The weekly update and the field note
Write the one-page weekly update, the memo a sponsor forwards upward unedited, and the field note that changes your own product, from the week-4 numbers.
On this page
Three one-page documents leave a deployment every week or so, and each is judged by what its reader does with it. The weekly update passes if the champion forwards it to the sponsor unedited, the status memo if the sponsor forwards it to his own boss unedited, and the field note if it changes your product. All three are short narratives with a number in the first sentence.
The narrative form, at one page
Colin Bryar and Bill Carr's Working Backwards describes the narrative memo that replaced slides in senior meetings: full sentences, read in silence, so the discussion starts from the same facts. A bullet reading "commercial: retrieval issues" lets the writer skip the cause; a sentence has to say that the governing clause is in an endorsement the index does not hold, and the reader can judge the fix.
The weekly update carries five things, in order:
- The number, in the reader's units, with what it was measured on.
- The delta since the last update, and its cause.
- The decision needed, one, stated so it can be answered yes or no.
- From whom, a person.
- By when, a date and time, and what slips if it is later.
Everything else supports those five or is cut. It goes out every Friday, good week or not, because, as champion, blocker, executive sponsor says, a missed update is read as a bad week.
The mistake: the activity update
The update most engineers write on a Friday afternoon, in week 4 of the claims pilot:
Week 4. This week we finished labelling, started the shadow run, met the security team on the field list, onboarded the deputy and started the endorsement analysis. Next week: shadow at full volume, commercial fixes, adjudication. Risks: champion departure.
Every sentence is true, and the sponsor can act on none of them. With no number, he cannot tell if the week was good. "Shadow at full volume" reads as a plan, when it needs two hours a day of someone's time that only he can give, and nobody has asked. "Risks: champion departure" names the event without its consequence or its handling.
Week 4, written the day the champion left
Friday of week 4: the champion leaves today, the VP is away until Monday, and the shadow run started on Wednesday at 100 claims a day, ahead of the week-6 gate that crawl, walk, run sizes. The update:
Claims-triage pilot, week 4 of 8.
The tool makes the same call as the handlers' adjudicated labels, escalate to a senior handler or approve, on 91% of the 140 gold cases settled so far, against the 95% line.
Since last Friday the last 60 cases were labelled, so the 200-case gold set is complete, and the shadow run started on live claims. Six of the 13 misses have one cause: on commercial policies the governing clause is in an endorsement, which the index does not hold as a separate document. The fix is built next week.
Decision needed: two hours a day of the claims QA lead's time in weeks 5 and 6, to rule on shadow disagreements, so the run goes to 500 claims a day on Monday. From the VP of Claims, by Monday 09:00. Each day later leaves the week-6 gate 100 commercial claims short of the 1,000 it needs.
Ownership. The Head of Claims Operations leaves today. Her deputy acts from Monday; he ran today's Friday meeting with her in the room, his name is beside hers on the definition of done, and the last 8 labelling disagreements are his to settle by Wednesday.
Week 8 holds. It moves only if the deputy cannot present the week-6 reading, in which case the result moves to week 10, still before the budget closes in week 11.
Read it against the five: a number the VP signed, in his words; a delta with a cause; one decision with a person, a time and the cost of lateness. The departure is there with its handling done, because a sponsor who hears it elsewhere reads every later page as the version that leaves things out, the failure the champion-left case turns on.
Two numbers are absent on purpose: the shadow run's first 300 claims, whose 60 commercial ones carry a band of at least ±6 points, and about ±12 if commercial reads near 70%, and cost per claim, which is inside its line and asks nothing of the reader.
The status memo the sponsor forwards unedited
On Monday of week 7, after the per-segment gate, the VP has to tell the COO something, and left to him it will be "91%". The memo is written for him to forward, in the COO's units:
The claims-triage pilot passed its live checkpoint on Friday. Running silently beside the handlers on 5,000 live claims, it made the same call as they did on whether a claim needs a senior handler for 96% of motor and property claims, which are 80% of the floor's volume, and from Tuesday two teams use it on those claims. Commercial claims, at 70%, stay out until a fix is checked at the end of week 8. The target, triage 30% faster on motor and property claims, is measured in week 8 on the two teams using it against the two that are not, and is reported on schedule; all four teams use it from week 9.
Read it as the COO, new to the pilot. "Shadow", "walk", "adjudicated" and "segment" are gone; the 96% says what it is a percentage of; the commercial number is there with its date, so it cannot surface later as news. A memo the VP has to edit will be edited, and the edit is usually the commercial sentence.
The field note back to product
The field note's reader is a PM deciding what the product does for every customer, the reader working with an FDE and with research describes. Three parts: the failure, the data, and what generalises. Written at the end of week 6, from the QA lead's rulings:
Failure. On commercial claims the tool surfaces the base policy's clause when the governing wording is in an endorsement. Data. The QA lead ruled the tool wrong on 300 of 1,000 commercial shadow claims, and put 210 of the 300 on this cause. The customer's policy system stores each endorsement as a separate document linked to its policy. Our ingestion appends endorsements to the base text, so retrieval ranks the base clause above the endorsement that overrides it. What generalises. Any customer whose policy system stores endorsements separately will hit this on about one commercial claim in five (210 of 1,000 here); ask in discovery. Motor and property carry few endorsements, so the 70% does not generalise. We built per-endorsement indexing with an effective date in four days; as the default, it would remove the failure before a pilot starts.
To a PM who has never met this customer, "retrieval quality issues on commercial" earns a nod; a cause, a count and a mechanism to check elsewhere earns a ticket.
In an interview
The behavioural round asks for a time a stakeholder left, and the strong answer quotes the document: "that afternoon's update led with the agreement number, asked the VP for one decision by Monday 09:00, and had the departure and its handling in it." The case round asks what you would send the sponsor, and the interviewer listens for the number, one decision, a name and a date.
The mistake that loses points is an update with no decision in it. An update that asks for nothing trains the sponsor to stop reading, and the week you need a decision, nobody reads the page it is on.
Check yourself
1. Week 5. The shadow run is at 500 a day, the gate is next Friday, and nothing needs deciding. Do you send an update?
Yes, and the first line says so: "no decision needed this week". A skipped update is read as a bad week; a no-decision update says the plan is holding and names the next decision, walk by segment, due Monday 09:00 of week 7. If anything in the next fortnight depends on the sponsor, this update asks for it now.
2. The VP asks for the update fortnightly, in weeks 3, 5 and 7, to save your Friday morning. Estimate what that would have cost this pilot and decide.
Two things. The departure would have reached him from someone else, with no week-4 update to carry it. And the week-6 gate, read on a Friday, would have waited for week 7's update, so walk starts on Tuesday of week 8 and gives the week-8 time reading 3 working days of data instead of 8. The saving: three Friday mornings, weeks 4, 6 and 8, at about two hours each, six hours in all. Keep it weekly; offer five lines on a week with no decision.