1. Define the operating constraint

Illustrative scenario: an approved order is captured in a CRM and manually entered into an accounting system. Finance needs to know whether an invoice exists and which exceptions need review.

Scoped question: how should one approved order become one accounting record, with a controlled recovery path? This example excludes warehouse, payment and inventory workflows.

2. Record the system and responsibility view

Illustrative system boundary
StageOwnerRequired decision
Approved order in CRMSales/process ownerWhat counts as approved and which identifier stays stable?
Exchange and exception handlingIntegration ownerHow is progress recorded and an interruption recovered?
Accounting recordFinance ownerWhat is validated and how is a duplicate detected?

3. Make risk actionable

Illustrative risk register
RiskConsequenceResponse to evaluateOwner
Repeated delivery of an orderA duplicate financial recordStable external identifier and destination duplicate protectionIntegration + finance
A rejected record is invisibleWork remains unfinishedVisible exception queue, alert and review procedureProcess owner
Interface ownership is unclearRecovery depends on one personNamed operating responsibility and handover materialBusiness sponsor

4. Compare options and record a decision

Option A: use a supported connector if it meets the required identifiers, validation and recovery behaviour. Option B: build a scoped integration if the connector cannot satisfy those controls.

Provisional decision: investigate the supported interface and connector before commissioning custom work. Record the interface/version evidence, gaps, support responsibilities and conditions that would change the recommendation. No vendor compatibility is assumed here.

5. Define acceptance and measurement

These are proposed checks for the fictional scenario, not reported JVS project results.

6. Example AI evaluation worksheet

For a separate fictional document-extraction task, establish acceptance before evaluating a model. Never infer a production accuracy figure from this example.

Synthetic test conditionExpected behaviourReview boundary
Required field is absentReturn an explicit missing-field resultHuman review; no invented value
Input conflicts with itselfFlag the conflict and show the source referencesHold any downstream action
Output has an invalid formatReject it before system entryRecord the failure for investigation
Provider is unavailableUse the agreed fallback or review queueNo silent completion

What is agreed before paid work begins

The introductory conversation defines the question, available evidence, participants and constraints. A written scope then states deliverables, exclusions, workshops, analysis period, briefing date and fixed fee. Implementation is a separate commitment.

Explore the five assessment offers →

Request a 30-minute introductory conversation →

Written by Jashin Sookdhoo · Reviewed 13 October 2026 · Illustrative sample