Start with what you know.
Answer all three questions. Unknowns are useful scoping information, not a failed score.
Applied AI / Implementation decisions
A good prototype.
Now make it dependable.
AI implementation connects a model to a real job, the right data and the systems around it. Production adds the evidence, controls and ownership that let people rely on the result.
Check your starting pointThree scoping questions. No score, email gate or data upload.
01 / Choose the kind of help
Advice, implementation and operation
are different responsibilities.
- Consult
- Choose the job, test the business case and define the boundaries. Useful when the team needs a decision before it needs a build.
- Build & integrate
- Connect approved sources and tools, implement the workflow and test the result against agreed examples. A demo is not acceptance.
- Deploy & operate
- Release into an agreed environment. Name the owner, monitor failures, manage changes and agree ongoing support. Maintenance is a scope, not an assumption.
02 / An honest starting point
What needs
to be true?
This check helps prepare a scoping conversation. It cannot assess your security, economics or technical feasibility. Your selections stay on this page and are not submitted.
Download the editable workflow brief ↗03 / Evidence before release
Production is a set of obligations.
Not a more impressive prompt.
An acceptance set
Representative inputs, expected behavior, unacceptable errors and known edge cases. Retain the source for review; do not judge the system only by its best answer.
A bounded tool path
Read and write permissions, validation, human approval where needed and safe handling of repeated requests. Prefer ordinary rules where the decision is predictable.
A failure route
Timeouts, unavailable sources, ambiguous output and uncertain external actions need explicit behavior. An unknown send or write is not a reason to repeat it blindly.
A usable operating record
Track failures, model and system changes, review decisions and actual usage costs. Define support, retention, recovery and who can stop the workflow.
Build, buy
or keep it simple?
Buy an existing tool when its workflow, permissions and operating limits fit the job. Evaluate integration and exit costs, not just the demo.
Build a scoped system when the business process or integration gap is material enough to justify ownership. Custom work creates a maintenance responsibility.
Use ordinary automation when inputs and rules are predictable. AI is useful where interpretation matters; it does not make a fragile process sound.
Pause the build when the task, data permission or owner is unresolved. A short discovery effort may be the useful deliverable.
04 / Inspect what the proof establishes
A capability example
is not a client outcome.
Our public build examples identify internal implementations and illustrative demonstrations. Use them to examine interfaces, integrations and review boundaries, not to infer client deployment or a financial return.
Bring one job worth improving.
Talk through the work, available systems, permissions and operating owner. We can define whether the useful next step is discovery, an implementation or something simpler.
Discuss an implementationExplore Applied AI