AI Operating Model Diagnostic Kit

One workflow, answered and closed out

What one AI-enabled workflow looks like once it has been answered.

A customer service team — the kit's own fictional Northstar Service Team — has been drafting inquiry replies with a generative AI assistant. Response times improved. Nobody can currently say what a reply has to satisfy before it is sent, or who is called when one turns out to be wrong.

This example is fictional. It does not describe a client, an engagement, or a real organization. The questions, dimensions, watch signals, and template fields are taken from the kit as shipped; the answers, the priorities, and the actions are invented. The team and its six answers are the same ones the kit's own bundled example uses, so what is shown here and what is delivered do not disagree.

1. 01

   Situation

   Adoption succeeded, and the operating questions were never asked.

   Nothing has gone wrong. The team is at the point where the assistant is load-bearing for a customer-facing workflow, and the arrangements around it — acceptance, escalation, records, review capacity — were never decided, only inherited from how one person happened to work. The kit is scoped for exactly this: one workflow, not an enterprise rating.

   • Inquiry replies are drafted with AI and sent by the same person who drafted them.
   • The department head approves everything, by custom rather than by rule.
   • There is no stated condition a draft must meet before it goes to a customer.
   • One person, 60–90 minutes; a group of three to six comparing views, 120–180 minutes.
2. 02

   The work

   Answer the questionnaire against one workflow, in four values.

   The workbook ships with the question, its dimension, and a watch signal already written; the Answer column is empty. Answers are Yes, Partly, No, or Unknown, and Unknown is a first-class answer rather than a failure to complete the sheet — the watch signal beside each question is what tells the reader whether a comfortable Yes is actually a Yes.

   Question IDIn the kit	DimensionIn the kit	QuestionIn the kit	AnswerYou fill in	Watch signalIn the kit
   Q01	D1 Purpose and priorities	Are the outcome this AI use should improve and the things that can wait shared?	Yes	Using AI is the purpose
   Q04	D2 Decisions and accountability	Are decision-makers clear for starting use, changing scope, and stopping use?	Partly	Someone approves
   Q08	D3 Delegation and human review	Are acceptance conditions and reviewers defined for AI output?	No	Output is used as-is
   Q10	D4 Exceptions and handover	For unexpected output, misuse, or an unanswered case, are the contact and initial response known?	Unknown	Reliance on individual goodwill
   Q13	D5 Records and visibility	Are important decisions, conditions of use, and review results retained so they can be explained later?	Partly	Information is scattered across chats or devices
   Q16	D6 Capacity and review load	Are the time and expertise required for review and approval estimated?	No	Every item receives the same depth of review

   Question ID· in the kit

   Q01

   Dimension· in the kit

   D1 Purpose and priorities

   Question· in the kit

   Are the outcome this AI use should improve and the things that can wait shared?

   Answer· you fill in

   Yes

   Watch signal· in the kit

   Using AI is the purpose

   Question ID· in the kit

   Q04

   Dimension· in the kit

   D2 Decisions and accountability

   Question· in the kit

   Are decision-makers clear for starting use, changing scope, and stopping use?

   Answer· you fill in

   Partly

   Watch signal· in the kit

   Someone approves

   Question ID· in the kit

   Q08

   Dimension· in the kit

   D3 Delegation and human review

   Question· in the kit

   Are acceptance conditions and reviewers defined for AI output?

   Answer· you fill in

   No

   Watch signal· in the kit

   Output is used as-is

   Question ID· in the kit

   Q10

   Dimension· in the kit

   D4 Exceptions and handover

   Question· in the kit

   For unexpected output, misuse, or an unanswered case, are the contact and initial response known?

   Answer· you fill in

   Unknown

   Watch signal· in the kit

   Reliance on individual goodwill

   Question ID· in the kit

   Q13

   Dimension· in the kit

   D5 Records and visibility

   Question· in the kit

   Are important decisions, conditions of use, and review results retained so they can be explained later?

   Answer· you fill in

   Partly

   Watch signal· in the kit

   Information is scattered across chats or devices

   Question ID· in the kit

   Q16

   Dimension· in the kit

   D6 Capacity and review load

   Question· in the kit

   Are the time and expertise required for review and approval estimated?

   Answer· you fill in

   No

   Watch signal· in the kit

   Every item receives the same depth of review

   6 of 20 questions. Answer ships empty; the rest of each row arrives written. Two further columns, Evidence or example and Notes, are also empty on delivery and are not shown here.

3. 03

   What becomes visible

   The Yes on purpose is what makes the No on acceptance expensive.

   Read as a set rather than a list, the six answers describe one shape: use is ahead of the arrangements around it. The guide names that pattern and names the five questions that, answered No or Unknown, are candidates for an accountable-owner decision before use expands. Four of this team's six are on that list.

   • Q01 is a genuine Yes — the workflow is bounded and the outcome is agreed. That is the strength, and it is why the rest matters.
   • Q08 and Q16 are the same fact seen twice: with no acceptance condition, the only available control is reviewing everything, which is why the department head is the bottleneck.
   • Q10 stays Unknown. Nobody in the room could name the contact for an incorrect reply, and that is recorded rather than guessed.
   • The guide's archetype for this pattern is Use leads; operations lag — a description of the situation, not a grade for the organization.
4. 04

   Result and next decision

   Three priorities, each with a named decision owner and a date.

   The Result Summary is where the pass ends. It caps priorities at three by design — the field is labelled that way in the shipped template — and it carries a 30/60/90-day view in which every row has a decision-maker and a stated definition of done. What the team can now take to their own leadership is not a rating; it is a short list with names on it.

   • The department head is no longer reviewing every reply because there is no other control — that was the finding, and it now has an owner and a date.

   Output

   Result Summary Template

   Record the scope and the answers, then close with strengths, critical gaps, at most three priority improvements, the issues that need an accountable owner, and a dated 30/60/90-day view.

   Workflow: customer inquiry reply drafting   Respondents: service leader, two agents, operating owner
   Strengths: purpose and workflow are bounded; the team agrees what the AI is for
   Critical gaps: output acceptance, initial exception response, review capacity
   Result archetype: Use leads; operations lag
   Priority improvements (maximum three)
     1. Acceptance criteria and a return path for inquiry drafts
     2. A named contact and first response for an incorrect reply
     3. One agreed location for decisions and review results
   Issues needing an accountable-owner decision: which inquiries may use AI support, and who can stop use
   30 days — Differentiate acceptance conditions and reviewers by impact
     Decision-maker: service leader   Done when: an agent can say, unaided, what a reply must satisfy
   60 days — Define the contact, stop, correction, and recording sequence
     Decision-maker: service leader   Done when: the sequence has been walked once against a real past case
   90 days — Agree the minimum record location for decisions and review results
     Decision-maker: operating owner   Done when: last quarter's decisions can be found by someone who was not there

Six of the 20 questions are shown, and one filled Result Summary. Withheld: the other 14 questions, the four maturity levels and four result archetypes the guide uses to read the answers, and the Improvement Action template that carries each priority on its own sheet. This page is meant to show that the diagnosis closes, not to be the diagnosis.