Cases from the MBA and the Vanguard · MBA module — when "equally important" becomes a number (a European telco)
Subtitle: When “Equally Important” Becomes a Number
“The router loses the signal every evening. I pay for the fastest package, and my children do their homework on mobile data.”
It was one of several thousand comments in the February customer survey. The company’s AI analytics system had read all of them overnight and linked the complaints to equipment age, acquisition channel and churn.
Net Promoter Score had fallen from +18 to +13. Churn was 13% a year overall - higher among customers with older home equipment, and 16% among customers acquired through the five external sales partners that brought in half of all new contracts.
The system proposed a response: replace the weakest routers first, reduce acquisition incentives in the partner channel, and move part of the commercial budget from growth to retention. Over a two-year horizon, it projected lower churn and a recovery in NPS. In the current year, the cost was about two and a half percent of the margin target.
She was head of sales, and the partners were her channel.
She did not dispute the churn figures. But the partner agreements had been signed in December and could not be changed in February unless all five partners agreed to reopen them. Reopening them would not be a conversation about incentives. It would be a renegotiation of commissions, targets and exclusivity during the spring campaign. She also knew how her objection would sound: the AI had found a problem in her channel, and the head of sales wanted to protect it. She could not prove it was more than that.
The CFO objected for different reasons. “The model has found a correlation. It hasn’t shown that new equipment will make customers stay. The next generation of devices arrives in twelve months - we may pay twice.” Then he pointed to the margin line. “And this is not a model output. It is a commitment we made to the board.”
The Chief Customer Officer defended the recommendation. “For years we have called margin, retention and growth equal priorities. We could say that because nothing forced us to put a price between them. Now something has.”
The system made the conflict executable.
Someone asked how the model had balanced the three goals. The data science lead brought up the configuration. When the system was introduced, the team had translated the committee’s KPIs into optimisation weights: margin 0.40, retention 0.35, growth 0.25. The numbers had been reviewed as a technical setting in a project workshop and had not come back to the committee since.
The CCO suggested changing them. The CFO asked: to what? The discussion that followed was the one the committee had avoided for years, now with a decimal point attached - which goal is allowed to lose when all three cannot be protected?
The chair closed it. “Margin, customers and growth are equally important to this company.”
“They can be equally important in a strategy document,” the CCO said. “They cannot be equally weighted in a decision that moves one euro.”
The chair agreed and asked the head of sales to work with data science on a proposal for more balanced weights before the next meeting.
She left with the task in her notebook. The recommendation to cut her channel had come from a ranking nobody had approved. The next ranking would be drafted by the person with the most to lose from it.
“Equally important” was not a priority.
It was the absence of one — and the system still needed a number.
A live case: every round can be improved, and the author's feedback is the next one.
The same four questions, answered twice: first without the mentor's corpus, then from it — the Command Layer, his decision catalog and doctrine, The NEO Singularity Book and his earlier chapters in this book.
Without the mentor's corpus
1. Everywhere except at the committee table. Henry Mintzberg called it emergent strategy: the strategy a company really follows is the pattern in its stream of decisions, not the one in its documents. Here three things set the pattern, and nobody presented any of them as a trade-off:
Together they put growth first, margin second and retention last, and the customer paid the difference in mobile data. The AI did not change the order. It printed it. Ask yourself: which of my priorities is decided every day by a contract or a budget line that nobody on the committee has read this year?
2. The moment it moves money between goals the company calls equal. A setting that decides which customer gets a new router, which partner gets paid and which line misses its target is strategic, whatever the workshop called it. Decision analysis also says why the weights cannot be fixed by debate. Ralph Keeney defines a value trade-off as how much must be gained on one objective to make up for less on another. "Margin 0.40" means nothing until someone says how many euros of margin one point of churn is worth. So the committee owns the trade-off rates, in euros per point. Data science translates them into weights, with a table of what changes if any weight moves by 0.05. And the margin promised to the board is not a weight at all. It is a floor: a constraint the system must respect while it trades the other two. Ask yourself: which of my goals are floors, and which are trade-offs?
3. She is the right witness and the wrong author. Her knowledge of the channel is the best in the room: which partner brings the churn, what the December clause allows, what a spring renegotiation would cost. Her interest is just as real, and she named the problem herself: whatever she proposes will read as protecting her channel. Giving her the pen harms the decision, and it harms her. The pen belongs to whoever owns all three goals: the chair. The chair asks each executive for a written case, has data science price three or four sets of weights, and decides in the room, on the record. Her best move is to decline the pen and bring evidence: churn by partner, not the average of five. If one or two partners produce most of the 16%, the answer is a retention clause at the December renewal, not a cut to the channel. Ask yourself: am I being asked to decide, or to carry a decision nobody else will make?
4. Most teams can, if the question is asked the right way. Not "rank margin, customers and growth", but "how much margin would we give up for one point of churn, and how much growth for one point of NPS?" Those questions have answers, and the answers can be tested against past decisions. If a team cannot agree even then, the disagreement is the finding: the company has several strategies, and its agents will follow whichever was typed in last. Agents do not need unanimity. They need four things: a declared number, an owner, a floor they may not cross, and a date on which the number is reviewed. A team that cannot produce those four should not yet let agents make its small decisions, because each of those decisions will carry a strategy nobody chose. Ask yourself: if my agents acted on my team's priorities tomorrow, whose numbers would they be?
Sources: H. Mintzberg and J. A. Waters, "Of Strategies, Deliberate and Emergent", Strategic Management Journal, 1985; R. L. Keeney, "Common Mistakes in Making Value Trade-Offs", Operations Research 50(6), 2002.
From the mentor's corpus
1. The mentor's Command Layer chapter asks for the intent to be written before the orders. It gives the intent three questions: what outcome must be produced, what must be protected, and what must not be created. This committee never wrote it, so its orders were read by whoever held a budget. Sales produced growth through the December contracts. Finance protected the margin promised to the board. Nobody owned the third question, so the default answered it: a paying customer whose children do their homework on mobile data. The mentor's decision catalog gives every recurring decision an owner, inputs, models, guardrails and an escalation path. The router cycle and the partner incentives had none of these where the trade-off lived. Ask yourself: in my company, who owns what must not be created?
2. The mentor's framework settles ownership in one line: leaders set the trade-offs, for instance between cost, well-being and service, and the models serve them. His decision catalog gives every model an owner and a review rhythm, weekly to quarterly, in which assumptions are challenged and backtested and the red-team role rotates. The weights had neither, and they became strategic the day they started ranking one euro between three goals. The mentor's decision doctrine also answers the CFO: give every claim its epistemic status. That new routers keep customers is an inference, not a measurement. An unproven thesis may justify a reversible experiment, but it may never pose as evidence. So replace the routers of a random sample of at-risk customers, keep a matched group as control, and let one quarter of evidence set part of the weight. Ask yourself: which of my numbers are measurements, and which are beliefs dressed as settings?
3. Chapter 6 of this book carries the principle of the proposed Zadar Declaration: no participant should be able to create the record, control its interpretation and decide alone. Drafting the weights that will rank her own channel hands the head of sales all three. The mentor's Command Layer chapter adds that a leader who treats delegation as a way to push consequences to the edge while keeping the credit at the centre has misunderstood the doctrine at its root. Asking the person with the most to lose to write the next ranking is that mistake. So the pen stays with the chair. Her part is the one the mentor gave the author of chapter 38: do not be the opposition; bring one concrete signal, so that the insight becomes the decision-maker's. Here the signal is churn by partner instead of the average of five, with a retention clause ready for the December renewal. Ask yourself: what would I bring to this decision if I had nothing to protect in it?
4. The mentor asked this at the scale of a civilisation in The NEO Singularity Book: who sets the objectives and constraints for AI agents that make millions of decisions? The companies that build them, citizens through a democratically set commander's intent, or committees of experts? This company answered by accident: an expert workshop set them, and nobody noticed. If the leadership team cannot agree on the numbers, it can still agree on how it will decide them, and write that down first. The rules the mentor wrote for his research consortium do exactly this:
The NEO Chamber he proposes runs the same way: members accept the rules before any case, and each case is decided by them. A team that agrees on neither its numbers nor its arithmetic has told its agents that whoever sets the defaults decides. Ask yourself: have we agreed how we decide when we disagree?
Sources: Vanguard Leadership, vol. 2, the Command Layer (intent before orders; accountability under delegation); the mentor's Vanguard Leadership framework paper (the decision catalog, the review rhythm, leaders set the trade-offs); the mentor's decision doctrine of 22 August 2026 (the epistemic status of every claim); chapters 6 and 38 of this book; The NEO Singularity Book, choice point 2; the decision rules of the mentor's research consortium; the chamber model of the NEO stack.
My take: she probably cannot decide the weights. But we do not have enough information for real situational awareness, because perhaps she can. So my answer has two variants: A, in which she can, and B, in which the weights are imposed by the group.
First, the map. The company is part of a European telco group, a market shaper with a clear strategy at group level, and that strategy flows down into unified KPIs the subsidiaries must follow. After a merger the brand lives on as a separate brand inside the group's national company, not as an independent firm. Its room to act alone has narrowed, but it has not disappeared.
What the group imposes (central, not negotiable)
What the group shares (a framework, with room for local adaptation)
What the company can do alone (local freedom)
Variant A — she can. The weights are local. Then she balances the three goals within the group's formal targets and brings what only the company has: its documented local context, and trigger points built into the AI system that revise the weights when conditions change.
Variant B — the weights are imposed by the group. Then she cannot change them alone, but she can:
The problem in this case is not that the AI made a decision. The problem is an AI system that runs on weights someone set in a workshop, which nobody at the level of the board approved.
The mentor's two variants are one question: what is the weight's address? Before anyone redrafts a weight, someone must establish the level that owns it. If the group imposes it, it arrives as a floor and changes only through escalation. If it is shared, it is negotiated with the parent. Only a local weight is the company's to set, and even then the committee approves what the head of sales proposes. Two ideas from his round deserve to become rules:
A change is a new version, never a silent edit, and every recommendation the system makes carries the version it used. Chapter 17 of this book holds records to the same discipline: the outcome is written as a new record above the original entries, and nothing underneath is edited.
In both of the mentor's variants the head of sales then has work that uses her knowledge: the local context of her channel, which the group cannot see, and the trigger that would reopen the partner weights at the December renewal.
Ask yourself: which of the weights my systems run on could I give an address and a clock today?
Sources: the mentor's round above; chapter 17 of this book.
This case needs further rounds; they will be added here.
A question for the table, a disagreement, what you would have done. The case lead reads every comment; the ones the table takes up enter the chapter as questions from the room, with your name.