← Blog

The Service-Business Software Multiple: How AI Orchestration Rerates a Roll-Up

There is a thesis hiding inside a lot of service-business roll-ups, and most of the people running them have not named it yet.

The standard roll-up is arithmetic. You buy a fragmented sector, one of pest control or HVAC or commercial landscaping or accounting, at a services multiple. You tuck in smaller operators at lower multiples, you bolt them onto a shared back office, and you sell the consolidated entity at the same services multiple you started with, just bigger. The value comes from the spread between what you paid and what you sell at, plus whatever cost you stripped out. It works. It has worked for decades. But the multiple does not move.

What moves the multiple is changing what the buyer thinks they are buying.

Two multiples, one business

Hold two numbers in your head. A service business trades at a services multiple, call it the high single digits. A software business trades at a software multiple, well into the teens, sometimes higher. Same revenue, very different price, because buyers pay for different things. They pay a services multiple for labor that has to be hired, trained, and managed. They pay a software multiple for a system that scales without proportional headcount, that has gross margins a labor business cannot touch, and that is hard for a competitor to copy.

The interesting question is what would have to be true for a service business to earn something closer to the second number.

The answer is not to stop doing the service. The skilled human work is the point. A plumber, an auditor, a field technician, a nurse, that is the value the customer pays for, and AI is nowhere near replacing the judgment in those roles. The answer is to put AI and software in the orchestration layer that surrounds the human, so the business starts to look and behave like a platform that happens to deploy skilled people, rather than a labor pool that happens to have software.

What goes in the orchestration layer

The orchestration layer is everything between the customer’s problem and the skilled person solving it.

It is the intake that captures and structures the job. The scheduling and routing that puts the right person in front of the right work with the least wasted time. The pricing and quoting that used to live in a senior estimator’s head. The quality checks, the follow-up, the parts ordering, the invoicing, the warranty tracking. The knowledge that lets a two-year technician perform like a ten-year one because the system is feeding them the next right step.

In a typical service business, all of that is done by people, in tools that do not talk to each other, differently in every branch and every acquired company. That is exactly the work AI is now good at. Not the plumbing. The orchestration of the plumbing.

When you move that layer from people and spreadsheets to software and AI, three things happen. Utilization goes up because the system squeezes the slack out of scheduling. Margins go up because the orchestration cost stops scaling one-to-one with revenue. And the business becomes repeatable, because the way work gets done lives in the system rather than in the tenure of whoever has been there longest.

Now every tuck-in stops being arithmetic and becomes a transformation. You are not just adding the acquired company’s revenue to yours. You are dropping its operations onto your orchestration layer, lifting its utilization and margin to your standard, and making it run the way the platform runs. The acquisition gets better the moment it joins, not just bigger. That is the story that earns a different multiple, and it is the story I wrote about in the context of a single sector in the garage door roll-up data bet.

Why the multiple is real and not a slide

A buyer is not going to pay a software multiple because you put the word AI in the CIM. They will pay it if they believe the orchestration is durable, transferable, and verifiable. Durable means it keeps working after you leave. Transferable means it survives the next tuck-in. Verifiable means their diligence team can confirm it actually does what you say.

All three of those run through the data layer. This is the part most of the roll-up thesis skips, and it is the part that decides whether the rerating is real.

The orchestration only works if the data underneath it is standardized and trustworthy across every entity you have acquired. An AI scheduler cannot route work across twelve branches if each branch defines a job, a customer, a technician, and a service line differently. A pricing model cannot learn from history if the history is six incompatible systems that were never reconciled. A quality system cannot flag a problem if half the acquired companies record outcomes in free text and the other half do not record them at all.

The reason this matters more in a roll-up than in a single company is that every acquisition arrives with its own data, its own definitions, its own systems, and its own quiet mess. The whole premise of orchestration is one layer running across all of them. If the data is not standardized, you do not have one platform. You have a holding company with a logo on top of twelve businesses that still run their own way, and a buyer’s diligence team will find that out in the first fortnight.

The two questions a buyer will ask

When you take this story to market, the buyer will test the thesis with two questions, and they are the same two questions I have watched PE buyers ask of any acquisitive business.

Are the acquisitions actually integrated, or just owned? They mean operationally and on the data, not legally. They are asking whether the orchestration layer genuinely runs the acquired entities or whether each one is still an island.

Can you produce a single consolidated view across every entity? Margin by service line across the whole platform. Utilization by branch on one definition. Customer retention measured the same way everywhere. If you can produce that quickly and it reconciles, you have proven the platform is real. If it takes three people a week and the numbers do not agree, you have proven the opposite, and the multiple collapses back to services no matter what the AI demo looked like.

This is the bifurcation playing out inside a single deal. The companies that can answer those two questions cleanly trade like platforms. The ones that cannot trade like labor businesses, and the gap between the two is not gradual. I laid out that dynamic in why some companies exit at a premium, and it applies with more force here, because the entire premium in this strategy is staked on integration the buyer can verify.

Sequence the data layer first, not last

The mistake is to treat the data work as cleanup before the exit. By then it is a fire drill, and the buyer can see the seams.

The sequence that works runs the other way. Standardize the data layer as part of the integration playbook, before you scale the AI on top of it. Every tuck-in gets brought onto the same definitions, the same core systems, the same data standards in the first ninety days, the same way you would standardize a chart of accounts. That standardization is what lets the orchestration layer actually span the platform. It is also what lets you prove, at any moment, that the consolidated view is real.

Do it in that order and the data foundation is doing two jobs at once. During the hold, it is the thing that makes the AI orchestration work, which is what drives the operational gains the whole thesis depends on. At exit, it is the thing that lets a buyer verify the platform in days instead of doubting it for weeks. Same investment, two returns. This is the argument behind treating data as the edge rather than the afterthought, which I made in why record multiples make data your edge.

If you want a structured read on whether your data layer is actually carrying that weight across the entities you have acquired, the VCP Data Score is built to surface exactly the gaps a buyer’s diligence team would find.

The thesis, stated plainly

Keep the skilled human doing the human work. Put AI and software in the orchestration layer around them. Standardize the data across every entity so the orchestration actually spans the platform and a buyer can verify it. Do that, and a business that should trade at a services multiple can start to earn something closer to a software multiple, and every tuck-in becomes a transformation rather than a line in a spreadsheet.

The AI gets the attention because it is the visible part. But the AI is only as good as the data it orchestrates, and in a roll-up the data is the hard part, because it arrives broken, twelve different ways, with every deal you close.

The orchestration layer is the strategy. The data foundation is what makes the multiple real.

For a weekly brief on PE exit dynamics, data readiness, and portfolio company value creation, subscribe to Inside the Data Room.