The deal from the engineer's seat · Stakeholders

The feature request that adds six weeks

Choose which kind of no to give a sponsor who wants decline letters drafted by week 8, cost it in reviewer-hours, and name what would make it a yes.

casehardExercise · 4 min · 18 of 64

On this page

The sponsor's request is reasonable, visible, and six weeks long. The answer has to keep him, keep the date, and be given in the room.

The situation

Week 3 of 8. In the Friday meeting the VP of Claims asks whether the pilot can also draft the letter sent to the claimant when cover is declined after triage: about 15% of 1,200 claims a day, roughly 180 letters. The letters go out under the insurer's name and are regulated; a wrong one is a complaint to the ombudsman. Today a handler writes each from one of nine templates in about six minutes. Costed in your head: one week to agree the templates and the prohibited phrasings with the legal team; two weeks to build a rubric-graded eval of 100 letters, each graded by two reviewers in about four minutes, since free text has no exact match; one to two weeks of prompt iteration; one week assisted, with a handler editing every letter. Six weeks of calendar, and the two senior handlers who would grade them are labelling the triage gold set, which stands at 140 of 200. The VP adds that the board sees a demo in week 8 and "the letter is the bit they'll understand". Agreement on the triage eval is 90% against the 95% line. Sales is not in the room.

The question

Which of the three kinds of no do you give, what do you say to the VP in the room, and what would have to be true for the answer to be yes?

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Not yet, with a date, because the request moves the week-8 date and needs the two reviewers the triage eval is waiting on. In the room: "Yes, and not in this eight weeks. The letter is about six weeks: a week with legal on the templates, two weeks for a 100-letter eval your reviewers grade by hand, two weeks of iteration, a week assisted. It needs the two reviewers labelling the triage set, which is at 140 of 200, so starting now moves the triage number from week 8 to week 14. For the board in week 8 I'd show the triage number against the 30% you promised the COO, and a letter drafted live on one real declined claim with the handler's edits visible, as a preview of phase 2. I'll send you the plan with both dates tomorrow."

The reviewer cost: 100 letters × 2 × 4 minutes ≈ 13 hours, from the two people the triage set is waiting on; the preview is real data, never a polished sample. The page carries the six-week breakdown and the choice: week 8 with phase 2 from week 9, or week 14 with both. Sales gets it the same day as the VP.

It becomes a yes if the request is the deal: if the VP says the COO signs only with the letter in scope. Then the answer is a re-signed scoping doc with the week-14 date, the triage volume gate cut from 90% to 80% to free reviewer time, a legal owner named for the templates, and sales told the same day. A request with the signer's name on it is a scope change; one from a Friday meeting is phase 2, as saying no sets out.

Rubric

picks "not yet" and gives the reason in terms of the date and the shared reviewers
costs the request in reviewer-hours and calendar weeks, with the breakdown
protects the number the sponsor promised upward, and says so in the room
offers the board something real for week 8 without a polished sample
sends the page with both dates to the sponsor and to sales the next day
states the condition that turns the no into a yes, and what the yes then costs

Where answers go wrong

"Sure, we can add letter drafting; the model is good at text." Loses because the eval for free text does not exist and the reviewers who would build it are already the bottleneck; week 8 arrives with neither number.

"No, it's a regulatory risk; we shouldn't draft letters." Loses because it is a refusal without a date or an alternative, and the sponsor hears "not interested"; the risk is real and is handled by the assisted week with a handler editing every letter, which the answer should have named.

"Do a quick version on three sample claims for the board demo." Loses because a demo on chosen samples is the demo the prototype lesson warns against, and the board's question in week 9 is why the live letters look different.