Melting Ice — Summary

A fund model for a roll-up has a known shape. Entry multiple on a fragmented industry, a thesis about consolidation, a cost structure that improves with scale, an exit multiple above the entry. The model has been run thousands of times across thirty years, and across thirty years it has mostly worked. The variable that decides whether it works in the next ten is not in the model. The variable is the half-life of the arbitrage the industry sits on, and most deal teams have no field for it. This is about the single most expensive error available to capital in the transition, and the error is the most intuitive move there is. When arbitrages are being destroyed, the assets that sit on them get cheaper, and cheap assets in a fragmented industry are exactly what the roll-up playbook was built to buy. The instinct is sound by every rule capital has learned. It is also, in this environment, a way to spend real money acquiring a liability disguised, for the length of the diligence period, as an asset.

The standard move deserves respect. Acquire fragmented operators sitting on a durable arbitrage, consolidate them, professionalize management, strip redundant cost, raise prices where the position allows, exit the larger entity at a higher multiple. For thirty years it produced real returns, for a structural reason easy to forget because it was always true: the arbitrage was durable. The spread persisted across the hold period, so consolidating the operators consolidated a real and lasting spread. The durability was so reliable it dropped out of the analysis entirely. No one modeled the half-life of a dental practice’s local position or a regional insurer’s information advantage, because there was no half-life worth modeling. The spread was treated as bedrock, and for thirty years it behaved like bedrock, which is why a generation of deal models has no field for the possibility that it is not. That world is the one the audit ends.

The inversion is simple to state and hard to act on, because acting on it means refusing deals that look, by every traditional metric, attractive. If the arbitrage is dissolving, the roll-up consolidates a position the audit is about to erase. The fund acquires at multiples that price the spread as durable, holds through a period in which the spread closes, and discovers at exit that it has assembled a larger version of a thing that no longer has the value it paid for. It did everything the playbook required, and none of those operations addressed the actual problem, because the problem was that the spread itself was on a dissolution schedule. This is the purchase of melting ice. The asset is real at the moment of purchase, colder and harder and more solid-looking than the price suggests, which is exactly why it looks like a bargain. And it melts in the hold period, not because the fund managed it poorly but because melting is what that kind of ice does, and the rate of melt was the one variable the model did not contain. A vertical roll-up of a dissolving arbitrage is value destruction wearing the costume of value creation, and the costume lasts exactly as long as the diligence period.

The central read of the transition is the distinction between two kinds of ice. A melting arbitrage is one the audit is closing on a knowable schedule, information, complexity, access, the consumer edge of attention, the spread real today and diminishing structurally, not cyclically. A compounding arbitrage is one the audit makes steeper or cannot reach, scale and the instrument, which compound because the audit runs on them, and relationship, which survives because the audit cannot touch it. The single most important diligence question is which kind of ice the target is made of, and most deal models do not ask it, because for thirty years the answer was assumed. The assumption is now the risk.

Estimating the rate of melt is tractable, in four steps. First, identify which arbitrage class the target’s margin sits on; if it is a dissolving class, the default assumption is melt and the burden is on the deal to prove otherwise. Second, identify which architecture reaches that class and how far it has matured, because the classes fall as their responsible engines come due; information melts fast because its engine arrived first and loud, while a structured-data or risk-pricing position may have more time, though the time is shortening and the direction is fixed. Third, decompose the target into layers, because no real business is a single arbitrage; it is a stack, and the layers melt at different rates. Fourth, estimate what fraction of the current margin is attached to melting layers versus durable ones. That is the number that should drive the price, and it is almost never the number in the model. Two failure modes recur, both from skipping the decomposition. The blended multiple prices a target as a single durable thing when a melting layer carries most of the current margin, so the headline EBITDA is real today and structurally impaired across the hold, and the model never sees it because it never separated the layers. The misread engine assumes a margin is durable because the loud engine does not reach it, while a quieter engine, tabular or forecasting or anomaly-detection, is maturing toward exactly that layer on a schedule the deal team is not watching. The only protection is to ask which engine reaches this specific margin and how far along it is, rather than calibrating on the engine in the headlines.

The deliverable is a discipline, not an image. Before any roll-up, classify the target’s arbitrage stack by fate. Estimate each layer’s half-life against the maturation of its responsible engine. Price only the compounding and durable layers as going concerns. Treat the dissolving layers as melting: assign them whatever cash they will throw off before the melt completes, and pay nothing for persistence beyond it. The fund that does this survives, because it stops paying durable prices for temporary spreads. The fund that does not applies the thirty-year playbook to a world that changed underneath it and funds, with its own capital, the erosion of the assets it just bought. The playbook is not wrong. It is correct for a world in which arbitrages were durable, and that world is ending one class at a time. The strategy concentrates exactly in the fragmented, information-heavy, access-heavy industries that the audit reaches first, which means the playbook is most active precisely where the melt is most certain. A great deal of capital is likely paying for persistence that the very technology driving the thesis is scheduled to remove.