quality of growth

A buyout return can be taken apart. Across 3,830 buyout deals entered globally between 2010 and 2022, the average deal turned 1.0× of invested capital into 3.0×. The 2.0× created above cost came from four places, and they were not equal.

Read the middle two bars first. Revenue growth contributed 0.8×, four times what margin expansion contributed. The industry talks constantly about margin. The data says growth did the work.

Now read the two dashed bars. Leverage and multiple expansion together contributed 1.2×. They are the reason the model worked.

Both are going. Debt has fallen from 44% of entry multiples to 37%, and it costs more than it did. Entry multiples reached a record 11.8× in 2025, which leaves very little room for an exit richer than the entry. Four in five sponsors say they expect multiples to stay flat.

So 1.2× of contribution is being removed, and the 1.0× that remains has to cover it. That is not a marginal tightening. It is a different job.

The effect on the required growth rate is easy to compute and hard to look at.

Three of these inputs are observable at entry: the multiple paid, the leverage raised, the cost of the debt. The exit multiple is not observable. It is a choice. Once those four and the cash-conversion assumption are fixed, the required growth is arithmetic.

Which raises the question of where that growth is supposed to come from.

Required EBITDA growth is not one lever. It is two. Revenue can grow, margin can improve, or both. Which one carries the plan is decided at underwriting, and sponsors report deciding it the same way over and over.

Asked what margin improvement they assumed on their most recent flagship deal, a quarter said none at all. Another 49% said less than 300 basis points. Three percent underwrote a decline.

Read that again. Three quarters of the market is buying companies at record multiples on the assumption that margins will be roughly what they are today.

Which means revenue has to do the rest. The share it has to do is computable from four inputs every deal team already has in the model.

One condition is doing quiet work inside that arithmetic, and it is the difference between this and a growth plan. Growth Dependency assumes revenue arrives at or above today's margin. Revenue that dilutes margin does not move EBITDA, and EBITDA is what the business gets priced on at exit. Growth bought with discounting, through unprofitable channels, or with a cost to serve nobody modelled does not count here.

Growth that costs more than it earns is not growth. It is revenue.

Read the bar as the market. At the assumption half of it reports, Growth Dependency is 87%. Where no margin improvement is underwritten it is 100%, by definition. Where margin is assumed to fall it passes 100%, because revenue then has to cover the decline as well as the growth. More than three quarters of the market sits at 87% or higher.

Weighted across the shares sponsors actually describe, Growth Dependency is 84%.

Eighty-four cents of every dollar of required earnings growth is a bet on revenue. Made at entry. On evidence that is rarely examined with the rigour applied to the earnings.

A bet that size would normally attract a workstream. This one does not, and the reason is worth understanding before it is criticised.

Look at when margin improvement actually shows up across deals exited since 2019.

Roughly one percent of the total accrues in each of the early years. Four percent arrives in the penultimate year. Six percent in the final one. Half the improvement lands in the last two years, once the asset is being dressed for sale.

The obvious reading is that management gets serious when the exit approaches. There is a simpler mechanism.

A growth assumption that is wrong produces almost no signal for a long time. Pipeline looks adequate. Bookings land within tolerance. Each quarter's variance is small enough to be attributed to timing, a slipped renewal, a slow start by a new rep. Nothing in the reporting pack says the engine cannot get there. It says the engine is a little behind.

Then year four arrives and the gap is no longer deniable. By then the runway is measured in months, and the only levers that work in months are the ones that cost margin.

This is not a failure of attention. It is a feedback loop that runs longer than the patience of the people inside it.

The commercial evidence agrees. Among more than eleven hundred commercial leaders, 86% were confident of hitting their 2025 growth target. Fifty-eight percent did. The share falling short of revenue targets rose from 33% to 42% in two years, and 91% are confident about 2026.

Optimism is not in short supply. Evidence is.

Sponsors were asked why deals they pursued in 2025 did not close. Seller valuation expectations, 36%. Diligence red flags where fundamentals such as earnings quality or customer churn did not hold up, 33%. Competitive intensity, 25%. Macro uncertainty, 6%.

Growth capability appears nowhere.

Deals die because the earnings are not what they seemed, or because the price is wrong. They do not die because the buyer concluded the company could not produce the growth the model required. That absence does not mean the question is always answered well. It means it is rarely asked in a form that could return a no.

The cause is structural, not cultural. Every workstream touches the growth number. Each one stops just short of it.

Underwriting value creation during diligence is becoming standard practice. Saying so and doing it with rigour are different things. Sizing a prize is not the same as pricing the capability required to capture it, in dollars and in months, against the specific number in the model.

The same gap is now opening in the technology being deployed to close it.

Every sponsor is deploying AI somewhere in the portfolio. The question is where it is aimed.

Revenue growth produced four fifths of operating value creation. Ten percent of sponsors expect AI to produce revenue growth next year. Forty-six percent expect cost savings.

The lever that has always carried the return is the one AI is least often pointed at.

There is a reason, and it is not stupidity. Cost is legible. A headcount reduction can be counted this quarter and defended in a board pack. Revenue capability takes longer to build, resists attribution, and gets tested by a market rather than by a spreadsheet. Under a five-year hold and quarterly reporting, the legible thing wins.

Which is precisely the trap the previous section described, arriving one cycle earlier. Around 60% of companies have realised no measurable value from AI at all, and roughly 70% of the return comes from redesigning how work is done rather than from the tools themselves. A hold can end before a cost programme pays back.

Pointed at the commercial engine, the same technology does something different. It shortens the ramp on new sellers. It raises the hit rate on the accounts that matter. It compresses the interval between a pricing decision and its effect on the book. Those are revenue capabilities, they are underwritable at entry, and they are what the return now depends on.

Growth misses do not usually announce themselves as losses. They show up as a deal that worked, slightly, and a fund that did not.

Take the deal from the first section. Record entry multiple, 37% debt, flat exit, underwriting 12.2% annual EBITDA growth for a 2.5× over five years. Now deliver 9% instead of 12.2%, which is roughly what a supported case looks like when the engine is honestly assessed.

The deal still works. It returns 2.07×, which is real money on real capital. But the gross IRR falls from 20.1% to 15.7%, and limited partners increasingly want north of 20% net from a buyout fund before they will commit to the next one (Preqin, ILPA).

So the sponsor does the rational thing and holds on to fix it. Two more years of work, and the multiple recovers to 2.71×.

The return does not. Seven years at 2.71× is a 15.3% IRR. The fund did more work, created more value, and earned less than it would have by exiting a worse deal on time.

This is what 32,000 unsold companies and four years of suppressed distributions look like from inside a single deal. Not disaster. Just a return that quietly stopped being top quartile, on an asset nobody would describe as a failure.

The cost of testing the assumption before signing is a fraction of one percent of equity. The cost of testing it in year four is the difference between those two numbers.

A company can grow quickly while its economics deteriorate, its system depends on heroics, or its capability walks out with a founder. Speed is not quality, and neither is volume. A business that is bought and sold on a multiple of EBITDA needs growth that reaches EBITDA. Quality is what survives contact with an owner who needs it to repeat, profitably, on schedule.

Percentages are where accountability hides

Every material lever should be translated into an equation. The equation does not prove the forecast. It makes the assumptions visible enough to argue with.

Suppose a plan requires $12 million of incremental new-logo revenue. At a $200,000 average first-year value with 75% realising in-year, that is 80 wins. At a 25% win rate, 320 qualified opportunities. At 40 opportunities per productive seller per year, eight fully productive sellers.

The question changes immediately. It is no longer whether the market can support $12 million. It is whether this company can recruit, ramp, enable, manage and feed eight productive sellers early enough for the revenue to land inside the hold. That question has an answer, and the answer is knowable before signing.

Evidence gets a ladder

Assertion, anecdote, point-in-time, longitudinal, triangulated. Each level carries a permitted use. No material upside enters the supported case on assertion or anecdote alone. A plausible but unproven lever is not rejected. It sits behind conditions, or in upside, until the evidence changes.

What it looks like applied

Consider an illustrative composite. A $72 million technical-services platform, five acquisitions deep, growing 15% a year, underwritten at 12.5% for five years. Four points of that growth came from acquisitions, two from catch-up pricing, one from project timing. The sustainable organic baseline was 8%.

Six growth levers were identified, summing to $18.9 million of year-five incremental revenue. The distance between the current engine and the underwritten case was $19.0 million. Side by side, the plan appears to close the gap almost exactly. That coincidence is the trap.

The levers overlapped. Price affected volume. Cross-sell and new-logo growth competed for the same enablement and delivery capacity. New-market entry needed senior sellers already counted in the core build. After evidence, timing, overlap and the binding constraint, $7.3 million survived into the conditional case, and the verdict was not that 12.5% was impossible. It was that the investor had not yet earned the right to treat it as the base case.

All of it resolves into one picture: the growth management reported, the growth the model requires, and exactly where the difference has to come from.

Read left to right, it answers four questions in order. What did the reported growth consist of? What survives normalisation? How much does today's engine support without heroics? And how far is that from the number in the model?

The dashed bar is the only one that matters commercially. It is not a criticism of the thesis. It is the part of the thesis that has to be funded, sequenced and proven rather than assumed, and it is the part that decides whether year four is a victory lap or a scramble.

A score compresses judgment into a number nobody can argue with. A bridge exposes every step, so an investment committee can disagree with one line without discarding the analysis. That is the difference between a rating and an underwriting.