The first piece in this series focused on the most expensive habit in private equity: rebuilding digital capability from scratch on every deal, so that the firm pays full freight for the same work repeatedly and never captures the compounding. That habit is expensive because it never looks like a crisis. It just looks like the way the work gets done.

The second most expensive habit in private equity is the habit of adding AI to a company that was never built to hold it.

Here is how it plays out, and it plays out the same way in portfolio after portfolio. An AI mandate comes down, from the firm, the board or the sheer gravitational pull of the moment. The portfolio company responds in the way any capable team responds to a mandate; it moves. Teams buy licenses, switch on copilots, stand up a minimum viable product (MVP) and build an agent or two. Within a quarter, there's a demo that genuinely impresses the investment committee. Something is happening. The box is checked.

And then, quietly, nothing scales. The MVP that dazzled in the demo doesn't survive upon contact with the actual business. The ROI comes in thin, then thinner. Six months later, the initiative has the specific, familiar smell of a thing that is being kept alive out of embarrassment rather than value. Leadership wonders why the technology underdelivered.

The technology didn't underdeliver. The technology was fine. It was asked to stand on a foundation that wasn't there.

The roof before the house

AI is not a bolt-on. It is the top of a stack, and everything it does depends on what sits beneath it. Point an AI system at your data, and it inherits the state of your data. Wire it into your workflows, and it inherits the state of your workflows. Ask it to make decisions, and it inherits your decision rights, your accountability, your definition of what a good outcome even is. AI has no foundation of its own. It borrows the one you already have, and if that foundation is broken, bespoke or missing, the intelligence you paid for faithfully amplifies the mess.

Adding AI before the foundation exists is like building the roof before the house. You can admire the roof. You can put it in the deck. But there is nothing underneath to hold it up, and the moment you lean on it, it comes down.

The stack, from the ground up, is not mysterious. Data is the floor. Cloud and infrastructure provide the underlying foundation. Core operating systems, including the ERP and CRM systems that run finance, operations and the commercial engine, give AI its context and reach. Commercial systems, digital, go-to-market and payments are where the revenue upside shows up. And AI is the crown that sits on top of all of it. You cannot install the crown and work down. You build up, in order, or you don't build at all.

This is why ‘we added AI’ and ‘AI created value’ are two completely different sentences. The gap between them is the foundation, which can either be built deliberately now or paid for later through stalled pilots and wasted quarters.

The foundation is the ultimate horizontal capability

 The first piece in this series argued that PE firms should industrialize common horizontal capabilities at portfolio level rather than rebuild them deal by deal: industrialize the common, customize the rare

The AI foundation is the purest horizontal capability there is. At the data, cloud and core-systems layer, a healthcare services business and an industrial manufacturer share many of the same foundational needs. They both need clean, governed and accessible data. They both need disciplined cloud. They both need operating systems that aren't held together with tape. None of that is sector specific.  Too often, these capabilities are rebuilt from scratch inside portfolio companies that receive an AI mandate, which means the second-most expensive habit is really the first one wearing a more fashionable coat. It's the same traditional tax, paid again, and this time it's also the thing blocking the AI everyone's so eager to deploy.

The move is the one we have argued from the start of this series. The firm doesn't build the foundation deal by deal. It industrializes the foundation once, at the firm level, and plugs it into each portfolio company's operating model. Build it once; every company draws on it. That is the platform play, applied to the exact moment the whole industry is trying to rush past it.

Half of the foundation everyone skips

Now to the part that separates the firms that will get value from AI from the firms that will get a graveyard of pilots. Because even the firms disciplined enough to build the foundation usually build only half of it.

The foundation has two dimensions, not one.

The first is the platform dimension: the systems. Data, cloud, ERP, CRM, digital and payments. This is the half everyone can see, because it's the half you can buy. It shows up as a line item; it has a vendor; it has an implementation plan. When a firm says, "we're investing in the foundation," this is almost always what they mean.

The second is the operating model dimension: how work actually flows. The decision rights, the workflow design, the team structures, the accountability and the definition of a good outcome. This is the half nobody can point to on an invoice, which is exactly why it gets skipped. And it is the half that decides everything.

A data platform installed underneath an unchanged workflow does not transform the work. It can simply make the old process faster while making flawed outputs appear more authoritative. An AI agent dropped into an approval chain that still takes four weeks, doesn't collapse the four weeks; it produces a beautiful recommendation that then waits four weeks. Intelligence layered onto a broken operating model doesn't fix the operating model. It accelerates it in whatever direction it was already pointed.

The systems are necessary, but they are not sufficient. The foundation only creates value when the people whose work runs on it also change how that work is done. The two dimensions are not a menu. They are a sequence and a pair: build the platforms, redesign the operating model that runs on them and then give AI something real to stand on. Skip the sequence, and you get speed with no traction. Skip the second dimension, and you get a faster version of the thing you were trying to leave behind.

Why does AI punish the shortcut?

It's worth being precise about why this is more urgent now than it was two waves of technology ago, because the instinct to move first and build later isn't irrational; it's just calibrated to a world that no longer exists.

When the lever is AI, a shaky foundation doesn't just cap the upside. It manufactures downside. AI on poor-quality data can produce confident, plausible errors at scale. AI applied to an unchanged workflow can also entrench that workflow by making it cheaper to maintain. The failure mode of premature AI isn't a modest miss. It's automating your worst process at speed and trusting the output because it came out of something that sounds smart.

That's why sequencing matters more now, not less. The faster and more capable the thing on top of the stack, the more it costs to build it on nothing.

"But we don't have time to build foundations"

The strongest objection is the honest one, so it deserves a real answer. It goes: this is a nice theory, but our competitors are shipping AI right now, and if we spend a year building foundations, we'll lose the window.

Two things are true in response, and together they dismantle the objection.

The first: visible AI adoption does not necessarily translate into scaled impact. Many organizations have adopted AI but still struggle to turn experimentation into consistent business value. Speed without sequencing is not necessarily a head start.

The second is the advantage PE firms can bring at portfolio level. Foundation-first does not have to be the slow path because the firm can industrialize common data, cloud and operating-model capabilities once and reuse them across portfolio companies. This can provide a faster route to scale than requiring each company to rebuild those capabilities independently.

Where the value lands

The payoff shows up in three places.

It lands in the exit multiple.  A portfolio company sitting on a shared, AI-ready foundation, with workflows redesigned around it and AI operating at scale, can be a more valuable and credible asset than one carrying a collection of pilots that never scaled. Buyers can assess the difference between AI embedded in the business and AI confined to demonstrations.

It lands in the sourcing and diligence edge. A firm that can deliver the foundation can underwrite AI value at diligence with a straight face, because it knows precisely what it takes to make that value real and it already owns the machine that produces it. That's a sharper, more confident bid than a competitor pricing in AI upside, they have no repeatable way to deliver.

And it lands in the fund-level story. A repeatable, industrialized foundation, both dimensions of it, is a cross-fund asset, not a per-deal cost. "We have a proven way to make portfolio companies AI-ready, fast, across the portfolio" is a materially stronger narrative to LPs than "we're excited about AI," which is, at this point, indistinguishable from every other pitch in the room.

What to do on Monday morning

The discipline this asks for is almost entirely a discipline of restraint, followed by sequence.

Resist making AI the first move. When the mandate lands, the instinct is to deploy; the correct response is to check what the deployment would stand on. For each portfolio company, map the foundation in both dimensions: which platforms are in place and how far the operating model has genuinely been redesigned rather than merely automated. Most companies will not be strong on either, and will be honest on the second dimension only under pressure, so apply the pressure.

Then identify the foundation capability most consistently missing across the portfolio, often data, and industrialize it at firm level so the next company can inherit it rather than rebuild it. Pair platform investment with the operating model redesign that makes it valuable, then deploy AI where both foundations are sufficiently mature.

The second-most expensive habit in private equity looks exactly like ambition. It looks like a firm moving fast on one of the most important technology shifts of the decade. It looks like progress, right up until someone leans on the roof and remembers there was never a house.

At HCLTech, we have built a platform playbook that brings together the sequencing and foundational dimensions needed to accelerate AI deployment across the portfolio.