Focus Arbitrage

The real story in AI and private equity isn't about tools. It's about what happens when the cost of building software collapses toward zero.

Most people in PE are still thinking about AI as a procurement decision. Which vendor? Which use case? How do we "adopt AI"? That framing misses the structural shift underneath. And it's not because people are dumb. It's because they're pattern-matching to the wrong decade.

If you came up through PE in the 2010s, you watched firms adopt Salesforce, build data warehouses, and roll out various SaaS tools across portfolios. The playbook was clear: evaluate vendors, negotiate contracts, implement with consultants, train users, measure adoption. Technology was something you bought and deployed.

AI looks like more of that. Another category of software to evaluate. Another wave of vendors pitching transformation. But here's what's different: the vendors aren't the story anymore. The story is that you can now build things that used to require a development team.

A year ago, building a custom workflow to extract data from CIMs, normalize it, and pipe it into your deal screening process would have meant hiring engineers or engaging a consulting firm for a six-figure project. Today, someone on your team with domain expertise and a weekend can prototype a working version. I watched an associate at a middle-market firm build exactly this: a CIM parser that extracted key metrics into a standardized format. Two days of work. It now saves their deal team four to six hours per new opportunity. That's not incrementally better. That's a different category of capability.

Right now, every PE firm is in roughly the same position. Everyone is experimenting. Everyone is running pilots. Everyone is "exploring AI." The gap between leaders and laggards is maybe 12 to 18 months of execution. That window is closing. The firms that figure out how to convert scattered experiments into compounding operational advantages will pull away. The firms that keep "trying everything" will look up in three years and wonder why their competitors seem to move twice as fast.

I've started calling this the Focus Arbitrage. In a moment when everyone is distracted by optionality, the returns to concentrated execution are abnormally high.

Here's how it actually works. The bottleneck has shifted. It used to be expensive to build software. Now it's expensive to decide what to build, get organizational buy-in, and actually use the thing you built. The winning firms won't be the ones with the biggest AI budgets. They'll be the ones who ruthlessly prioritize a small number of high-ROI initiatives and execute them completely. The hard part isn't building anymore. The hard part is focus.

This connects to something unsexy but increasingly strategic: your data infrastructure. Every AI application depends on context. The model needs to know about your deals, your relationships, your portfolio companies, your investment thesis. Firms that have clean, accessible data can ship AI applications fast. Firms with institutional knowledge trapped in people's heads and scattered across email threads will spend months on data cleanup before they can build anything useful. Context is what makes AI actually useful for your specific workflows, which means the boring work of data hygiene suddenly matters.

But most firms get the sequencing backwards. They try to build the data warehouse first, thinking they need perfect infrastructure before AI can work. This kills momentum. The project takes eighteen months, burns political capital, and by the time it's done, the original sponsors have moved on to other priorities. What actually works: start with a narrow use case you can ship in weeks using whatever data you already have access to. Maybe it's ugly. Maybe you're pulling from three different spreadsheets and a CRM export. Doesn't matter. Get something working, get people using it, and let them feel the pain of the bad data themselves. Now you have internal champions who understand why the infrastructure investment matters. The data warehouse proposal that would have died in committee six months ago suddenly has advocates who've lived the problem. You've earned the right to ask for the bigger bet.

Which brings me to ownership. I watched one firm create an "AI council" that met monthly to review pilots and debate priorities. After six months, they had a SharePoint folder full of slide decks and no production systems. A different firm put a single VP in charge with a clear mandate and weekly access to the managing partners. Within four months, they had three tools in daily use. It doesn't need to be a senior hire. An analyst or associate with the right combination of domain knowledge and technical curiosity can drive more progress than a committee of partners. What matters is that someone wakes up every day with AI as their primary job, not their side project.

That owner needs to think about two levers that most firms mistakenly treat as separate. There's firm-level AI: making the fund itself better through improved relationship coverage, faster deal processing, smarter market analysis, more efficient reporting. And there's portfolio-level AI: making portfolio companies better through operational improvements, back-office automation, revenue acceleration. These levers compound when you connect them. Learnings from portfolio deployments inform what you build at the fund level. Tools you build for the fund can be adapted for portfolio companies. A win in one place creates leverage in the other. Treating them as separate workstreams leaves value on the table.

The portfolio angle deserves more attention than it gets. Right now, many companies are within 12 to 24 months of each other in AI adoption. That creates a rare moment of shared timing and shared problems. The learnings from one portfolio company can inform what you build internally. The tools you build internally can be adapted for portfolio companies. Wins can be distributed across the portfolio faster than any individual company could move alone. This is one of the underrated advantages of the PE model in the AI era. You have a natural laboratory of related companies facing similar challenges. The firms that treat this as a coordinated learning system will compound faster than those treating each company as an isolated experiment.

One more thing worth saying plainly: production beats pilots. The pilot graveyard is real. Firms run dozens of experiments, declare success based on demo day excitement, and then watch the prototypes gather dust because nobody owns ongoing maintenance and adoption. Pilots are easy. Production is hard. Production means someone is accountable for uptime, for user adoption, for continuous improvement. It means building the unglamorous infrastructure that keeps things running. One production system that people actually use tends to deliver more value than ten pilots that impressed executives in a conference room.

So what does all this add up to? AI in PE is not a tools story. It's an operating model story. Focus beats experimentation sprawl. Ownership beats committees. Production beats pilots. And once you have the context layer and one flagship win, the whole system starts compounding.

The window for focus arbitrage is open now. Everyone is distracted, everyone is exploring, and the returns to concentrated execution are unusually high. In three years, that won't be true anymore. The leaders will have pulled away, and catching up will require buying what they built, not building it yourself.

The cost of software is going to zero. The cost of distraction is going up. The firms that do the hard work of choosing, committing, and shipping will pull ahead. Everyone else will wonder what happened.