Two competing strategies are bidding for the same post-SaaS opportunity. The first is services-as-software: build a company from inception that sells the work, prices against outcomes, and uses AI as the substrate that compresses the cost of delivery. The second is the AI-enabled roll-up: buy a fragmented services market at services multiples, deploy AI to compress costs, exit at software multiples. Both have committed capital. Both have early operational results. They are not the same bet, and they are not equally right.
The AI roll-up thesis and the capital behind it
The roll-up thesis is intuitive in the way that most category-error theses are intuitive. Services trade at one to two times revenue. Software trades at five to ten times. Buy services, deploy software, exit at the spread. General Catalyst earmarked $1.5B from an $8B fund for the strategy in late 2024 (the Creation Strategy). Thrive Holdings committed $1B+ in April 2025. Bessemer is co-investing. Lightspeed has roll-up plays in engineering services and healthcare. Total committed capital across the strategy now exceeds three billion dollars, and the portfolio results have been operationally interesting: Long Lake at $672M raised in HOA management, Crescendo at $500M valuation in call centres with 60-65% gross margins (roughly four times industry average), Crete at $300M+ revenue and the fastest-growing accounting firm of 2025, Eudia at $105M Series A in legal.
The financial counter: Concentrix, Genpact, and the multiple gap
The Fortune piece from June 2025 named the financial counter directly, and the data behind it is the kind that does not go away. Concentrix and Genpact have deployed gen-AI at more than a thousand customers each. EBITDA margins remain around ten percent. Multiples remain in the single digits. AI-transformed BPO firms (Concentrix, Genpact, Infosys) trade at five to twenty-three times EV/EBITDA. Pure software (Salesforce, ServiceNow, Workday) trades at twenty-two to ninety-two. Operational improvement, the data says, is not the same thing as business model transformation. The Fortune line is worth quoting in full: "AI roll-ups may still deliver returns, but not the kind VCs are underwriting. At best, tech-enabled PE: operationally heavy, valuation-capped, and unlikely to scale like software."
The mechanism behind this gap is not mysterious. The customer of a roll-up firm is buying services, not software. The price they are willing to pay is bounded by what services cost in the market, because the customer can always go back to a non-roll-up provider in the same category. AI-driven cost reduction therefore gets competed away into lower prices rather than captured as margin expansion. PolyAI ran exactly this experiment in 2019, in the call-centre vertical, and walked away from it. The Fortune piece notes the case explicitly. The bet was the same one the Creation Strategy is now making at scale, and the conclusion the experimenter reached six years ago was that the bet does not work as a financial thesis even when the operational story is real.
How services-as-software is structurally different
Services-as-software, by contrast, builds the company as a services firm from inception. There is no acquisition step. There is no acquired entity with a legacy P&L to defend. There is no reference price set by what the non-AI version of the same service costs. The customer is comparing the offering to the in-house team they do not have, not to a roll-up acquisition or a traditional services firm. The pricing reference shifts from cost-of-service to cost-of-labour-displaced, and the labour budget is roughly six times the software budget by Sequoia's framing or fifteen times by Bhargava's.
Operational proofs of services-as-software across verticals
The operational proofs of services-as-software are now visible across multiple verticals. Crosby in legal: $18B addressable contract review market, volume-based pricing per contract, fifty-eight-minute median turnaround. Eudia, also in legal: ARR went from $2M to $20M in twelve months after acquiring an alternative legal services provider and operating as an Arizona ABS firm; clients include DHL, Intuit, Cargill. Manifest OS in business immigration: $60M Series A at a $750M valuation, fixed-fee outcomes-based pricing. Crescendo in call centres: gross margins four times industry average. The pattern across these is not a coincidence. Services firm built from inception. Outcome pricing built into the model. Vertical depth that compounds over time. The shape repeats.
The single question that separates roll-ups from services-as-software
What separates these from a roll-up at the strategic level is the answer to a single question. Did the firm exist before the AI? In a roll-up, the answer is yes; the AI is a margin lever applied to a pre-existing business. In services-as-software, the answer is no; the firm exists because of the AI, and the entire engagement model and pricing structure were built around what the AI makes possible. The first kind of firm has a legacy customer base that was paying services rates for services delivery; pricing power is bounded by what those customers used to pay. The second kind of firm has customers who never had any version of the service before, who are comparing the offering against the cost of doing the work in-house or not doing it at all; pricing power is bounded by the labour budget being addressed, not by what the previous services vendor charged.
A reasonable counter to this is that the roll-up portfolio results are real. Long Lake's fundraise is real. Crete's revenue is real. The early operational outcomes are not fake, and the firms running this strategy are sophisticated. The distinction worth making at this point is between operational success and financial-thesis success. The portfolio numbers prove the operational story (you can buy a services firm and improve its margins with AI). They do not yet prove the financial story (those margin improvements translate to software multiples at exit). Concentrix, with the largest gen-AI deployment in the BPO category, is the natural-experiment refutation of the financial story. The roll-ups may still deliver returns, but the returns will look like tech-enabled PE returns, not venture returns. The capital committed to the strategy will get returned at the multiple of the underlying services category, not at the multiple of software. That may still be a good outcome for the LPs of those funds. It is a different outcome from the one the playbook implies.
The category-defining companies in this decade will be services-as-software firms, not roll-ups. The capital that recognises this earliest will end up holding the positions in those firms. The capital that bets on the multiple-arbitrage version of the thesis will end up holding companies that look like good operating businesses without the exit multiples that justify the venture model. The two paths diverge at the moment of incorporation. The strategy worth backing is the one that builds the firm with software economics from the start, not the one that buys the firm and tries to retrofit them.