Choosing the first useful customer workflow

This worked application uses explicit planning assumptions to demonstrate FIELD. Quantities below are inputs, calculated implications or proposed targets; they are not reported customer results. The engagement decision remains conditional on the evidence described.

Planning basis and engagement boundary

F — Frame the outcome

Select one useful workflow for 20 pilot users in four weeks; the sponsor decides whether evidence supports expanding beyond the pilot.

Value measure: Net human minutes per eligible completed task, attempted-use rate and ongoing support cost.

I — Investigate the workflow

Observe document lookup, invoice exceptions and contract intake. Compare task frequency, baseline effort, error consequences and access feasibility before selecting a candidate.

First evidence task: Observe representative tasks and time verification, not just initial answer generation.

E — Engineer the solution

Use the smallest complete authorized workflow and its resource budget. Start with retrieval/search where that tests value sooner than autonomous actions.

The implementation assessment below makes the authority, workload, resource and failure assumptions concrete. Its numbers are calculations from the stated scenario, not a vendor performance guarantee.

Frame the problem

Talk with the 20 pilot users and observe actual cases across the three proposed workflows. Score frequency, baseline effort, consequence of error, data access and feasibility. For a worked choice, select document lookup if permissions/source access are ready and invoice/contract workflows have unresolved action authority. Narrow to answering policy questions with citations; leave business changes outside the pilot.

Implementation plan

Source connector→versioned parsing/indexing→authenticated retrieval→optional reranking→one grounded model call→citation/answerability checks→user response and feedback. Start with lexical search as a baseline. Store tenant/user identity, document versions, queries in a privacy-appropriate form, retrieved references and feedback. Human escalation handles missing/conflicting evidence.

Sizing

Assume 20 users×20 questions/day=400/day, or 0.0139/s over an eight-hour day. A chosen 2/s test peak intentionally probes burst capacity rather than deriving peak from average. At 2,000 input/300 output tokens per question, daily demand is 800k input and 120k output tokens. A 4-logical-CPU/8GiB application host is an initial pilot configuration, not a proven capacity choice; corpus/index requirements are separately measured. Scaling to 2,000 users under the same behavior multiplies daily work by 100, but peak/skew still need observation.

Acceptance and delivery

Week 1: workflow/baseline and permissions; week 2: working search/RAG; week 3: representative evaluation and error repair; week 4: controlled user pilot and handover. Measure verified task correctness, citation support, net handling time including review, repeated usage and cost/success. If manual lookup takes eight minutes and the new workflow takes five including verification, hypothetical saving is three minutes/task; do not count those savings before observation.

Changed conditions

Refuse to remove a necessary approval merely to improve projected ROI. Offer a narrower class of reversible low-risk tasks with independently enforced checks, or retain review and report its cost. User distrust prompts evidence/UI/workflow investigation, not a marketing claim.

L — Launch with evidence

Compare verified task completion and total human effort against the current workflow; include abandoned tasks. Twenty users establish pilot evidence, not company-wide impact.

Critical decision condition: No expansion without an agreed outcome denominator and permitted data path.

Attach raw artifacts, the tested configuration, population, denominator, date and limitations to the release record. Missing or stale evidence holds the dependent decision. A favorable value forecast cannot offset a failed critical condition.

D — Develop ownership and reuse

Have the workflow lead own ongoing measurement and an operator own support. Expand only after the selected task remains useful without daily delivery engineer intervention.

Candidate reusable asset: Outcome-contract template and one governed connector.

Commercial decision and next commitment

Accountable decision owner: Business sponsor with workflow and data owners.

The initial commitment is the bounded journey above, with the stated exclusions. Confirm access, customer staffing and acceptance ownership during discovery; price and schedule unresolved dependencies explicitly. A new source, tenant class, language, action or operating envelope requires an impact review rather than silent scope growth.

Approve only the next exposure level supported by the evidence. The next useful investment is: observe representative tasks and time verification, not just initial answer generation.

Method and references

Apply the FIELD method, economic model and evidence rules. Methodological and technical influences are listed in the source register.