitable data, a responsible owner and a realistic route into day-to-day work.
A better alternative
Would a process change, standard software, analytics or simple automation solve it more safely and cheaply?
Data readiness
Is the necessary data available, lawful to use, sufficiently representative and good enough to support the decision?
Risk & responsibility
Who is accountable for outputs, errors, privacy, security and the effect on the people who will use—or be affected by—the system?
Delivery reality
Can this integrate with real workflows, earn user trust and be supported after the prototype team moves on?
Measurable value
Is there a baseline, a useful success measure and a clear point at which we will stop rather than defend a failing idea?
AI FIRST-CHECK
Ten questions. No hype. No automatic “yes”.
This is a quick way to test the shape of an idea. It looks separately at opportunity, organisational readiness and risk, then suggests what deserves attention next. It is not project approval.
A TYPICAL AI PROJECT
From honest question to useful service.
Each phase produces evidence for the next decision. Approval is not permanent: if the evidence weakens, the project changes direction or stops.
Frame the problem
Agree the outcome, users, current baseline and constraints before choosing a technology.
Challenge the AI assumption
Compare AI against simpler options. Choosing not to use AI is a valid—and sometimes best—project outcome.
Discover data and risk
Understand data quality, access, security, privacy, failure modes and who carries accountability.
Run a controlled proof
Test the riskiest assumptions with a small, time-boxed experiment—not a disguised production rollout.
Validate with real users
Measure usefulness, accuracy, safety, adoption and unintended consequences in the workflow that matters.
Design production properly
Build ownership, controls, monitoring, integration, support and rollback into the service.
Roll out, measure and review
Track the promised outcome, review drift and risk, and improve or retire the system when it no longer earns its place.
WARNING SIGNS
When enthusiasm is running ahead of evidence.
- A solution looking for a problem
- A prompt demo being treated as a product
- No accountable business owner
- No baseline for the promised improvement
- Confidential data used without clear controls
- Output quality judged by the project team alone
- Success defined only as adoption or usage
- No rollback, review or retirement plan
My position
AI will not repair a broken process, unclear ownership or poor-quality data. It often amplifies them. Fixing the foundation may be the highest-value technology decision.
BEFORE YOU COMMIT
Bring me the idea while ‘no’ is still an acceptable answer.
When capacity allows, I can provide an independent approval review, shape a responsible proof of value, or help challenge a project that has moved faster than its foundations.
Discuss an AI project →