Haute Lumière

Commerce · II.07 · MMXXVI · daylight

La Bourse  /  Volume II  /  Nº II.07  /  Workbook — the executive

WORKBOOK — THE CORPORATE EXECUTIVE

Chapter II.07 · Information, Signal, and Price

For the executive with a P&L, a board, and a reporting calendar. The language of the firm, without apology. Everything here is costed, and most of it is already supported by numbers your own finance function produces.

---

THE PREMISE, STATED COMMERCIALLY

Your business runs on signals it has never priced. Some of them are prices, and those are cheap, fast and structurally blind past one dimension. The rest are measurements you either buy, build, or do without — and at present you are almost certainly buying some that change nothing and doing without some that would change a great deal.

This workbook gives you the arithmetic that sorts them. It is the same arithmetic your capital committee already uses, applied to a category it has never had a line for: information as a purchasable asset with a computable return.

Three numbers do the work.

• k — the count of attributes that would change a given decision. 1/k is the share of the decision a single price can be expected to carry.
• EVPI — the expected value of perfect information on one attribute. Howard, 1966. It is a hard ceiling on what any measurement of that attribute can be worth, and it is computable before you spend anything.
• Cost per bit — what a channel costs against what it delivers. For comparison, general equity price discovery costs about 10,421 dollars a bit at the chapter's inputs, and a bit aimed at one decision was worth 95,950 — a ratio of 9.21.

That last comparison is an order of magnitude rather than an identity, and it carries the operating rule of the whole workbook: instrument the decision, not the world.

---

PART ONE — DISCOVERY

Days 1–30: find what your firm already knows and never transmits

Exercise 1.1 — The five-place sweep for information (one week)

Every firm has signals in it that nobody costed. Go and find them.

You are assembling a list of dimensions your firm already perceives and does not transmit. Every one of them is a candidate channel with a zero acquisition cost on the sensing side, because somebody is already sensing it.

Exercise 1.2 — Weigh the working channel first (half a day)

Before anyone in the room says the word externality, establish what price discovery is doing for you. Take one commodity input and compute what price dispersion across your sources costs you, and what it would cost if you had no market price at all. Kerala's fisheries are the clean case: price dispersion collapsed and waste went to essentially zero when one number arrived earlier, with fishermen's profits up about 8 percent and consumer prices down about 4 percent at the same time.

Run this exercise first, every time. A meeting that opens on the limits of prices loses the finance function in the first four minutes. A meeting that opens on what price discovery is worth to this business, quantified, does not.

Exercise 1.3 — Name a vector (2 hours, with the decision owner in the room)

Take one recurring decision with real money on it — supplier selection, capital allocation, product specification, site choice. List every attribute that would change it. State k. State 1/k.

At k = 12, a price carries 8.3 percent of a randomly oriented preference gradient and leaves eleven dimensions — 91.7 percent — exactly indistinguishable. Put that on the slide. It reframes the conversation from are our prices right to which of the eleven are we flying blind on, and do we mind — and the second question is one an executive team can actually answer.

---

PART TWO — THE ARITHMETIC

Days 31–45: rank by value of information, not by importance

Exercise 2.1 — The EVPI table (one day)

For each attribute on the vector, estimate three numbers with the decision owner and the relevant functional head. Rough is fine; the ranking is robust to error in a way the levels are not.

• p — probability of the state that would change the decision
• L — loss from acting as if good when bad
• c — cost of the remedy taken regardless

  without information        min(p L, c)
  with perfect information   p c
  EVPI                       min(p L, c) - p c
  bits supplied              H(p)
  value per bit              EVPI / H(p)

Worked: p = 0.10, L = 1,000,000, c = 50,000. pL = 100,000; c = 50,000; pc = 5,000; EVPI = 45,000; H(0.10) = 0.4690; value per bit 95,950.

Two things your team will get wrong the first time.

Confusing important with worth measuring. An attribute on which you are already certain carries no bits and can change nothing, however much it matters. H(p) goes to zero at both ends and so does EVPI.

Missing the ceiling. EVPI is bounded above by min(pL, c). In the worked case no measurement of that attribute can ever be worth more than 50,000 a year. Put the ceiling in the paper before you go to procurement, and the negotiation is finished before it starts.

Exercise 2.2 — The EVPI curve (90 minutes)

Walk p and show the committee the shape:

  p = 1%    EVPI    9,500    H(p) 0.0808    117,584 per bit
  p = 5%    EVPI   47,500    H(p) 0.2864    165,854 per bit
  p = 10%   EVPI   45,000    H(p) 0.4690     95,950 per bit
  p = 30%   EVPI   35,000    H(p) 0.8813     39,714 per bit
  p = 50%   EVPI   25,000    H(p) 1.0000     25,000 per bit

It peaks at p = c/L and is zero at both ends. Measure what you cannot guess, and guess the rest. That is a theorem and it is also, conveniently, the best cost-discipline argument you will ever bring to an information budget.

Exercise 2.3 — Compute your floor before anyone else does (half a day)

Every channel has a scale below which it loses money. Use the chapter's form on whatever you are proposing:

  MRV cost 20,000/yr, carbon 80/t

  500,000 tCO2e/yr  ->  EUR  0.04 per tonne of measurement
   25,000 tCO2e/yr  ->  EUR  0.80
    5,000 tCO2e/yr  ->  EUR  4.00
    1,000 tCO2e/yr  ->  EUR 20.00

  abatement needed to pay for measuring    20,000 / 80  =    250 tCO2e
  at a 10% abatement rate, the floor        250 / 0.10  =  2,500 tCO2e/yr

The EU wrote its own floor into law an order of magnitude higher: Article 27 of Directive 2003/87/EC excludes installations under 25,000 tonnes, and Regulation (EU) 2018/2066 lets de minimis source streams under 1,000 tonnes use simplified methods.

Bring your floor to the board yourself. A regime that does not know its floor is reopened at the floor, by holders who computed it first — which is exactly what the Commission's Omnibus package of 26 February 2025 did to sustainability reporting, proposing to remove roughly four-fifths of companies from scope at a stated saving of 6.3 billion euros a year: about 157,500 per company, or 143 per datapoint across the eleven hundred in ESRS Set 1.

---

PART THREE — DESIGN

Days 46–60: the instrument

The structure. A Channel Budget Facility: a ring-fenced, pooled measurement vehicle whose covenant is written in value per decision-bit rather than in datapoints.

Balance-sheet treatment. The subscription is operating expense. Do not capitalise a measurement system — its useful economic life is exactly as long as the decision uncertainty it resolves, which is not a life your auditors will accept. Where the attested record carries a market price, as an allowance or certificate does, recognise that instrument separately on its own terms. Keeping the channel and the commodity apart is what makes the whole thing auditable.

Counterparty. Start with the party who already buys your output and already wants the attribute — a customer, not a regulator. A customer-funded channel has somebody inside it whose money depends on its accuracy. A compliance-funded one does not, and will degrade into a template within three cycles. This is Grossman–Stiglitz generalised, and it is the single most reliable prediction in the chapter.

---

PART FOUR — DESTINY AND DELIGHT

Days 61–90: run it, publish it, and make it boring

Exercise 4.1 — One period, against a signed baseline (30 days)

Run the channel once. The baseline is agreed and signed before deployment by the decision owner and finance. An unagreed baseline is not a baseline; it is a future dispute you will lose on burden of proof.

Exercise 4.2 — Publish cost per bit and value per bit, side by side (one page)

One page, to the person who controls the next allocation. Four lines:

1. The decision, and how often it recurs.
2. p, L, c, EVPI — the value.
3. The channel cost, and the cost per bit.
4. The ratio, and what you propose to do with it.

Do not present it as a sustainability initiative. Present it as an information arbitrage with a computed return and a stated ceiling, because that is what it is, and because the second framing gets a different meeting.

Exercise 4.3 — Design for four bits (half a day, with whoever reads it)

Whatever your channel produces, compress the output to four bits and then to one. The binding constraint is your reader's processing capacity, not your sensing capacity — Sims's rational inattention, and the most under-used result in corporate reporting. Chile's front-of-pack octagon is a few bits and moved a national diet; a forty-number panel on the back of the same package had been there for years.

---

THE FAILURE MODES, NAMED

So you can see them coming

It tries to transmit everything. A channel wider than its receiver delivers zero. This is how thousand-datapoint regimes fail and it is not anyone's fault.

It is mandated below its floor. The small holders compute the floor, and the regime is reopened — and the reopening takes the honest measurements down with the rest.

It measures a proxy. The proxy gets optimised and the attribute does not. Goodhart's mechanism, and the most reliable failure in the set. The defence is Ostrom's: monitors accountable to the monitored, graduated sanctions, and a cheap dispute step.

It is funded as compliance. Nobody inside earns anything from its accuracy, so its accuracy decays at the speed of staff turnover.

It never retires anything. The budget accretes, the attention does not, and eventually the whole thing is cut in one line by someone who did not read it.

---

THE NINETY DAYS ON ONE PAGE

---

BOARD PAPER TEMPLATE

Decision requested. Approve a Channel Budget Facility of [amount] to instrument [attribute] for [decision], recurring [frequency].

What price discovery currently delivers. [Quantified. Open here.]

What it structurally cannot. [k, 1/k, and the named dimensions in the kernel. One paragraph, no adjectives.]

Value. p = [ ], L = [ ], c = [ ]. EVPI = [ ]. Ceiling = min(pL, c) = [ ]. Bits supplied H(p) = [ ]. Value per bit = [ ].

Cost. Channel [ ] per year. Cost per bit [ ]. Ratio [ ].

Floor and pooling. Break-even scale [ ]. Members required [ ]. Secured [ ].

Retirement. This facility switches off when [stated condition], releasing [amount].

What we are choosing not to measure, and why. [The honest list. This paragraph is what makes the rest credible.]

---

WHERE YOUR OWN NUMBERS ALREADY SUPPORT THIS

You do not need a new budget line to start, and you should not ask for one in the first quarter. Four places in the accounts already carry the case.

The quality cost line. Whatever you spend on rework, scrap, warranty and recall is the price you are already paying for attributes you did not observe early enough. An EVPI computed against that line is not a projection; it is a measurement of something that has already happened, repeatedly, and your finance function has the series.

The supplier due diligence budget. You are already buying attribute information — audits, questionnaires, site visits. Almost none of it is ranked by value per bit, and a great deal of it is collected on attributes where the answer has not changed in six years. Rank what you already buy before proposing anything new; the first pass typically frees money rather than asking for it.

The insurance premium. An underwriter has already priced some of your kernel and will usually tell you which dimensions moved the quote. That is a free ranking of attributes by somebody whose own capital is behind the answer.

The customer's specification. Every attribute a customer has ever written into a contract is an attribute they would pay to have attested. That list is the counterparty test from the Design movement, already written, sitting in the contracts folder.

The order of the asks. Rank the due diligence you already buy, retire what has stopped moving, and fund the first new channel out of the released budget. The first Channel Budget Facility in any firm should be cost-neutral in year one, and it can be, because most firms are already buying bits they do not read.

---

APPRECIATIVE QUESTIONS FOR YOUR LEADERSHIP TEAM

1. Where has one number arriving earlier or from somewhere new made us measurably better off? What did it cost to move it?
2. Which of our people read something about quality that none of our systems record — and what would it take to give that a channel?
3. If our customers had four bits of attention at the moment of choice, which four would we want them to have?
4. Whose pay here depends on one of our measurements being accurate? If nobody's, who should it be?