The Three Pressure Points On Private Equity Backed Companies
AI creates an impossible roadmap.
Private Equity acquisitions used to be relatively deterministic. Skilled investors could guess the percentage growth of EBITDA they could create with a playbook of 10-20 repeatable changes they could apply to a business. Most of these changes were industry agnostic: renegotiate contracts, increase prices, trim management, etc.
Private Equity investors discovered early that the ways in which companies tended to be bloated, biased or simply irrational made certain improvements almost guaranteed to work. Most of these programs required very little resources too. An operator could come in from the PE firm, get installed as the CEO or CFO and often implement most of the playbook in the first year or two.
From my conversations with PE-backed CEOs, I think AI has changed this and blown the playbook wide open.
PE-backed CEOs now have to make trade-offs across 3-different roadmaps.
MODERNIZATION
This is the first and most direct descendant of traditional PE interventions. Legacy companies have legacy software, legacy processes and legacy attitudes. All those things have to change. It’s the software side that’s most interesting, however.
Operating teams could rarely afford to come in and modernize an entire software stack. For one, they would have had to bring in a significant amount of additional engineers and that’s not great for improving EBITDA over a short-term horizon. The second reason was that improvements didn't have to be holistic. If a company is making $15M a year using legacy software, they can probably make $30M a year using that same software by doubling prices. Hold things constant if they are hard to improve and move the needles that are easier and faster to move. Third, never rewrite software or so it used to be the case.
Now there’s more reason to consider refactoring:
AI is making engineering so productive that rewrites feel attainable;
Competitive pressures are higher – both existing competitors can move faster and new startups can enter and chip away at market share;
There’s an awareness that having legacy code makes agentic engineering less effective or could prevent attracting the type of talent that can make the most of agentic engineering;
The idea that a rewrite could be difficult would imply a lack of organizational comfort over AI tooling. When Anthropic engineers talk about leveraging looping, why are we talking about human bottlenecks for something as deterministic as a software rewrite?
OPERATIONS
The second big roadmap item is expecting that everyone can do their job better with AI.
It’s just not obvious how go get there yet.
Does it mean completely scrapping the vendor list and looking for more AI-native applications?
Does it mean running one hundred internal transformation projects touching every single department?
Does it mean equipping everyone in functional teams to use Claude and build tools for themselves even if they are non-technical?
Or perhaps building a super agent that can understand the entire company and help everyone do their job?
The truth lies in finding the right combination of these ingredients and while operations can be the most impactful area, it’s harder to impact because the people with the domain expertise are often not the ones with engineering experience and not the most AI-native either.
PRODUCTS
Finally, companies are paranoid about how relevant their existing products & services are and how they should imbue them with intelligence.
At one point everyone wanted a chatbot, now having an agent or MCP server on the roadmap is table stakes. But the best companies are seeing OpenAI and Anthropic as direct competitors and are rethinking their strategic moats entirely.
Satya Nadella’s and Alex Karp's comments on sovereign AI are no doubt part of their GTM but that doesn't mean they are wrong. Company moats won't just be reduced to proprietary data and physical assets, there is room to build AI-powered products around proprietary capabilities and use them to create feedback loops that create more data or improve the company’s own proprietary models.
If Prime Intellect making $100M in ARR is any indication, there’s a lot more proprietary model development happening behind the scenes than we may think.
Not only do PE-backed companies have to decide how to juggle the 3 possible roadmaps in parallel, they have to resource all of the work too.
And from where? All three roadmap items require a degree of AI-nativity that neither the PE-backed company nor even the PE operating team may already possess.