The Friday Paper

The continuing case in this book is invented. Its organisation, people and events represent recurring situations rather than a disguised account of one company.

At 9.07 on Monday morning, Maya Sen was asked for the company’s AI strategy.

The request came near the end of the executive meeting, after the financial forecast and before a discussion about office leases. It occupied less than four minutes.

Tom, the chair, had attended an industry dinner the previous Thursday. Two people at his table had spoken about AI agents. One said his company expected large productivity gains. The other said organisations without an AI strategy would soon struggle to compete.

Tom placed his coffee beside the meeting papers.

“We need to get ahead of this,” he said. “What’s our AI strategy?”

Maya, the chief operating officer, looked towards Daniel, who led technology. Daniel looked back at her.

The company did not have an AI strategy. It did, however, have AI.

A sales manager had built a tool that could analyse tender documents. It had impressed his team until it attributed a customer requirement to the wrong section of a contract. A person caught the error before the proposal left the building. The incident had not been recorded because nobody was sure where it belonged.

Customer service wanted an assistant that could search old cases. The team had spent months asking for its knowledge base to be cleaned. Some procedures contradicted one another. Product information sat in shared drives, old tickets and the memories of two experienced employees.

A large customer had also sent an assurance questionnaire. One question asked whether the company maintained a register of AI systems and their approved uses. The response was due on Friday, alongside the paper Tom now wanted for the board.

“Could we start with the service desk?” the chief financial officer asked. “If AI handles the repetitive enquiries, there should be a saving.”

Leila, who ran customer operations, closed the folder in front of her.

“Which enquiries are repetitive?” she asked.

The chief financial officer glanced at his notes. “You tell us. You run the team.”

“We record the same problem under six different names,” Leila said. “The reports make six problems look like one large category and one problem look like six small ones. If the assistant learns from that, it will answer quickly. I’m less sure it will answer correctly.”

The room went quiet for a moment.

Daniel said he could arrange demonstrations from three providers. Tom nodded.

“Good. Let’s move. Maya, can you give us a recommendation for the board paper by Friday? We have a budget decision next week.”

By 9.16, Maya owned an AI strategy, a customer assurance response and a deadline tied to funding.

FIG-P.1

One request creates five decisions

Strategy

What direction can we defend?

Assurance

What can we truthfully evidence?

Existing use

What needs a boundary now?

Service failure

What work needs diagnosis?

Budget

What spending can we authorise?

Back at her desk, she opened a new document and typed:

AI Strategy

She began with a list of possible uses. Customer support. Tender analysis. Internal search. Invoice processing. Marketing content.

It was the sort of list the board expected. It was also easy to produce. She could add a coloured matrix, rank the ideas by value and feasibility, and recommend two pilots. Daniel could obtain vendor estimates. Finance could attach a savings range.

The paper would look finished.

FIG-P.2

Two possible board papers

Expected paper

Use-case matrix; vendor estimates; forecast savings. Assumptions can disappear inside the presentation.

Maya’s paper

Current exposure; decision owners; knowns and unknowns; bounded next steps. Investment direction still has to follow.

Her first interruption came from the sales director.

“The tender tool is still a trial,” he said from the doorway. “It hasn’t touched customer data.”

“Where did the tender documents come from?”

“The sales folder.”

“Those are customer documents.”

He leaned against the doorframe. “They’re commercial documents. We didn’t upload personal information.”

“Do we know what the provider retains?”

“Daniel’s team can check. The point is, it saved hours.”

“How many?”

“We haven’t measured it formally.”

He wanted the tool included in the board paper as evidence that the company had already begun. He also wanted the assurance response to avoid language that might alarm the customer.

“We caught the contract error,” he said. “That’s what review is for.”

After he left, Maya stared at the list on her screen. One line now contained a claimed saving with no baseline, a customer document used under uncertain terms, an error that had not been recorded and a reviewer who happened to notice it.

The line still read: Tender analysis.

At 10.42, Leila called.

“I sounded difficult in there,” she said.

“You sounded as though you knew something the rest of us didn’t.”

“I know the knowledge base is a mess. I also know my team is tired. If a decent assistant could find the right procedure, they’d use it tomorrow.”

“So you want the pilot?”

“I want the problem fixed. If the pilot helps, good. But if we put a fluent answer on top of contradictory instructions, customers will get the contradiction faster.”

Leila described a case from the previous week. A customer had called twice about a service credit. The first employee followed an old procedure and declined it. The second found the revised rule and approved it. The customer had spent forty minutes correcting a decision the company should never have made.

“Which version would the assistant have found?” Maya asked.

“That depends on what we feed it. Nobody has decided which sources are current.”

Maya saved the use-case list in a draft email to Tom. Her cursor rested over Send. It would meet the request she had been given and buy her time, but it would also turn five guesses into the company’s apparent direction.

She closed the email without sending it.

She did not yet know whether the service desk needed AI. She knew the customer had already paid for the organisation’s confusion with time and frustration. She knew Leila’s team wanted relief, while finance wanted a saving it could put into next week’s budget. She knew a software demonstration would not settle either matter.

The customer questionnaire asked the company to identify each AI system, its approved purpose and the person accountable for it. Maya could not answer any of the three questions for the tender tool. She borrowed the first question for a blank page:

What has happened?

The chair had requested an AI strategy. A budget decision was approaching. A customer wanted assurance. Employees were already experimenting. At least one experiment had used customer material and produced an error.

Below that, she wrote:

What decision is needed by Friday?

The company could not responsibly decide its entire AI strategy in four days. It could decide which existing uses needed immediate boundaries, which proposed uses deserved diagnosis, and who would own that work.

Maya called Daniel and Leila. The three of them agreed on a provisional first response.

Until the provider terms had been checked, the sales team would stop uploading new customer tenders to the trial tool. The existing material would not be deleted until Daniel had established what records were needed and what deletion would actually remove. Customer service would not begin a vendor pilot that week. Leila would nominate one service problem for a short diagnosis, starting with the conflicting service-credit procedures. Daniel would identify AI features already enabled in the company’s main software.

These were temporary boundaries, not a complete inventory. Maya wrote that distinction into the paper.

At 4.36, she sent Tom a one-page note.

It did not contain a roadmap or a savings forecast. It recorded the request, the pressure behind it, what was known, what remained uncertain and three decisions the board could responsibly make on Friday.

Tom replied six minutes later.

I expected options. This reads like a list of questions.

Maya read the message twice. She reopened the opportunity matrix and wondered whether she had made the wrong call. On Friday she would have to defend a paper that did not resemble the one the chair had asked for.