Sunita, 38: production planner, twelve years in, drowning in Excel

Planned for release phase 2. The page is complete as a draft; interactive parts run in the browser and do not yet save to an account.

1. Step 1 — Define the End State

   "The daily report exists by 9 a.m. without me building it, and is right — and I use the four hours to prevent the downtime the report keeps telling me about after the fact. The plant head sees a planner who anticipates, not a clerk who reconciles." Audiences: the plant head, the three shift supervisors, and the maintenance lead who acts on what she flags.

2. Step 2 — Map the Flow

   Her list is long and precise, because she knows the plant: where each data source comes from, what breaks in each, which checks she does by eye. AI review adds one thing she had not seen: the reconciliation logic she runs in her head is itself a specification — if she writes it down, it can be automated by the IT team without any AI at all. And one thing she pushes back on: it suggests a predictive downtime model; she says the plant's data is not clean enough to predict anything yet, and adds a smaller task — cleaning the data — instead.

3. Step 3 — Assign Tasks

   Writing down the reconciliation rules: hers — context only in her head. Turning those rules into a script: AI-assisted, verifiable against a month of old reports. Spotting the anomaly that matters in today's numbers: hers, always. Drafting the plain-English morning note to supervisors: AI generates from her flagged items; she edits. Deciding what to do about a machine: hers and maintenance's.

4. Step 4 — Select the Tool

   A data-and-analytics category tool that reads her three files and applies rules, built once because it runs daily; a conversational tool only for the morning note. She does not need a chat interface for the reconciliation, and says so.

5. Step 5 — Prepare and Direct

   Criteria: the automated report must match my hand-built report for thirty consecutive days before I trust it; every discrepancy must be flagged, not silently resolved; the morning note must be under 120 words and name a machine, a shift and an action.

6. Steps 6 and 7 — Execute and Evaluate

   The script matches her report on day one — and she does not trust it, per her criterion. Day nine it diverges: a supervisor changed a sheet format. The tool had "resolved" it silently; she had written the flag-don't-resolve rule and the script had not honoured it. Fixed. Thirty days pass. The morning note drafts are fine and slightly too polite; she edits them into her own register, which supervisors read faster.

7. Step 8 — Integrate and Stress-Test

   Integration is the four hours: what does she do with them? Small fix: the note goes out at 8:40, before the supervisors' huddle, not after. Structural: her end state said "anticipate", and nothing in her flow anticipates — everything is faster reporting. A Step 1 to Step 2 trace. She adds a weekly task: look at the thirty-day pattern of flags and walk the floor to the three machines that recur. That is where the planner she wanted to be lives.

What changed

Twelve years of judgement did not get replaced. It got four hours a day to work in. The plant head forwards head office a note about "AI adoption"; Sunita's version of the story is that she wrote down what she already knew.

A composite case, not a client engagement; the person is invented, the situation is not.