Rollout, handover and writing · Rollout

Fill the RACI

Lay out the post-handover RACI for the claims pilot with one accountable person per row, and settle a security lead's request and a sponsor's suggestion.

designeasyExercise · 3 min · 55 of 64

On this page

The grid you fill in week 7 is the one that runs when you are gone. Two people have already proposed cells, and both proposals sound reasonable.

The situation

Week 7 of the 8-week claims-triage pilot. Handover is Friday of week 8, and the grid you are filling runs from week 9, without you. Five people go in the columns: the VP of Claims, the sponsor; the deputy, acting Head of Claims Operations since the champion left in week 4; the claims QA lead, named as owner in the definition of done; the IT platform on-call lead, whose rota of five has never seen the prompt; and the information-security lead. Six decisions need a row: approve a prompt change; re-index the policy wording when an endorsement is issued, about twice a month; rotate the model API key, every 90 days; roll a segment back to shadow; answer an alert outside working hours; and decide in phase 2 whether small motor claims move to run. The security lead asks to be consulted on every prompt change; her review queue runs two weeks. The VP suggests your company's support team be accountable for the prompt, "since they built it".

The question

Lay out the six-by-five grid with exactly one A per row, and decide what you do with the security lead's request and with the VP's suggestion.

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Decision	VP	Claims ops	QA lead	IT on-call	Security
Approve a prompt change		I	A, R	I	C on fields
Re-index the wording		I	A	R
Rotate the model key			I	R	A
Roll a segment back	I	A, R	I	I
Out-of-hours alert		I		A, R
Small motor to run	A	C	R	I	C

One A per row, every A at the customer, and the alert goes to the rota, not a person. IT runs the re-index; the QA lead answers for it by re-running the eval. Without her A, the index changes and nobody reads the result, the drift the RACI lesson traces.

Security's request is narrowed, not refused. She is consulted on any change to which fields enter the prompt, because that is what her threat model covers, and informed of wording changes. Consulted on every change through a two-week queue, the prompt runs two weeks behind each wording change it has to follow, and the wording takes effect on its date regardless.

The VP's suggestion is declined. An accountable party outside the building cannot approve a change at the customer's pace or answer the customer's auditor, and the row empties the day the support contract lapses. Our support team can be C, by contract, for 30 days.

The QA lead now holds two of the six As, more than anyone, so her hours go in the handover doc: about four a month, approved by the VP. The run decision is the VP's, because it removes a person from the claim, and security is consulted because an output that acts is a new line on her threat model.

Rubric

puts exactly one A in each row, and every A on a person at the customer
sends the out-of-hours alert to the rota, not to a named engineer or the vendor
narrows security's consultation to changes in what reaches the prompt, with the queue arithmetic
declines the vendor as accountable and offers a bounded, dated C instead
costs the QA lead's hours and puts them where the VP can approve them
gives the run decision to the sponsor and consults security on it

Where answers go wrong

"The QA lead and claims operations jointly accountable for prompt changes." Loses because two As let each wait for the other; the first change after handover sits until someone decides it is theirs.

"Vendor support accountable for the prompt, since they built it." Loses because the row empties when the contract lapses, and when the customer's auditor asks who approved a change, nobody at the customer can answer.

"Security consulted on everything; safer that way." Loses because a two-week review on every prompt change leaves the prompt two weeks behind each wording change it follows, a quality failure nobody consented to and nobody measures.