Nathan Benaich is the founder of Air Street Capital and author of AI Reports. Nikola Mrkšić is CEO of Polyai.
Throughout the world of technology investment, investors are betting on fascinating papers. Generated AI transforms low margin service businesses into high margin software companies. Several well-known platform ventures have committed billions to this strategy and have betted. Here’s how to get to the paper:
First, we acquire traditional business process outsourcing (BPO) companies such as call centres and accounting companies with a modest valuation of 1x revenue. These businesses typically operate at 10-15% EBITDA (revenue before interest, tax, depreciation, and amortization), overwhelmed by the military of humans performing repetitive tasks, facing the greatest structural resistance of automation.
Next, deploy generated AI to automate core workflows, reduce staffing, and extend EBITDA margins to over 40%. What once needed hundreds of accountants or call center agents could now be done by a small number of people managing AI systems.
Third, we are leaving the newly created AI-enabled service company with software multiples, as we recognize that buyers and public markets have transformed human-heavy service businesses into scalable AI businesses. If a traditional BPO trades with 6x EBITDA, the software company commands more than 20x.
On paper, it’s a great ruling. In reality, it’s a miracle. It is based on basic category errors: disrupt business model transformations and operational improvements. Yes, AI can make your workflow more efficient. No, it doesn’t turn a service company into a software company.
In fact, five years ago, a now remarkable AI company ran this exact experiment and walked away. That discovery should serve as a warning to today’s followers. Let’s dig deeper.
69X rating Canyon
The worst evidence for AI rollup papers is clearly hidden in the public market. Traded by “AI-converted” BPO companies (Constix, GenPact, and Infosys), which invest heavily in automation, at 5-23X EV/EBITDA (EBITDA’s Enterprise Value). The pure software counterparts such as Salesforce, ServiceNow, and Workday are command ratings for 22-92x EV/EBITDA. Here is the chart to tell the story:
That’s not a gap that can be built in a press release on Openai, humanity, or Gemini’s partnership. This is the fundamental difference in how the market evaluates human-dependent businesses and true software platforms.
Consider Concerix, which is often cited as a success story for BPO conversion. They launched their Gen-AI products in 2024, and now they are Expanding With over 1,000 customers, the company’s EV/EBITDA multiples remain in a low single digit, and the EBITDA margin is still around 10% hovering. The market’s message is clear. Automating workflows does not change the basic business model.
Polyey’s prophecy
In 2019, PolyAI, a leading conversational AI company, spent six months investigating whether to acquire an existing human-driven contact center to accelerate growth. After visiting over 10 contact centers, building relationships with three major BPOs and analysing opportunities by hiring industry advisors, the answer was clear.
“Business process outsourcing companies are not trusted to innovate, not rewarded for innovation, not allowed to innovate,” reads the board deck.
The structural barriers it has identified remain the same today:
- The Illusion of Control: Buying BPO does not mean owning the business you support. You simply rent the right to supply labor on client terms. The high-tech stack, processes and approval remain firmly in the hands of the client. AI deployments still require permission, integration and monitoring. You are not controlled. You are a replaceable vendor.
- Price Trap: Most service businesses charge hourly. Improved efficiency that reduces billable times cannibalize revenue directly. As Polyai discovered, BPOS promises innovation to win contracts, and protects margins by maximizing billable times. This is a business model that is fundamentally at odds with automation.
- Zero switching cost: If a 10-year service contract was once the norm, it is becoming more and more common to see a term of office of less than three years. This reduces the ability to recover UPRONT AI investments, especially when there is little client lock-in, no network effects, and no moats.
Polyai chose to maintain the software company and partnered with BPO rather than obtaining it. It’s like today Important It has exceeded $500 million, along with customers such as PG&E, Marriott and FedEx. Meanwhile, BPOs are trading in single-digit multiples as they consider purchasing.
Why is there no difference this time?
Here are the things investors lack: The service business is no coincidence. They are inefficient by design. Inefficiency is the product. Clients pay for flexibility, customization, and someone who takes responsibility when things go wrong.
Not only does it automate humans, it also reduces costs, but it fundamentally changes what you are selling. BPO technology features were not a constraint. Clients who wanted the software would have already purchased the software.
The most successful service companies understand this. They don’t use AI to reinforce humans and replace them. They maintain margins through pricing capabilities and relationships, rather than reducing staff cuts. Because ultimately they are still trading at multiples of the service.
Lessons of History
AI rollup papers represent familiar patterns of technology investment. This is a confusion between technical capabilities and business model transformation. I’ve seen this movie before.
In the early 2000s, followers thought that e-commerce would change retail margins. Amazon has proven them right by building a native digital retailer rather than getting and converting Sears or Barnes & Noble. In the 2010s, investors believed that software would eat traditional industries. Rather than remodeling the old business, the winners built a new software native business.
The same lesson applies today, but the scope is narrower. AI could well translate the corner of professional services, especially when existing companies are being encouraged to adopt new tools by private equity owners with clear control and incentives. This has been seen in sectors such as healthcare and financial services, where PE companies are driving the adoption of AI-driven tools. But this is different from the AI rollup paper that VCs are chasing. This assumes that by simply embedding AI, a service business with a low-headed workforce can be transformed into a software-like platform. For these companies, transformation doesn’t come from owning a service layer. It comes from a new AI native company with fundamentally different economics.
Conclusion: You own software, not services
AI rollup papers are venture capital attempts to mediate multiple gaps between services and software. But there’s a reason for that gap. Even in highly automated businesses, services businesses face different constraints, different economics, and different customer relationships than software companies.
Polyai saw it in 2019. The open market is watching that now. The AI revolution is real. The opportunity to improve your service business with AI is authentic. What is the idea that this improvement will turn them into a software company? Just as it wasn’t in 2019, it’s not realistic today.
AI Rollup may continue to deliver returns, but no VC of the type has been underwritten. At best, they are high-tech private equity. Operationally heavy, rated and no software-like scaling.
The opinions expressed in Fortune.com’s comments are the views of their authors and do not necessarily reflect any opinions or beliefs. luck.
read more:
