Chapter 15
The Answer for Tom
At 7.46 on Wednesday morning, Maya deleted the title AI Strategy from the first page of her board paper.
She typed it again.
Tom had asked the question months earlier, at 9.07 on a Monday morning after an industry dinner. “What’s our AI strategy?” Maya had not been able to answer it then. The organisation had tools in use that nobody had yet verified, a vendor demonstration booked, and a service-credit process whose conflicting procedures were already making ordinary work difficult.
Now she had an answer. She was having trouble fitting it beneath the title.
Daniel had sent her a list of available capabilities. The CFO had sent the cost view. Nadia had sent the legal scan. Leila had sent the service-credit renewal and the handover to the incoming customer operations lead. There was also a register of uses, proposed uses and uses that remained closed.
Maya arranged the papers on the table.
She did not want to tell the board that every promising task needed an assistant. She did not want to present the service-credit retrieval feature as proof that the organisation should automate the next customer decision. She needed to say where AI could help, what the company had learned about operating it, and which work deserved authority and resources next.
She wrote one sentence.
We will use AI where it improves a defined piece of work under an authority we can explain, support, review and withdraw.
It was a beginning. Tom would ask what they intended to do.
The answer they could have given earlier
Maya called Daniel before the board paper was due.
“If we had written this after the demonstration, what would it have said?”
“Choose a platform. Set up pilots. Train staff.”
“Would it have been wrong?”
“Some of it might have been useful.”
That was fair. The demonstration had revealed a real possibility. A system could help an adviser find a passage in a procedure. Its polished answer had also relied on the superseded procedure dated 11 June 2025. The citation was genuine. It made the error findable. It did not make the old rule current.
Live work had changed the question again: a source edit marked minor altered an approval answer, Maya paused the feature, and it returned in a narrower form (Chapters 8 to 10).
The company could have bought a product before learning any of that. It would still have needed to resolve which procedure was current, who approved a credit, how reversals were recorded and who could change a source.
Daniel listened as Maya read the opening sentence of the paper.
“It sounds cautious,” he said.
“It is a strategy, not an apology.”
He asked what the board would be asked to fund. That was the harder question. The sentence described a discipline. It did not yet choose priorities.
Maya opened the Business Case Template. She would give the board a small set of decisions: continue the bounded service-credit use; repair and maintain the sources and records on which several possible uses depended; allow defined routine assistance within standing permission; and investigate a consequential small-claims proposal without granting it decision authority.
The tender trial remained closed pending its unresolved questions. The marketing assistant remained a separate register entry, not a benefit smuggled into the service-credit case. Neither would become an approved priority merely because it appeared on a slide.
A strategy chooses what deserves the next decision, not only what looks possible.
The number beside the choice
At 10.22, the CFO arrived with the cost worksheet she had used for the limited retrieval use.
“We need to be clear which of these costs will grow,” she said.
Maya asked whether the business case could promise that widening the adviser group would spread the same cost across more users.
“Only if the work stays the same,” the CFO said. “More users may produce more questions, reports and source checks. We have not established that it scales without more support.”
The benefit view was equally careful. In some cases the feature helped an adviser find a procedure, especially when the question used a customer’s language rather than the document title. In others it added a step between the adviser and a source they already knew. The team had not shown a cash saving. It had shown a bounded service benefit that could be examined over another review period.
Maya wanted that distinction in the board paper. Tom had asked for strategy. If she offered a return figure built from every minute an adviser might save, the board could authorise expansion on a benefit that had not been earned.
The CFO suggested that each proposed priority carry two cost lines: what it would take to reach the next decision, and what it would take to operate the use if that decision were yes. An investigation into small claims had a research and design cost. It did not yet have an approved deployment cost or benefit. A routine meeting-summary use had a much smaller continuing burden. The service-credit feature had actual operating evidence, including the cost of a pause and repair.
What would it cost to reach your next decision, and what would a yes commit you to?
Guidance
Take one proposed priority and separate the work needed to investigate it from the work needed to operate it. Mark what has been measured and what is still estimated. Do not turn approval to investigate into an assumed commitment to deploy.
Your answers are saved only in this browser on this device. Nothing is sent anywhere.
Maya put the distinction into the Business Case Template: for each priority, the board would see what had been measured, what was an estimate and what remained unknown (see Chapter 11).
“Will they accept that?” Daniel asked later.
“They can reject the investment,” Maya said. “They should not have to guess what the number means.”
What the law question changed
Nadia came in after lunch. She had prepared a short account of the Legal and Regulatory Scan, but Maya did not want to turn the strategy paper into a legal memo.
Tom had asked what the law required. Nadia’s scan had shown why the answer had to follow each use. The organisation already had responsibilities in handling customer information, making and recording decisions, dealing with staff material and honouring agreements. Introducing an assistant did not replace those responsibilities. Nor did a general statement about AI settle how they applied to every workflow.
For the service-credit use, the team could describe the information boundary, source ownership, case approval and records. For the approved internal meeting-summary use, the output remained a draft checked by a participant before entering the ordinary record. The tender proposal raised a question about customer documents and a proposed data route that the team had referred for specialist advice. The trial remained closed while that question was unresolved.
Nadia had also established a regular scan of the register and triggers for an earlier review. A changed audience, new data, new place, altered contract or stronger authority for an output could change the questions. The board did not need to approve each search Nadia performed. It did need to understand that the work would recur and that a material open question could hold a proposed use.
Maya drafted a paragraph.
Each proposed use will be examined against the rules and commitments that already govern its underlying work. We will seek specialist advice where the decision exceeds the team’s knowledge or authority. We will review that position when the use, relevant rules or circumstances change.
Nadia read it.
“Add that an unresolved question stays unresolved in the decision record,” she said. “Otherwise someone will hear ‘we did a scan’ and assume it was cleared.”
Maya added the line.
The small-claims proposal made the point plain. A tool that helped a worker find a claims procedure was one kind of use. A system whose output could decline a claim was another. The board could authorise investigation of the latter without pretending the legal, service and appeal questions had already been answered.
Nadia did not ask Maya to frighten Tom with every possible failure. She asked her to keep the proposed authority honest.
The arrangement that would survive them
At 8.51 the next morning, Leila and the incoming customer operations lead joined Maya.
Leila’s regional move was nearly due. The service-credit retrieval feature would continue, but responsibility for its sources and outcome would move with the customer operations role. Daniel’s technology role would remain separate. The register now showed the handover date, the current decision, the narrowed boundary, the review trigger and the ordinary fallback.
Maya asked the incoming lead to explain the position as if Tom had never heard of the experiment.
“A small adviser group can use it to find and check service-credit procedures,” she said. “I own the source and service outcome, and Daniel can disable it. We review it before authority expires and sooner if the conditions change.”
Leila looked satisfied. She had not prompted the answer.
Maya put the operating arrangement in the strategy paper. She resisted drawing a large organisation chart. The board needed to know who could say yes, who owned an outcome, how a use came back for review and who could stop it. It did not need to see every meeting invitation.
Leila asked where the standing customer panel belonged. It was already part of how the organisation heard from people affected by its service processes. It could help them understand whether an explanation, delay or route to challenge made sense outside the company. It could inform a consequential proposal such as small claims. It would not approve a system or settle a person’s case.
Maya placed it beside the internal challenge function as a source of outside experience. Different jobs, both useful.
Governance holds when the next person can run it without the original team in the room.
The incoming lead asked whether she could authorise an experiment with the same retrieval feature for refunds.
“Owning customer operations lets you bring the proposal,” Maya said. “It does not extend the service-credit permission. The problem, sources, data and consequence need their own decision.”
That answer would go into the staff-facing guidance as well. The standing arrangement had to make sensible small uses easy enough to propose while preventing permission from spreading by resemblance.
Three routes for new work
Maya built the middle of the board presentation around the three rows the company had been using.
The first was routine use within permission, the second a bounded experiment, the third a consequential system (see the Introduction and Chapter 14). A small-claims decline proposal belonged in the third row even if its technical parts looked similar to procedure retrieval.
Tom would recognise the distinction. It made strategy less dependent on product names. The same capability could show a passage to an adviser, propose an outcome to a reviewer or take effect as a refusal. The organisation had to decide which of those jobs it meant to permit.
Daniel read the slide and asked about uses discovered after staff had begun them. The initial register had contained unverified tender, service-credit and marketing activity. A strategy that addressed only future proposals would miss the next use that appeared through convenience.
Maya added an entry route. Staff could report an existing use without being asked to defend its past existence on the spot. The owner would contain it if necessary, establish what it did and bring the appropriate decision forward. An unverified use would not gain authority merely because it had become familiar.
She also added a refusal route. A proposal could be declined when the process problem was better repaired another way, when the source could not be trusted, when the cost exceeded the supported benefit, or when the organisation could not own the consequence.
What finding would lead you to decline one of your current proposals?
Guidance
Choose a proposal with a named problem and compare it with repairing the existing process. Identify a finding that would make the proposed use unjustified, even if the technology worked. Consider who would record that decision and preserve what the investigation taught the team.
“No” was a strategic choice, not a failure to be ambitious.
The sales director would not like seeing the tender tool remain closed. Maya planned to tell him the reason plainly: a proposed use of customer documents had raised questions about permission and the data route that needed to be answered before a trial. If those answers supported a bounded test, it could return with a proper proposal.
The board paper without a matrix
At 6.12 that evening, Maya printed the draft Board and Management Presentation.
The first page answered Tom’s question directly. The company would use AI for defined work where a person could establish a benefit, give the output a limited role, maintain the sources and records, and withdraw the permission when its reasons no longer held.
The next pages showed the choices behind that sentence. The service-credit use had been narrowed and renewed. The approved meeting-summary use could continue within its routine boundary. The tender trial was closed pending its specific question. The marketing assistant remained to be verified on its own terms. Source and record repair mattered even where no AI tool was selected. The small-claims proposal was ready for investigation, not decision authority.
The CFO’s cost view sat beside those priorities. It showed the work of operating a use, not only paying for software. Nadia’s scan showed how existing obligations and open specialist questions followed each entry. The standing arrangement showed who owned decisions after Leila’s move.
Daniel noticed there was no diagram with a target number of AI applications.
“Was that deliberate?” he asked.
“Yes. The number of applications is not the outcome.”
He suggested a page about technical capability. The board did need to know what the existing knowledge base could support and why the small internal build had been deferred. Maya gave it a short place in the evidence, connected to the decisions it affected. She did not make the technology inventory the centre of the strategy.
She had also prepared a Stakeholder Map for the board discussion. It showed advisers, team leads, source owners, technology, Finance, challengers, decision-makers and customers. Maya wanted Tom to see that a use did not end when someone enabled access. It moved work across people. The map made the handoffs visible without turning the presentation into a staffing plan.
The Project Plan Template held the proposed sequence. Source and record work would continue. The current retrieval use would reach its next review before any expansion. The tender question would be resolved before a trial could reopen. The small-claims team would investigate its decision path and bring back a proposal with consequences, evidence and review rights defined. Each item had a next decision, an owner and a condition for proceeding.
Maya left the deck on her desk overnight. In the morning she removed a sentence saying the organisation would “lead in responsible AI”. She could not tell Tom what action that sentence authorised. The rest of the paper was stronger without it.
What Tom challenged
At 9.07 on Friday, Tom began the board discussion with the same question he had asked months before.
“What’s our AI strategy?”
Maya gave him the short answer. Then she showed the choices.
Tom listened through the service-credit account. He asked why the board should care about one small retrieval feature when larger possibilities were appearing every week.
“Because it showed us what has to be true before a use deserves to grow,” Maya said. “A current source. A bounded role for the output. An owner of the customer outcome. A way to see errors, pause, recover and count the work. We can use that learning on larger proposals. We cannot treat the small use as evidence that they are safe.”
Another board member asked whether the company was moving too slowly. A rival could announce an automated claims process while they were still investigating theirs.
Maya did not claim competitors were wrong. She asked the board to distinguish announcing a capability from owning a decline decision when a customer challenged it. The proposed small-claims system might be worthwhile. The investigation would establish the rule, information, explanation, review route, cost and authority needed for it. The board could fund that work now.
The CFO was asked for the return on the service-credit use. She described the actual cost of preparing and operating the narrow use and the supported benefit of easier procedure discovery in some cases. She did not claim a cash saving. The next review would decide whether the benefit remained worth its continuing work.
Tom asked whether Maya had lost confidence in AI.
“No,” she said. “I have more confidence in the uses we can explain. I have less reason to make a general promise about every use.”
He turned to Nadia’s page. “And the rules?”
Nadia described the scan by use. Existing obligations attached to customer decisions, information, records and agreements. New or changed rules would be checked on a cadence and when a proposed use changed. Questions beyond the team’s expertise would go to specialists before authority was granted. She did not give the board a blanket statement that the company was compliant with every possible future requirement.
The board had wanted a strategy. It now had decisions to make.
What the board selected
The discussion ran past the allotted time. Tom asked Maya to put the proposed priorities on one page before the board voted.
She did.
First, maintain the narrowed service-credit retrieval use for its authorised group until the next review. Keep the ordinary search and team lead route available. Do not expand it by default. The customer operations role would own the sources and outcome after Leila’s move. Daniel’s team would retain the configuration and disable route.
Second, fund the ordinary source and record work that the experiment had exposed. Procedures needed clear status and ownership. Reversed decisions needed complete records after payment, not only a successful transaction. Those repairs would help service work whether or not retrieval continued.
Third, continue routine internal assistance inside its approved boundaries. The meeting-summary use could remain a checked draft for permitted notes. It would not become an automatic account of a disputed decision.
Fourth, keep the tender trial closed until the question about prospective-customer documents and the proposed processing route was answered. Then the sales director could bring back a bounded proposal or close it.
Fifth, authorise an investigation of the small-claims system. The investigation would compare the current process with possible assistance, identify what a decline would mean for a claimant, and specify the records, explanation, review route, cost and rules relevant to that decision. It would use the standing customer panel to understand the experience of receiving and challenging an outcome. It would not grant the system authority to decline a claim.
The board selected those priorities. It asked Maya to report progress through the normal management cadence, with a decision brought back whenever a proposal sought stronger authority.
FIG-15.1
The board’s selected priorities
| Priority | Authority granted now | Next decision |
|---|---|---|
| Service-credit retrieval | Continue the small, narrowed Assist use until its review | Renew, narrow, expand or retire on evidence |
| Procedure and case-record repair | Continue as owned service work | Check whether the repairs hold in ordinary cases |
| Routine meeting summaries | Continue within the existing checked-draft boundary | Review if input, audience or output role changes |
| Tender tool | Keep the trial closed | Consider a bounded proposal after the open question is resolved |
| Small-claims system | Investigate the process and possible routes | Decide whether any proposed authority, including decline authority, can be justified |
Tom read the final row twice.
“We are paying to investigate without promising to build?”
“Yes,” Maya said. “That is the decision we can make now.”
After the meeting
At 4.37, Maya returned to her office and opened the first Demand Note from the Monday when Tom had asked for a strategy.
The question had arrived before the organisation knew how many uses were already under way, why service-credit decisions varied, which procedure was current, or what a live assistant would cost to maintain. A polished strategy document then could have named aspirations. It could not have settled those facts.
She filed the board decision with the register. The service-credit entry retained its expiry. The tender entry remained closed. The small-claims entry changed from an unshaped idea to an authorised investigation with a named owner, boundary and next decision. The standing customer panel had a defined part in that investigation. None of those entries borrowed authority from the board’s general approval of the strategy.
Leila called before leaving for the day. The incoming customer operations lead had completed the source handover and knew when the next retrieval review was due. Priya would still be available to explain what advisers had seen in live work, but the new lead owned the outcome. Daniel had checked that source-change notices reached the right role.
“What did Tom say?” Leila asked.
“He asked why a small service-credit use mattered to a big strategy.”
“What did you tell him?”
“That it taught us how much work sits behind a small permission.”
Leila was quiet for a moment. “That sounds right.”
Maya looked at the board paper. It did not settle every future choice. It gave the organisation a way to make them in sequence. A repaired process could win over a tool. A bounded use could continue without expanding. An investigation could be funded without granting the authority it was investigating. A consequential proposal could return when its owner had evidence to support it.
Tom had his answer. The organisation would now have to live by it.
Use the Business Case Template to show the real alternatives and whole cost of each proposed use. Put the decisions, evidence and limits into a Board and Management Presentation. Use the Project Plan Template to give every priority a next decision, and the Stakeholder Map to show who owns and experiences its effects.
In the Epilogue, The Empty Chair, the small-claims proposal returns with a request for decline authority. The question is whether the standing arrangement can hold when the decision is harder than finding a procedure.
