How Operating Teams Make AI Roll-Ups Work

Picture a platform that has just completed its fifth acquisition of a regional accounting firm. The investment thesis assumes AI will lift margins across the group. On day one, the operating team inherits five practice management systems, five versions of the tax review process, and five approaches to billing clients. Some partners price returns by the hour, some by fixed fee, and at least one firm has never formally tracked turnaround time.
In late October, Thrive Holdings agreed to acquire ADT in a $15.3 billion all-cash transaction, adding a household name to a portfolio that already includes more than 75 businesses across accounting, IT services, and the built environment. A November market map from Dealroom counts 33 AI-native roll-ups across 14 industries, three of which are take-privates of listed companies. Venture firms, holding companies, and AI labs are all competing for services businesses on the premise that AI changes their economics.
Which workflow can the team improve, who owns it, and what result would justify doing it again at the next firm? That question anchored a recent episode of Private Capital Insiders, hosted by Frank Scarpelli, with guests Sohail and Raheel Jooma, co-founders of BowMerge, an AI services firm focused on value creation and post-merger integration. In their view, the AI roll-up remains a buy-and-build at its core. What changes is where the returns are expected to come from, and that shift puts far more of the burden on the people who run the businesses after the deal closes.
Pick a Workflow With a Clear Operating Payoff
The best place to start is a recurring service with measurable volume. In an accounting platform, tax preparation fits well. The team can count returns completed, track turnaround from document receipt to filing, measure how often work comes back for correction, and calculate the fully loaded cost of each completed engagement. Those four numbers form a baseline against which every later decision can be measured.
Sohail Jooma draws a useful distinction between two ways of applying AI to that workflow. The hybrid approach automates steps inside the process that already exists. Data entry from client documents gets faster, while the sequence of preparer, reviewer, and partner sign-off stays intact. The pure approach starts over. "The pure approach is more, okay, what is done today? What are the first principles about running this sort of firm?" he explained. "And where can we identify areas that we can rebuild what work needs to get done so that it becomes fully transformational?"
For the people doing the work, the two paths feel very different. Under the hybrid model, a preparer's day gets lighter. Under a full redesign, the preparer role itself may change, with junior staff spending more of their time on exceptions and client questions and less on transcription. Both models depend on clarity about where professional judgment sits. Jooma was direct about the limits: "Where AI can't function for you is on the judgment side of saying, here's all the data that we've provided for you. It's up to the human to understand what the decision should be made based on that information."
Integrate the Data and Process the Pilot Depends On
Once the workflow is chosen, the integration question becomes concrete. A tax preparation pilot depends on client records, engagement histories, document intake, and a shared definition of what a completed return looks like. If each acquired firm runs a different CRM and defines a finished return differently, the AI is working from five incompatible foundations.
Raheel Jooma framed it in physical terms. "It's similar to the foundation of a house," he said. "If one part of your house has wood all over it and the other side has concrete as a foundation, it's gonna get uneven pretty fast." As he described it, the purpose of integration is to make sure the platform can look across everything it has bought and understand what is happening before adding anything on top.
That observation lines up with recent survey data. Accordion's PE AI Adoption Benchmark found that technology operating partners rank data infrastructure as the top barrier to scaling AI, cited by 71 percent of respondents, and that only 6 percent of portfolio companies are ready for AI deployment within their first 100 days of ownership.
What the team needs is the specific records and process definitions the chosen workflow depends on, made consistent enough to compare across firms, with a named owner accountable for that work. Who owns the data model on day one is a question the platform has to settle early. The Joomas noted that the right answer varies with what the sponsor is trying to achieve. A platform built to capture combined margins needs a common data model far sooner than a holding structure that keeps its businesses largely independent.
The deal team still has an important role here after the handoff. Diligence typically surfaces known data gaps, system overlaps, and the workflow assumptions behind the margin case. Those findings, along with a realistic integration budget, belong in the operating plan from the start. An investment memo that assumes AI-driven margin expansion without funding the integration work beneath it leaves the operating team to discover the shortfall on its own.
Translate Operating Improvements Into P&L Results
The portfolio CEO and CFO need a way to evaluate it that goes beyond hours saved. Compare cost per engagement, turnaround, rework rates, and service quality against the pre-pilot baseline, then trace them through to revenue and margin. This is where services economics can work against the thesis. Under hourly billing, faster work can mean less billable revenue. A tax return that once took ten hours and now takes four may simply generate less income unless something else changes. The team has to test where the gain will actually land. It may come through higher volume handled by the same staff, lower delivery costs on fixed-fee work, or a new pricing model that clients are willing to accept.
The Joomas see outcome-based pricing as the eventual answer, though it depends on evidence. Hourly billing offers no assurance of how long a job will take, while a proven workflow can promise a result. "If you could demonstrate the proof, then you have much more leverage in terms of how to price your outcome," Sohail Jooma said, "because you can guarantee something that an hourly billing couldn't really guarantee."
Lead the Rollout Across Acquired Businesses
Moving from pilot to platform is as much a leadership challenge as a technical one, and the roles need to be explicit. The operating partner coordinates priorities and resources across the platform. Portfolio company executives own adoption and results within their firms. Frontline staff, particularly long-tenured preparers and reviewers, show how the work actually gets done, including the exceptions that never appear in a process map. Product and engineering talent, increasingly embedded at the portfolio level, builds around that reality.
The hiring question behind this structure is whether a single leader can combine the instincts of a services operator, a technologist, and an integrator. Sohail Jooma was candid that such a person "would be a unicorn," adding that a strong leader starts with "the ability to talk less and listen more."
That emphasis on listening reflects the circumstances of most roll-ups. An executive arriving from outside will know less about the business than the employees who have spent decades inside it, or the founder who built it. Jooma also stressed transparency about where the platform is headed and empathy for employees who have just learned their firm was acquired. In his view, basic AI fluency is becoming table stakes, with particular emphasis on understanding where the tools help and where they fall short.
Prove the Result Can Travel to the Next Acquisition
The decision to scale should rest on a second test. Can another firm on the platform adopt the workflow, maintain service quality, and deliver a comparable economic gain once implementation costs are counted? Wherever the playbook had to change, that change should be documented, along with who at the local level can run it without central support.
This discipline appears to be uncommon. In the same Accordion benchmark, 41 percent of operating partners said their firms are deploying AI across multiple portfolio companies without an operational playbook. In a roll-up, that gap compounds with every acquisition, because each new firm inherits whatever was or was not codified at the last one.
The evidence also feeds directly back to the deal team. A workflow that has been reproduced at two or three firms gives the sponsor a more reliable basis for underwriting the next acquisition, including a realistic estimate of integration cost and time to value. At exit, it gives a future buyer something concrete to diligence: what the platform has already replicated, what it cost, and what it would take to do it again at the next target.
Taken together, the conversation points to a short operating checklist. Underwrite the cash flow and treat the AI story as upside. Prove the result on one workflow before scaling it. Fund integration first, since that is where these deals are most often won or lost. And test the exit case well before the platform reaches the exit.
