AI Value Creation for PE Multi-Site Roll-Ups

The roll-up math: AI only matters if it scales sublinearly with location count

Every roll-up has the same secret: the deal model assumes back-office cost grows slower than location count. Add a 20th clinic and you should not add a 20th full front desk. AI is the first tool that makes that assumption real instead of aspirational — but only if you have enough similar sites to amortize the build.

That’s the load-bearing caveat. Generalist portfolios face harder math because every deployment is custom and the unit economics fall apart; value creation needs 5–8+ similar companies to build a repeatable playbook (Withum). A roll-up is exactly that structure — one vertical, dozens of near-identical sites.

The same source is brutal about where AI doesn’t pay: AP automation, reporting dashboards, and customer intake move EBITDA 1–2% at best. If a vendor is selling you a dashboard, they’re selling AI theater. The money is in labor and revenue capture, not pretty charts.

Play #1 — AI call-handling: the cheapest thing a competitor beats you on

Multi-site service businesses leak revenue at the phone, and the leak compounds with every location. In dental, practices miss 30–40% of inbound calls, and 47% of appointment requests arrive after hours — each missed call worth ~$850 in first-year patient revenue, putting a 50-location DSO’s leak at $500K–$1M a year (TensorLinks).

The clearest proof this is real money: Avoca, an AI voice-agent startup for HVAC/plumbing/roofing, raised $125M+ and hit a $1B valuation in 2026 backed by Kleiner Perkins, Meritech, and General Catalyst, serving 800+ contractors (Fortune). The EBITDA mechanism is replacing a centralized call center: traditional DSO call centers run $500K–$2M/year; AI front-desk coverage runs roughly $799/month/location (TensorLinks — treat the single-vendor case number as directional).

Play #2 — Revenue-cycle automation: the back office that should scale sublinearly

For healthcare roll-ups, billing and collections are where headcount silently balloons site-by-site. This is the textbook sublinear play: one centralized, AI-assisted RCM function serving all locations instead of a biller per clinic.

The denial problem is worsening — providers reporting 10%+ denial rates rose from 30% (2022) to 41% (2025) (EY). The headline worth repeating to an IC: McKinsey estimates AI could cut cost-to-collect by 30–60% (via Notable Health). The skeptic’s flag: only 14% of providers use AI to reduce denials today — that gap is the opportunity and the warning that execution, not the tool, is the hard part.

Play #3 — No-show reduction & demand fill: small percentage, large dollars

No-shows are pure margin loss because the cost of the slot is already sunk. In multi-site dental, new-patient no-show rates run 30–40%; automated reminder cadences cut no-shows 25–40% versus manual calls (Weave). In home services the analog is dispatch: AI scheduling reports 10–20% job-capacity gains without added headcount (Profitability Partners).

The connective tissue is the cleanest line from AI to multiple: in a healthcare-services company running 55% labor-to-revenue, a 5% labor-cost cut drops 2.75 points to EBITDA — roughly $1.1M on a $40M company, ~$9M of enterprise value at 8x (Withum).

What kills these roll-ups

• AI theater. Dashboards, AP automation, intake portals — 1–2% EBITDA at best. If it doesn’t touch labor or revenue capture, it’s a science project.
• Too few similar sites. Below ~5–8 comparable locations, every deployment is custom and the unit economics collapse.
• Integration debt. AI call-handling is only as good as its hook into the practice/field-service system (Dentrix, OpenDental, ServiceTitan).
• The macro can swamp the savings. Veterinary roll-ups show this — KKR’s PetVet was marked at 88% of par amid declining visit volumes (Octus).
• The adoption gap. 74% of contractors call AI essential but only 25% actually use it. Buying the tool is not deploying it.