An honest AI readiness assessment: whether your data can support the use cases you have in mind, which ones are worth building, and what to fix first.
AI projects fail on data, not models
The failure mode is almost never the model. It is that the data the use case depends on is incomplete, inconsistently labelled, locked in a system with no API, or simply not collected. Teams discover this three months into a build, after the budget is committed and expectations are set. The second most common failure is choosing a use case that is technically achievable and commercially pointless, because it was picked from a vendor deck rather than from where the cost actually sits.
What the appraisal covers
- A structured inventory of candidate use cases, including ones your team has not considered
- Data feasibility per use case: what it needs, what exists, and what the gap costs to close
- Value estimation grounded in current process cost, so the ranking is commercial rather than technical
- A build, buy, or wait recommendation per use case, with reasoning
- Integration and access assessment, since many blockers are systems rather than data
- A sequenced plan starting with the use case most likely to succeed and prove value
How we work
Workshop candidate use cases with the people who own the affected processes
Profile the underlying data for completeness, consistency, and accessibility
Score each use case on feasibility and value, and be explicit about what fails
Deliver the ranked plan with a recommended starting point and what to fix before the rest
Typical stack
Frequently asked questions
When that is the answer, yes, and it has happened. The most common version is not "never" but "not this use case yet": fix the data dependency first, or start with a narrower version that will actually work. An appraisal that approves everything is worthless as a decision input.
It depends entirely on the use case, which is the point of appraising rather than generalizing. A retrieval assistant over documentation tolerates messy data well. A forecasting model needs consistent history. An agent acting on budgets needs data that is correct and current, because a wrong action costs real money.
Yes, and quite often they win. For plenty of use cases a mature product is cheaper and better than anything custom. We flag those explicitly so you do not commission a build that a licence would solve.
Typically two to three weeks: a workshop round, data profiling, then findings and a review session. Faster if the candidate use cases are already narrow and well defined.
Go deeper
Get an AI readiness read
Tell us the use cases you are considering and we will assess whether your data can support them.
Start a project