The typical phases
A useful way to think about timing is by phase, not by a single number. Each phase has a clear exit, so you always know where you stand and can stop if the results don't justify going further.
Most projects run these phases in sequence, though discovery and a proof of concept can overlap on a small scope.
- Discovery and scoping: aligning on the use case, the success metric, and where the data lives
- Proof of concept: a narrow, throwaway build that proves the approach works on your real data
- Pilot: a limited release to a small group, with monitoring and a rollback path
- Production rollout: hardening, access controls, error handling, and wider release
- Handoff: documentation, monitoring, and a plan for ongoing tuning
What speeds it up or slows it down
Two teams asking for "the same" AI feature can land months apart. The difference is almost never the model itself; it's the environment around it.
- Clean, accessible data versus data spread across systems with no API
- A single approver versus multi-department sign-off, legal, and security review
- One integration point versus syncing several tools that each have their own rules
- Clear acceptance criteria versus "we'll know it when we see it"
- Whether a human must review output before it reaches a customer
- Compliance requirements (healthcare, finance) that add validation steps
How we keep it honest and on schedule
We plan in short, visible increments so you see working software early and can redirect before the budget is spent. Timelines in a proposal are estimates tied to assumptions we write down; when an assumption changes, we tell you what it does to the schedule.
- A written scope with a named success metric before any code
- Weekly check-ins with a working build, not status slides
- A go/no-go decision after the proof of concept, before larger spend
- Early testing on your real data instead of a polished demo dataset
More on ai integrations
Frequently asked questions
Can you give me an exact number of weeks up front?
Not honestly, until we've looked at your data and the systems involved. We can give a phase-by-phase estimate quickly after a short discovery call, and we tie every estimate to written assumptions so you can see exactly what would change it.
What's the fastest thing you can ship?
Usually a narrow proof of concept on a single, well-defined task with data we can access. It's meant to prove the approach and inform the real estimate, not to be the finished product.
Do we have to commit to the whole project at once?
No. We structure the work so there's a natural stopping point after the proof of concept and after the pilot. You decide to continue based on results, not a signed commitment to the full build.