Writing · AI Implementation
I get called into AI projects after they have failed once. The pattern is consistent enough to write down: the project was lost during scoping, months before anyone wrote code.
Large AI projects get scoped and priced at month zero, when nobody yet understands the real constraint. The vendor needs a number to close the deal. The buyer needs a number to get budget approval. Both sides commit to a scope that describes assumptions, not reality. Then delivery proceeds against those assumptions long after the operation has quietly invalidated them.
The misunderstanding does not disappear. It gets buried inside the contract and paid for in months of delivery and six figures of spend. This is not an AI-specific failure. It is a scoping failure that AI budgets have made more expensive.
The workflow being automated exists only as a story management tells. The people doing the work run a different process, with exceptions nobody scoped. The model automates the story and breaks on the reality.
ProcessNobody inside the company owns the system after handover. A prototype without an operational owner is a demo with hosting costs.
OwnershipThe expensive problem is almost never where the org chart says it is. Automating a step that was never the bottleneck produces a faster queue in front of the same wall.
ConstraintThe pilot worked on twenty clean documents. Production means thousands of messy ones, edge cases, cost controls, and evaluation. Prototype quality wrapped in enterprise pricing fails in the gap between those two.
ProductionThe fix is to discover the actual scope before pricing the work that follows. In practice that means interviewing the operators doing the work, not only management. Tracing the workflows as they actually run. Reading the systems and the numbers. Comparing the stated narrative to observable outcomes.
That is what the two-week operating diagnostic produces: a ranked constraint map, an automation and elimination register with effort estimates, and one fully scoped first fix. Sometimes the honest finding is that AI is not the answer at all. That finding gets delivered anyway, because it is cheaper than the alternative: discovering it in month five of a fixed-scope build.
If a project has already failed once, the same method works in reverse: identify the root causes, keep the parts worth keeping, and rebuild only those. A documented example of what the diagnostic-first approach produced for a law firm is in the case studies, and the specific service for legal and professional services firms is in AI automation for law firms. For investors assessing someone else's AI claims before committing capital, the equivalent work is technical and operational due diligence.
A free 30-minute conversation is enough to tell whether the problem is understood well enough to price. If it is not, two weeks of diagnosis costs less than one month of misdirected build.
Book the conversation →