The cost of knowing

The cost of rigorous commercial research has collapsed in the last year. This is our view of what that does to markets built on scarce information and to the business of supplying it.

Information asymmetry in private markets

When one side of a trade knows more than the other, the market misprices and under-trades. A buyer who cannot judge quality will pay for the average quality of what is on offer. At that price the sellers of anything better than average walk away; the average of what remains falls; the price follows it down, and the next tier of good sellers leaves in turn. The gap between what a business is worth and what a buyer who cannot see it will pay is the lemons discount.1 This leads to mispricing, capital sent to the wrong places, and fewer trades than there ought to be — we believe technology can now close that gap at scale.

Figure 1

The cost of knowing

Share of deals done with an informed buyer, against the cost of becoming informed.

In a market with a price, the finding-out has been done by market participants. Hayek’s insight was that the relevant knowledge is scattered — held in pieces by the people closest to each part of the work, never gathered in one place. A price gathers it anyway: the number collects what everyone who paid to find something out has learned. A listed company has such a number.2

A private company has no such number. No trading crowd has done the finding-out on a buyer’s behalf, and there is no price to read the answer from. A private transaction carries two problems at once: the seller knows more, and nothing has aggregated what anyone else discovered. So the buyer produces the information first-hand and pays for it, in time or in money. There is nothing to free-ride on: the share of buyers who are informed is set by the cost of becoming informed, and by nothing else.

Figure 2

The lemons discount

What a buyer pays against what the business is actually worth.

Private investments are an attractive instance of this problem to solve, because the stakes are high. The worth of a piece of information rises with what the decision riding on it puts at risk: you pay to be sure in proportion to what being wrong would cost.3 But a diligence study has cost a large consulting firm’s fee whatever the size of the transaction, so there has been a threshold.

The same buyer holds its smaller transactions to a different epistemic standard: set by the cost of knowing, not by what can be known. Most small transactions, in what we see, pay for no external diligence at all.

Below the threshold, the lemons problem applies in full. The transactions that clear are disproportionately the ones that should not have — the seller who knows the business is worth less than the going rate is the keenest to take it. The gap is visible if not cleanly attributable: unlisted businesses change hands at discounts of roughly 15–30% to comparable listed ones, though the literature credits illiquidity for as much of that as information.4

Knowing got cheap

The information a decision needs is what research and consulting firms have produced, and its cost was set by labor. A study is, at bottom, three kinds of people-hours: the research, analytics and production that assemble the facts and turn them into a document; the interviews that reach what no document holds; and the judgment and expertise that decide what it all means. All three were labor, and strategic advisory was a textbook case of cost disease — an activity with little productivity growth whose price, against everything that was getting cheaper, rose for decades.5

In the last year that cost fell in all three parts at once, for three different reasons.

Figure 3

Three parts of a study, three cost curves

Cost of one study, illustrative shapes.

Research, analytics and production went first, and went furthest. This is extraction from registers and filings, benchmarking and statistics, and the making of the deliverable itself — the slides, the models. It is now done to a professional standard by software agents that did not exist a year earlier, and it arrived in two steps. An autumn-2025 generation — GPT-5, Claude Sonnet 4.5, Gemini 3, Claude Opus 4.5 — could run long, multi-step work on its own rather than answer one prompt at a time; by late 2025 the task a model could finish unaided had reached about five hours. A 2026 generation — GPT-5.5, Claude Sonnet 5, Claude Opus 5 — is rated above the human expert on professional deliverables.6 In September 2025 an evaluation built by professionals averaging fourteen years’ experience found the best model’s work matched or beat their own on 47.6% of deliverables; by mid-2026 the best models sat well above that baseline.7 For this part of a study, the cost of producing one more is now near zero.

Interviews went second. Finding the right respondents can be automated, and the interviews run at scale by voice agents working to a structured guide. Voice models crossed into production quality in 2025 and improved sharply through 2026. The expert networks shipped limited AI-moderated interviewing within the year — although their overall operating model is otherwise still very traditional. The experts on the calls are still paid, so this cost falls rather than disappears; but the labor around them — sourcing, scheduling, sitting in — was most of what a study spent here.

Superior judgment did not get cheaper, but it got leveraged. The judgment that decides what the evidence means is still the scarce input, and still costs what an experienced person’s hour costs. What has moved is how many of those hours a study needs. The machine now does the middle of the work — the assembling, the first drafting, the routine analysis — so the expensive judgment is spent only where it is required, and a single senior can stand behind far more studies than before.

This unbundling only makes sense now, because production was what held the bundle together. What a consulting firm sold was four things at once: judgment, expertise, production, and the brand that stood behind the answer. They came together because they had to. A client who wanted the judgment without the pyramid beneath it had nowhere to buy the production alone; an experienced individual could not turn a question into a finished study by themselves; so the only real alternative to one large firm was another.

Firms exist for a reason: using the open market has its own costs — search, bargaining, coordination — and a firm is worth having when those are high.8 Lower them, and work once done inside the firm moves outside it. Once production costs almost nothing, the knot comes undone. Judgment and expertise can be bought separately, and a small team with a capable machine can deliver a whole study end to end. The production line was the consulting firm’s moat, and it is the part that has gone.

Built like software

Consulting as it has been practiced has no repeatable, written method. The way a study is actually done — what gets checked, what gets rejected, when an answer is good enough to stand behind — lives in an experienced practitioner’s mind, and is applied by hand, in review, one deck at a time. It is not written down as an exact procedure a machine could read, run and check against. Every engagement is, in effect, built as if from nothing.

The alternative is to run knowledge work the way software is built. The method is written down for every small part of the work successfully completed; a machine iteratively checks each study step and corollary against it; and every study leaves the method better than it found it. Because the method accumulates, the cost of serving the next study keeps falling, as the written practice compounds. This is a firm-specific fall, on top of the technological one described earlier.

Two assets accumulate this way, and they move different lines. The first is the method itself, and it acts on supply: it pushes the cost of one more study down, which lowers the price and sells more. The second is the body of practitioner knowledge that builds up across engagements — generalized, holding no one client’s findings — and it acts on demand: the studies get better and harder to substitute, so a buyer would pay more for one and switch away less. Both assets end in more studies sold; only the first also lowers the price, while the second raises what the work is worth. They pull the price opposite ways even as they agree on volume.

Figure 4

Two things compound, and they move different lines

The method acts on supply; the knowledge on demand.

The method moves supply

The knowledge moves demand

The organization that results is small. It is a lean team of unusually capable people who think in systems, not a pyramid of juniors run from above. The method works only when the people running it grasp it whole and keep it flexible — able to see where it applies and, above all, where it does not. That distinction is the whole game. In a 2023 field experiment at one large consultancy, more than seven hundred consultants were given a leading AI assistant: on tasks inside the tool’s competence they were markedly faster and better, but on a task that fell outside it they got it wrong far more often — and nothing in the tool told them which they faced.9 Leverage that can be trusted comes from the system, not the assistant.

The large firms have little reason to follow, and would have to become a different firm to do it. Their margin is made on billed juniors — the pyramid is not a side effect of the model, it is the model. A method needing few of them takes that margin apart. The fee does a second job: a high price signals quality, and buyers read it as a mark of the work behind it.10 Pass the saving through, and the firm reprices its own hours across everything else it sells. So the choice is to keep the saving as margin, or pass it through and weaken the signal that carries the rest of the business.

The second reason is organizational. The model asks everyone inside it to understand the whole system — the only way to know where the method holds and where a person must override or grow it. The machine, as we currently see it, is a delicate weapon. A firm of specialized roles and many juniors can copy pieces — bolt on an assistant, ship an AI interview product — but not become a place where each person carries the whole. Its archive does not rescue it: a deck records the output, not the decisions behind it, so simplistic replication of past work returns pastiche.

Priced closer to the truth

Lower cost means a lower price, more studies bought, and more surplus in total. How much of the fall in cost reaches the price is a choice the sellers make — hold the old price and keep the difference, or pass it through and win the volume.

To the extent it is passed through, the picture is textbook economics. Every study worth more to a buyer than it cost to make, yet left unbought because the price sat above the buyer’s line, was value gone missing — a deadweight loss, the triangle of trades that should have happened and did not. A lower price shrinks that triangle, and the surplus it creates accrues to buyers and to the smaller transactions priced out before.

Figure 5

Where the threshold sits, and where it moves

What a study is worth against what it costs, by investment size.

A study’s value still rises with the size of the transaction; the cost line is what changed — no longer a fixed high fee but a curve dropping toward the floor. The point where value crosses cost — the smallest transaction that can justify a study — slides down to a far smaller number. The band between the old threshold and the new one is the population that can now carry real evidence for the first time.

Transactions get priced closer to what the assets are worth, and more of them get done. When the cost of information is low, the buyer need no longer price in the average, and the lemons discount narrows on every transaction that can now afford a look, and liquidity in the market rises.

Figure 6

The fees paid today are one point on the curve

The market for diligence, with price on the vertical axis.

Put together, the market for private investments becomes more efficient — never completely, but a great deal more than it is now. As the cost of becoming informed falls, more transactions are done by a buyer who knows what they are buying, and prices carry more of what was there to be known. Learning what a business is really worth, and what could be done with it, changes which get capital and on what terms.11 This process of discovery is socially useful, and making it cheaper produces more of it.