Rebuild the Factory: How GPs Should Be Thinking About AI

The private markets have spent years watching AI evolve from conference-room speculation to something GPs genuinely can't afford to ignore. The question now isn't whether AI will reshape fund operations—it's whether firms are building around it or just adding tools on top of what already exists.

At Super Return North America, Chief Solutions Officer Brandon Rembe sat down to talk through the state of AI in private markets: what the technology can do, what it can't, and what separates firms that are capturing value from those still waiting to act.

"Where we are seeing probably the biggest challenges today, but also the most opportunity, is just around data." That's how Rembe frames AI's moment in the private markets, and the framing matters.

Data is fragmented, scattered across disconnected systems, and inconsistently structured. Getting it right is one challenge. Getting value out of it is another.

That's where the foundation work begins. Structuring data correctly is what makes everything else possible, and AI is well-suited to exactly this problem. Converting messy, unstructured data into normalized, queryable data is precisely what the technology does well, and it's already separating firms that can operate with speed and confidence from those that can't.

The outcomes that follow are concrete: fundraising efficiency, DDQ response times, and fund administrator oversight. There's no part of a GP’s operations that can't be improved by AI. The work starts with data.

The firms struggling the most with AI share a common mistake: they treat it as a point solution problem. They look for the right tool for DDQ responses, another for investor reporting, and another for portfolio monitoring. Fifteen tools later, the operating model is even more complex, information is even more fragmented, and the underlying silos are only more entrenched. 

The more durable framing is harder: rather than asking how to inject AI into their current process, firms should be asking how they would reconstruct their business from the ground up if they were starting today. When the world shifted from steam to electric power, the result wasn't a better steam engine; it was an entirely different kind of factory. Firms gaining real traction with AI are asking that same foundational question.

Most GPs can't do this alone. Attracting AI talent is difficult, and training models at scale requires data infrastructure that few management companies have the resources to build. The firms seeing durable results are the ones partnering with organizations that already have the scale, the tooling, and the track record to make it work.

Every firm has a Doug. A partner or senior relationship manager who meets constantly, carries an enormous relationship context, and whose knowledge of what was discussed, what was promised, and what needs follow-up lives almost entirely in their own head.

Getting that knowledge out of Doug's head and into something actionable has always required chasing Doug down the hallway. AI changes that dynamic. When interactions are recorded, automatically summarized, and structured in a CRM in a compliant way, follow-up nearly writes itself. Relationship intelligence that used to be one person's institutional knowledge becomes an operational asset that the whole firm can act on. According to FTI Consulting's 2026 Private Equity AI Radar, firms that deployed AI-augmented data workflows reported a 73% reduction in manual data aggregation hours.

But recording interactions and building AI-assisted follow-up workflows isn't just an operational efficiency play. It improves relationship quality. GPs who aren't scrambling to recall what they said three months ago can show up more present, ask better questions, and act on what they learned faster.

What AI won't replace is the judgment behind those relationships. The ability to read a room, build genuine trust, understand what an LP actually needs, and translate that back to the GP, that remains human work.

In the private markets, there are very few places where being right 90% of the time is acceptable. Get an LP's figures wrong, respond to a DDQ with an error, or miscalculate a distribution; even a small mistake rate can cause real reputational damage.

This is where AI's accuracy trajectory matters. Humans processing data at scale typically operate at around 99.9% accuracy—three nines. AI models, especially when two models independently validate the same conclusion, are approaching four nines. And unlike humans, they don't lose focus on repetitive tasks or leave for another firm six months in.

The concern about hallucinations is real, but is frequently overstated. Humans make data entry errors, too, and those errors are less auditable, less systematic, and harder to correct at scale. 

The language around AI oversight matters. A "human in the loop" sounds like a compliance checkbox. An "expert in the loop" is something fundamentally different: a person with domain knowledge who can evaluate outputs, catch errors, and actively train the model when it gets something wrong.

Hire a Harvard MBA, sit them in a corner without context or feedback, and the results will disappoint. AI works exactly the same way. The firms building durable capability are treating model training as an ongoing operational investment, not a one-time configuration. Getting to four nines of accuracy—and staying there—requires expert oversight at every stage of the process.

The instinct when facing something disruptive is often to wait: to see which tools win out, which use cases prove themselves, which competitors stumble first. That instinct is understandable, and it's also dangerous.

AI is evolving faster than most firms can track. The knowledge work that defines fund operations today—DDQ responses, investor reporting, data reconciliation, meeting follow-up—is moving toward AI. The relationship work, the judgment calls, and the trust-building that define this industry will stay with people.

The right response isn't to resolve that tension. It's to start rebuilding: find the right partners, get data infrastructure in order, and treat the next 18 to 24 months as an opportunity to reconstruct operations around a technology that is, by any reasonable measure, the most significant shift in how knowledge work gets done since the internet.