AI Strategy
When Not to Use AI: The Discipline of Subtraction
The most valuable AI recommendation is often “don’t.” Here is how to make that call credibly.
A consultant who recommends AI for every problem is not consulting — they are selling. The discipline that matters most in this field is subtraction: identifying the processes that should not be automated, the models that are unnecessary, and the projects that should be cancelled before the second invoice.
Do not automate low-volume judgement work
A task performed 20 times a month by a senior person whose judgement is the point is a poor automation candidate. The engineering cost will never amortise, and the system will be worse than the person it replaces. Leave it alone.
Do not add AI to a broken process
Automating a broken process makes it break faster. If the workflow has 12 handoffs, three rework loops, and no clear owner, fix the workflow first. AI on top of chaos produces faster chaos, and the AI gets the blame for a problem it inherited.
Do not use a model where a rule works
If the decision is “amount over $10,000 requires manager approval,” that is a rule. Expressing it as a model adds cost, opacity, and a new failure mode. The temptation to use AI because it is available is real; resist it where determinism suffices.
Saying no credibly is what earns the right to say yes. A track record of killed projects and avoided automations is the most persuasive evidence that the ones you do recommend are worth doing.
