The AI services transformation may be harder than VCs think | TechCrunch

Venture capitalists have convinced themselves they’ve found the next big investing edge: using AI to wring software-like margins out of traditionally labor-intensive services businesses. The strategy involves acquiring mature professional services firms, implementing AI to automate tasks, then using the improved cash flow to roll up more companies.

Leading the charge is General Catalyst (GC), which has dedicated $1.5 billion of its latest fundraise to what it calls a “creation” strategy that’s focused on incubating AI-native software companies in specific verticals, then using those companies as acquisition vehicles to buy established firms — and their customers — in the same sectors. GC has placed bets across seven industries, from legal services to IT management, with plans to expand to up to 20 sectors altogether.

“Services globally is a $16 trillion revenue a year globally,” said Marc Bhargava, who leads GC’s related efforts, in a recent interview with TechCrunch. “In comparison, software is only $1 trillion globally,” he noted, adding that the allure of software investing has always been its higher margins. “As you get software to scale, there’s very little marginal cost and there’s a great deal of marginal revenue.” If you can automate services business, too, he said – tackling 30% to 50% of those companies with AI, and even automating up to 70% of those core tasks in the case of call centers – the math begins to look irresistible.

The improved cash flow then provides ammunition for acquiring additional companies at higher prices than traditional buyers can afford, creating what proponents see as a lucrative flywheel.

The game plan seems to be working. Take Titan MSP, one of General Catalyst’s portfolio companies. The investment firm provided $74 million over two tranches to help the company develop AI tools for managed service providers, then it acquired RFA, a well-known IT services firm. Through pilot programs, says Bhargava, Titan demonstrated it could automate 38% of typical MSP tasks. The company now plans to use its improved margins to acquire additional MSPs in a classic roll-up strategy.

Similarly, the firm incubated Eudia, which focuses on in-house legal departments rather than law firms. Eudia has signed up Fortune 100 clients including Chevron, Southwest Airlines, and Stripe, offering fixed-fee legal services powered by AI rather than traditional hourly billing. The company recently acquired Johnson Hanna, an alternative legal service provider, to expand its reach.

General Catalyst looks to double – at least – the EBITDA margin of those companies that it’s acquiring, Bhargava explained.

The powerhouse firm isn’t alone in this thinking. The venture firm Mayfield has carved out $100 million specifically for “AI teammates” investments and led the Series A for Gruve, an IT consulting startup that acquired a $5 million security consulting company and grew it to $15 million in revenue within six months while achieving an 80% gross margin, according to its founders.

“If 80% of the work will be done by AI, it can have an 80% to 90% gross margin,” Navin Chaddha, Mayfield’s managing director, told TechCrunch this summer. “You could have blended margins of 60% to 70% and produce 20% to 30% net income.”

Solo investor Elad Gil has been pursuing a similar strategy for three years, backing companies that acquire mature businesses and transform them with AI. “If you own the asset, you can [transform it] much more rapidly than if you’re just selling software as a vendor,” Gil said in an interview with TechCrunch this spring. “And because you take the gross margin of a company from, say, 10% to 40%, that’s a huge lift.”

But early warning signs suggest this whole services-industry metamorphosis may be more complicated than VCs anticipate. A recent study by researchers at Stanford Social Media Lab and BetterUp Labs that surveyed 1,150 full-time employees across industries found that 40% of those employees are having to shoulder more work because of what the researchers call “workslop” – AI-generated work that appears polished but lacks substance, creating more work (and headaches) for colleagues.

The trend is taking a toll on the organizations. Employees involved in the survey say they’re spending an average of nearly two hours dealing with each instance of workslop, including to first decipher it, then decide whether or not to send it back, and oftentimes just to fix it themselves.

Based on those participants’ estimates of time spent, along with their self-reported salaries, the authors of the survey estimate that workslop carries an invisible tax of $186 per month per person. “For an organization of 10,000 workers, given the estimated prevalence of workslop . . .this yields over $9 million per year in lost productivity,” they write in a new Harvard Business Review article.

Simply implementing AI doesn’t guarantee improved outcomes, in short.

Bhargava disputes the notion that AI is overhyped, arguing instead that all these implementation failures actually validate General Catalyst’s approach. “I think it kind of shows the opportunity, which is, it’s not easy to apply AI technology to these businesses,” he said. “If all the Fortune 100 and all these folks could just bring in a consulting firm, slap on some AI, get a contract with OpenAI, and transform their business, then obviously our thesis [would be] a little bit less robust. But the reality is, it’s really hard to transform a company with AI.”

He pointed to the technical sophistication required in AI as the most critical missing puzzle piece. “There’s a lot of different technology. It’s good at different things,” he said. “You really need these applied AI engineers from places like Rippling and Ramp and Figma and Scale, who have worked with the different models, understand their nuances, understand which ones are good for what, understand how to wrap it in software.” That complexity is exactly why General Catalyst’s strategy of pairing AI specialists with industry experts to build companies from the ground up makes sense, he argued.

Still, there’s no denying that workslop threatens to undermine the strategy’s core economics. The bigger question is how severe the problem is and whether or not that picture changes over time.

For the time being, if companies reduce staff as the AI efficiency thesis suggests they should, they’ll have fewer people available to catch and correct AI-generated errors. If they maintain current staffing levels to handle the additional work created by problematic AI output, the huge margin gains that VCs are counting on might never be realized.

It’s easy to argue that either scenario should perhaps slow the scaling plans that are central to the VCs’ roll-up strategy and that potentially undermine the numbers that make these deals attractive to them. But let’s face it; it will take more than a study or two to slow down most Silicon Valley investors.

In fact, because they typically acquire businesses with existing cash flow, General Catalyst says its “creation strategy” companies are already profitable.

“As long as AI technology continues to improve, and we see this massive investment and improvement in the models, I think there’ll just be more and more industries for us to help incubate companies,” Bhargava said.