AI Fit
Determine where AI belongs before you build it.
Evaluate the problem, workload, economics, architecture and risks before committing engineering resources.
The most expensive AI decision is the one nobody scored. A team picks an agent because agents are what people are building, and discovers nine months later that the answer was a database lookup with a good interface.
AI Fit takes a use case, the data behind it, the budget and the real constraints, and returns a decision you can defend: the recommended shape, what was ruled out and why, the cost drivers that actually move, and a brief you can hand to a board.
What you get
What the product actually produces.
- Recommended shape
- An architecture class — not a vendor, not a product name you will be locked into.
- Ruled-out alternatives
- What was considered and rejected, with the reason attached to each one.
- Named cost drivers
- What will actually drive spend. Never an invented price.
- Open questions
- The things the tool cannot decide for you, listed rather than glossed over.
The decisions it forces
Most AI mistakes are architecture mistakes made early.
Not model choice. The decision about what shape the system should be, taken before anyone had to defend it.
The most expensive AI decision is the one nobody scored.
- Workflow or agent?
- A deterministic workflow is cheaper, testable and easier to defend. An agent earns its complexity only when the path genuinely cannot be known in advance.
- Retrieval or a lookup?
- A surprising number of retrieval projects are a database query with a language interface on top. That is a good answer, not an embarrassing one.
- Where should this execute?
- Cloud, local or edge, decided from privacy, latency, policy and cost rather than from whichever vendor has a solutions architect on the call.
- What has to be true to afford it?
- The cost drivers that actually move the number, named — not a made-up price per user.
- Which duties already apply?
- Disclosure and content-marking obligations are in force now. That changes design, not just legal review.
How it works
Deterministic rules, not a model guessing.
You describe the use case in plain language. Published rules match that description to a catalogue of architecture shapes. Every finding quotes the sentence that triggered it, so the same input always produces the same output.
Decision path
- Use case
- Constraints
- Rule match
- Shape
- Ruled out
- Board brief
Nothing you type is stored. The recommendation is an architecture class rather than a vendor, because the vendor list changes every quarter and the shape of the problem does not.
The tool is open on the web. See how the rules are published.
Limits
What this is not.
Stated on the page rather than discovered in month three.
- It is deliberately not a vendor recommendation engine.
- It will not invent pricing it cannot source.
- It answers the architecture question, not the organisational one — that is AI Fit Teams.
Got the architecture, unsure about the organization?
AI Fit answers the technical question. AI Fit Teams answers whether your teams, workflows and governance can carry it. Thirty minutes is enough to work out which one you need.