Decision definition
A named user, workflow and consequence of error are more useful than a list of desired model capabilities.
Custom AI planning
Estimate a custom AI build from evidence, interfaces and operating risk
Start with what the system must decide, what evidence exists, what has to act on the result and who owns it when the project team leaves.
What the estimate measures
Production conditions, not model fashion
The public result is driven by evidence that can be inspected before a project is signed.
Evidence burden
The data and held-out outcomes needed to prove the decision determine whether the work starts with feasibility or build.
System surface
Identity, interfaces, permissions, latency and deployment boundaries are priced as production engineering.
Ownership
Release gates, observability, fallbacks, runbooks and change authority are deliverables rather than future chores.
Worked examples
How the inputs move the starting point
These examples explain the rules. They are not invented client results.
| Situation | Likely first path | Why |
|---|---|---|
| Broad opportunity list, no representative data or owner | Feasibility Study · from $9,500 | The project needs one decision and evidence that the available data can support it. |
| Named workflow, prototype, two APIs, numeric metrics but no held-out set | AI Pilot · $28K–60K | One real slice can establish the evaluation, integration and production architecture. |
| Existing operation, reviewed data, specified interfaces and release gate | Production Build · $75K–180K | The principal unknowns are resolved and the work can be phased against signed acceptance. |
Methodology v1.0.0
A versioned rule, open to inspection
The score has five dimensions: problem definition 20, data readiness 25, integration readiness 20, acceptance readiness 20 and operational ownership 15.
- 01The shared scoring model makes different modalities comparable at the project level without pretending their model work is identical.
- 02The recommendation is chosen from Pran’s public engagement bands. It never multiplies a day rate or produces a precise quote from questionnaire answers.
- 03Consequential decisions, restricted deployment and broad integration can move a ready project into the enterprise band because validation and commissioning expand.
Limitations
What this result cannot establish
- The estimate is a planning band derived from the answers supplied. It is not a quote, warranty, or substitute for reviewing real data and interfaces.
- Third-party licences, cloud or hardware consumption, travel, and unusual certification work are outside the build band unless a proposal explicitly includes them.
- No model, accuracy, savings, or return claim is inferred from a questionnaire. Those require a feasibility test or measured baseline.
Questions
Before you use the range
It scores the project conditions common to production work: outcome, data, interfaces, evaluation and ownership. Track-specific questions then shape the risks and acceptance criteria.
No. It selects one published planning band and explains why. A quote follows review of representative data, interfaces and the target environment.
That is a useful outcome. It means a specific unknown can invalidate a larger build and should be answered for a known smaller cost before committing.