Nathan Benaich is the founding father of Air Road Capital and creator of the State of AI Report. Nikola Mrkšić is the CEO of PolyAI.
Throughout the expertise investing world, buyers are scaling their bets on a seductive thesis: Generative AI will remodel low-margin service companies into high-margin software program firms. A number of well-known platform enterprise corporations have dedicated billions to this technique and have begun to make their bets. Right here’s how the thesis goes:
First, purchase conventional enterprise course of outsourcing (BPO) firms akin to name facilities and accounting corporations at modest valuations of 1x income. These companies usually function at 10-15% EBITDA (earnings earlier than curiosity, taxes, depreciation, and amortization) margins, weighed down by armies of human staff performing repetitive duties, and automation faces the best structural resistance.
Second, deploy generative AI to automate core workflows, minimize headcount, and increase EBITDA margins to 40% or extra. What as soon as required a whole lot of accountants or name heart brokers can now be achieved by a handful of individuals managing AI techniques.
Third, exit the newly minted AI-enabled companies firm at software program multiples as a result of patrons and public markets acknowledge you’ve reworked a human-heavy service enterprise right into a scalable AI enterprise. The place conventional BPOs commerce at 6x EBITDA, software program firms command 20x or extra.
On paper, it’s good arbitrage. In follow, it’s a mirage. It rests on a basic class error: complicated operational enchancment with enterprise mannequin transformation. Sure, AI could make workflows extra environment friendly. No, that doesn’t flip a companies firm right into a software program firm.
Certainly, 5 years in the past, a now notable AI firm ran this actual experiment, and walked away. Its findings ought to function a warning to right now’s believers. Let’s dig in.
The 69x valuation canyon
Essentially the most damning proof in opposition to the AI rollup thesis hides in plain sight on public markets. Immediately’s “AI-transformed” BPO corporations which have invested closely in automation—amongst them Concentrix, Genpact, and Infosys—commerce at 5-23x EV/EBITDA (enterprise worth to EBITDA). Their pure software program counterparts, akin to Salesforce, ServiceNow, and Workday, command valuations of 22-92x EV/EBITDA. Here’s a chart to inform the story:
That’s not a niche that may be bridged with press releases about OpenAI, Anthropic, or Gemini partnerships. It’s a basic distinction in how markets worth human-dependent companies versus true software program platforms.
Contemplate Concentrix, usually cited as a BPO transformation success story. Regardless of a serious push in launching their gen-AI merchandise in 2024 and now having deployments at over 1,000 prospects, the corporate’s EV/EBITDA a number of stays caught within the low single digits, and its EBITDA margin remains to be hovering round 10%. The market’s message is evident: Automating workflows doesn’t change your basic enterprise mannequin.
The PolyAI prophecy
In 2019 PolyAI, the main conversational AI firm, spent six months exploring whether or not to amass incumbent human-driven contact facilities to speed up its progress. After analyzing the chance by visiting over 10 contact facilities, constructing relationships with three main BPOs, and hiring business advisors, the reply was a transparent no.
“Enterprise Course of Outsourcing corporations usually are not trusted to innovate, not rewarded for innovating, and never allowed to innovate,” learn its board deck.
The structural limitations it recognized stay unchanged right now:
- The phantasm of management: Shopping for a BPO doesn’t imply proudly owning the enterprise you’re supporting. You’re merely renting the proper to provide labor on the consumer’s phrases. Tech stacks, processes, and approvals stay firmly within the consumer’s palms. AI deployments nonetheless require their permission, integration, and oversight. You’re not in management; you’re a replaceable vendor.
- The pricing lure: Most service companies invoice by the hour. Effectivity enhancements that cut back billable hours instantly cannibalize income. As PolyAI found, BPOs promise innovation to win contracts, then revert to maximizing billable hours to guard margins. It’s a enterprise mannequin essentially at odds with automation.
- Zero switching prices: The place 10-year service contracts had been as soon as the norm, it’s now more and more frequent to see three-year phrases or much less. This reduces the power to recoup up-front AI investments, notably when there’s little consumer lock-in, no community results, and no moat.
PolyAI selected to stay a software program firm, partnering with BPOs somewhat than buying them. Immediately, it’s valued at over $500 million with prospects like PG&E, Marriott, and FedEx. In the meantime, the BPOs they thought of shopping for nonetheless commerce at single-digit multiples.
Why this time isn’t completely different
Right here’s what buyers are lacking: Companies companies aren’t inefficient accidentally. They’re inefficient by design. The inefficiency is the product. Purchasers pay for flexibility, customization, and somebody in charge when issues go flawed.
Automating away the human doesn’t simply cut back prices, it essentially adjustments what you’re promoting. BPO expertise functionality has by no means been the constraint. And purchasers who wished software program would have already purchased software program.
Essentially the most profitable companies corporations perceive this. They use AI to reinforce their people, not change them. They keep margins by means of pricing energy and relationships, not by means of headcount discount. Finally, they nonetheless commerce at companies multiples as a result of that’s what they’re.
The teachings of historical past
The AI rollup thesis represents a well-known sample in expertise investing: the conflation of technological functionality with enterprise mannequin transformation. We’ve seen this film earlier than.
Within the early 2000s, believers thought e-commerce would remodel retail margins. Amazon proved them proper by constructing a local digital retailer, not by buying and reworking Sears or Barnes & Noble. Within the 2010s, buyers believed software program would eat conventional industries. The winners constructed new software-native companies somewhat than retrofit previous ones.
The identical lesson applies right now, however with a narrower scope. AI might effectively remodel some corners {of professional} companies, particularly when present corporations are pushed to undertake new instruments by personal fairness homeowners with clear management and incentives. We’ve seen this in sectors like well being care and monetary companies, the place PE corporations have pushed adoption of AI-driven tooling. However that is completely different from the AI rollup thesis that VCs are chasing—one which assumes low-margin, labor-heavy service companies might be was software-like platforms just by embedding AI. For these corporations, transformation gained’t come from proudly owning the service layer. It’s going to come from new, AI-native firms with essentially completely different economics.
The underside line: Personal the software program, not the service
The AI rollup thesis is enterprise capital’s try and arbitrage the a number of hole between companies and software program. However that hole exists for a motive. Companies companies, even extremely automated ones, face completely different constraints, completely different economics, and completely different buyer relationships than software program firms.
PolyAI noticed it in 2019. Public markets see it now. The AI revolution is actual. The chance to enhance companies companies with AI is actual. The concept that this enchancment transforms them into software program firms? It’s unlikely to be actual right now, simply because it wasn’t in 2019.
AI rollups should still ship returns, however not the sort VCs are underwriting. At greatest, they’re tech‑enabled personal fairness: operationally heavy, valuation‑capped, and unlikely to scale like software program.
The opinions expressed in Fortune.com commentary items are solely the views of their authors and don’t essentially mirror the opinions and beliefs of Fortune.
Learn extra:
