Advisory by default

Private equity operating teams are naturally positioned to guide the diffusion of AI tooling in their portfolios. They have control positions and concentrated ownership; investment theses are increasingly assuming efficiency gains from AI; and the teams’ mandates are explicitly to build enterprise value within portcos.

In the last few months, I’ve spoken to dozens of operating teams about how rollout is going. Surprisingly, more than half of teams are in a pure advisory role.

Five postures

The most hands-on operating teams (roughly 5%) are running small, dedicated internal teams who build and deploy AI tooling themselves. As an example, they might build a lightweight sales coaching agent with one portfolio company, harden it at the next portco, host it on GitHub, and then roll it out across the full portfolio. (We’ve only seen this model at middle- and lower-middle-market funds.)

Some (another 5%) are running dedicated internal teams whose mandate is to (1) embed with portfolio companies, (2) run AI opportunity diagnostics, (3) map existing workflows, and (4) refer the tool-building to an implementation partner. They’re often partnering with a CFO, CTO, or COO at each portco. There’s a healthy mix of focus on “where can we automate?” and “where can we unlock capacity or improve performance?” These teams are generally staffed by semi-technical ex-consultants.

Another ~20% of funds are supporting via portfolio assessment and benchmarking. As an example, they may run portfolio-wide surveys or workshops to evaluate AI maturity across functions and across orgs, then turn this assessment into a KPI packet for boards to monitor on a quarterly basis. The spirit of this approach is that once maturity is measured, portcos will naturally push to find AI opportunity within their own orgs. The belief is that portco execs are best-positioned to target and drive change given the fact that they hold the most company context and own strategy. The operating team’s role is to share data and help structure thinking.

~20% are hands-off on targeting, but deeply hands-on with impact measurement. Portfolio execs make decisions about where to invest (often backed by business cases) and set target KPIs. Operating teams, in turn, carefully monitor KPIs to assess project success and drive accountability. Some teams have even built their own data platforms, which portcos integrate with, to track real-time performance metrics.

Management antibodies

Why are more PE-backed firms not further along in AI deployments? Most often, the problems aren’t technical.

Political buy-in is hard, and middle-management is often a decision choke-point. Conversation that touches on automation opportunity is generally met with a strong negative reaction (sometimes spoken out loud, more often not). One team described this as “middle-management antibodies.”

But very few firms have actually made material headcount reductions related to AI. Headcount changes that have occurred are concentrated in customer service organizations. The AI tooling actually getting shipped into production is low-hanging fruit: lightweight tools that enhance existing workflows instead of completely automating them.

Data is another consistent roadblock. Legacy companies, especially those in the middle market, may have siloed, untrustworthy, or inaccessible data.

Process documentation is the third. It rarely exists. Many deployments start with a lengthy phase of current-state workflow mapping to inform what gets built (and how), which means starting from scratch with time-consuming process discovery.

Where deployment time goes

Whether a third-party firm, operating team, or internal transformation team is driving deployment, the phases of work look quite consistent. Projects start with opportunity diagnostic (Where’s the value? What’s feasible?). They advance to workflow mapping (How does this process work today? What’s target-state?). Then alignment (Where do we place a bet?). Spec and test cases (What do we build and how do we test it?). Building, and then extensive testing.

The vast majority of deployment time goes to the pre-build stages. The build stage is commoditizing fast. Thousands of AI implementation shops have appeared, and there is limited data on the degree to which outcomes vary from shop to shop.

Talent for the pre-build phase is short. PE operating teams are struggling to hire AI-savvy talent who can deploy into portcos and drive change. Implementation shops are struggling to staff FDEs. And enterprises are starved for AI transformation talent with experience in both AI tooling and transformation work (an exceptionally rare combination). Within enterprises, “AI Transformation” teams are often staffed from existing Digital Transformation, Change Management, or Process Excellence teams.

Wanted: Scalable deployment options

Put together: the market has converged on a process, most of the process happens before the build, and almost nobody has the tooling or the people for that part. This is the shape we described in What stays scarce. Judgment, relationships, and accountability stay scarce while gathering, mapping, and drafting become software.

Operating teams hold the scarce half. They have standing with management and a mandate to act. What they need is software and service partners who enable them to guide AI deployments across their portfolios in a way that is scalable, data-backed, and measurable.

That’s the gap we are building for at Cobalt. The platform watches how work actually happens, maps it, and drafts the specs and test cases with every claim cited to evidence, so a small senior team can run a portfolio-wide program without a bench of analysts.

PE teams will coalesce towards more hands-on enablement as the opportunity size continues to grow. The interesting question is which firms get there first.