SuperReturn 2026: Most PE-backed companies are running AI. Few are creating measurable EBITDA impact

The graveyard of AI pilots has never been fuller. Across the private equity industry, firms are reporting ambitious deployments, promising demos, and enthusiastic board updates. Yet EBITDA often has not moved. The technology works. The organizations around it do not. 

This is the defining challenge of the 2026 vintage of private equity AI. Not whether to invest in AI, but why the returns are so unevenly distributed, and what the funds that are genuinely winning have figured out that everyone else has not. The answer is not a better model, a smarter vendor, or a larger technology budget. It is a set of organizational choices that most firms are still getting wrong. 

OMMAX works across close to 4,000 transformation projects. The pattern that emerges from that volume of work is consistent with what the most sophisticated investors are now demonstrating in their portfolios: in an era where 74% of organizations are already deploying agentic AI in production and 77% consider it a strategic priority, the constraint is no longer access to the technology. It is the judgment to use it in the right places, the discipline to govern it properly, and the organizational will to drive it all the way to the bottom line.

The allocation problem that is quietly destroying returns 

Most firms are spending AI budget in exactly the wrong places. Yet pressure to demonstrate activity continues to pull investment toward the visible rather than the valuable. 

The visible is back-office automation: HR chatbots, invoice processing, report generation, and document handling. These are real use cases. They are also largely solved by third-party vendors. Workday, SAP, and their peers have been embedding AI into administrative workflows for years. A portfolio company building its own proprietary tool to answer HR policy questions is not generating a competitive advantage. It is reinventing a commodity at significant cost and distraction. 

The valuable lives in two places. The first is the product or service delivered to customers: embedding AI so deeply that it changes what the product can do, how fast it does it, and what a customer is willing to pay for it. The second is the front- and middle-office, the client-facing professionals, and the operations that support them. The first drives revenue growth. The second drives operating leverage. Together, they are where real margin expansion lives. 

The leading PE firms have made this distinction explicit in how they allocate. Permira, for example, reports that 45% of its portfolio companies now have live AI embedded in their products and services, and 85% are using it in front and middle-office operations. The back office has been deliberately left to commodity vendors. That is not a gap in their AI strategy. It is the strategy. 

What actual EBITDA impact looks like 

The clearest illustration of what happens when allocation is right comes from two portfolio company transformations in the Permira portfolio that have now produced verifiable, sustained results. Both demonstrate the same shift that is becoming visible across the most advanced PE portfolios: from efficiency gains at the margin to AI embedded so deeply in the product that it changes what the business is worth. 

Octus, the global credit intelligence provider, invested approximately 12% of revenue into technology to launch Credible AI, a product that fundamentally transformed how credit analysts access and synthesize information. ARR per client tripled. Publishing time dropped by 90%. Growth has sustained above 30%. None of that came from adding AI as a feature on top of an existing product. It came from rebuilding the core product around AI capabilities so thoroughly that the experience, the speed, and the quality of output became incomparable to anything competitors could offer. Customers paid more. More customers arrived. The compound effect is now durable. 

Acuity tells a different story with the same conclusion. A knowledge process outsourcing firm with 6,000 analysts, the investor committed approximately 20 million euros in one-off costs to build Agent Fleet, an AI platform that now operates alongside the analyst workforce rather than replacing it. Today, 15% of company revenue is generated by fully automated workflows with no human involvement, 30% of total workforce capacity runs through automation, and core workflow efficiency has improved by around 20%. The fear that AI would destroy the business became the catalyst for transforming it into something more valuable. 

What both cases share is not sophistication or scale. It is commitment. Real capital allocation, real organizational change, real accountability for outcomes. Not a pilot. An investment decision made with conviction. 

The demo-to-production gap is where most money quietly disappears 

Identifying a use case is not the hard part. Getting it into production, keeping it there, and improving it over time is where most AI investment actually goes: not into spectacular failures, but into the quiet accumulation of proofs of concept that never scale. 

The firms making consistent progress have learned to close this gap through three practices that look obvious in retrospect but are rarely executed well in practice. CVC has formalized this through its assess-enable-upskill framework, while Bridgepoint relies on portfolio-wide peer communities and reusable playbooks to accelerate adoption across more than 90 portfolio companies. The approaches differ in structure but share the same underlying logic: discipline compounds where improvisation does not. 

The first is building on existing expertise rather than importing new specialists. AI capability built on top of domain knowledge compounds. The most effective approach trains subject matter experts to become AI practitioners, not the other way around. An AI agent built to help analysts write SQL for data analysis identified 2.5 million euros in stock value potential at a single portfolio company within two weeks of deployment. That result came from people who understood what to look for, not just how to run a model. 

The second is blueprinting. One successful implementation is an experiment. Two is a pattern. A documented, reusable approach to a common problem, whether sales qualification, RFP analysis, or financial modelling, can be deployed across a portfolio without rebuilding from scratch each time. The medium, whether a mobile app connecting 800 executives or a WhatsApp community across 90 portfolio companies, matters less than the habit of sharing what works.

The third is measuring before deploying, not after. The single most reliable predictor of whether an AI initiative scales is whether a clear measurement was established at the point of deployment. Without it, the initiative eventually dies, not because it failed but because it could not prove it succeeded. Every agent that moves to production should have a dashboard tracking its impact from launch.

Token spend is the new cloud spend, and most firms are repeating the same mistake 

A decade ago, PE firms watched cloud bills balloon without clear attribution to business outcomes. The discipline that eventually brought those costs under control, FinOps applied as a genuine management practice, took years to mature. The firms that got there early built a cost advantage that compounded over time.

The same dynamic is playing out now with AI token spend. Costs are rising. EBITDA impact is lagging. The instinct has been to apply hard limits: rollbacks, caps, and leaderboards tracking individual consumption. That instinct is wrong. Hard limits do not teach organizations to spend intelligently. They teach them to avoid tools that might attract scrutiny. The right response is visibility into what is being spent and why, routing tasks to the most cost-appropriate model rather than defaulting to the most capable one for every step, and building incentives to optimize rather than avoid. A modern AI workflow involves five to ten model interactions within a single process, not all requiring the same capability. That discipline is not a compromise on quality. It is good engineering. 

Governance is what determines whether any of this compounds 

There is a pattern in OMMAX's transaction advisory and value creation work that appears with striking regularity: companies with technically capable AI systems that are strategically disconnected from where decisions actually get made. AI ownership sits in IT or engineering, while the commercial and operational functions that need to act on AI-generated insight report through entirely separate chains. The tooling works. The organization around it does not.

OMMAX's analysis of B2B software economics makes this point in financial terms. Software companies that fail to connect AI to execution authority, to the actual ability to initiate transactions, enforce business rules, and automate decisions, end up capturing incremental value rather than structural value. The same logic applies at the portfolio level. AI that does not connect to the operating model does not compound. It costs money and produces reports.

AI impact needs to be a CFO-owned metric, collected with the same governance and rigor as revenue or EBITDA, and reported through the same monthly processes that already exist. Productivity gains that do not trace to financial outcomes are not value creation. They are activity.

The leadership dimension is equally decisive. Bain Capital has highlighted that the single most consistent characteristic of top-quartile portfolio companies is not AI expertise but leadership curiosity: CEOs who actively test, learn, and engage with the technology themselves rather than delegating it as an IT workstream. CVC applies a formal assess-enable-upskill framework specifically to ensure AI capability sits at the executive level, not buried in engineering. The firms producing the strongest results are led by executives who treat AI as a CEO-level priority, who understand enough to cut through vendor noise, and who have connected AI performance directly to management incentives. Hiring for that profile has become one of the clearest differentiators between the top quartile and the rest.

The reframe that separates leaders from laggards 

There is no such thing as AI revenue. There is only revenue enabled by AI.

It sounds like a semantic distinction. It is not. It describes a failure mode that is already visible in the market: companies that have organized their narratives around AI as a category, reporting AI revenue lines and AI feature counts, and have lost sight of the actual question. Are we solving a real customer problem better than we could before? 

The companies producing the most durable returns in leading PE portfolios are not the ones with the most elaborate AI infrastructure. They are the ones who identified a problem customers actually had, built a genuinely better solution, and used AI as the mechanism. The AI is often invisible to the customer. The improvement in outcome is not. This is also what is now reshaping software valuations: the market has moved decisively away from rewarding feature breadth toward rewarding execution authority. Companies embedding AI into the execution layer are capturing structural value. Those merely adding features are not. For private equity, the implication is direct: AI readiness and maturity are no longer product roadmap topics. They are multiple protection topics, assessed at entry and tracked through exit.

Four things the best firms are doing that most are not 

The firms building durable AI capability in their portfolios share four observable characteristics worth encoding explicitly.

1. They answer the allocation question before any other. Before tools, models, or platforms, they establish where AI investment actually creates value and where it creates activity. Back-office functions handled by commodity vendors are not proprietary investment opportunities. The product and the front and middle office are. 
2. They build measurement infrastructure before they need it. CFO ownership of AI metrics. Biweekly review cadences. A clear, explicit map from specific use case to specific financial outcome. All of it is easy to defer until it is too late to matter. 
3. They apply FinOps discipline to token spend before costs become a political problem. Not hard limits, but visibility, routing intelligence, training, and incentive alignment. The organizations that get this right compound the cost advantage as AI embeds more deeply into every commercial and operational function. 
4. They resolve the leadership question in the right direction. AI is not an IT function that happens to touch the business. It is a business function that happens to require technical capability. The CEO either owns it as a strategic priority or the transformation stalls at middle management. Across close to 4,000 OMMAX projects, technology accounts for roughly 30% of what determines whether an AI transformation succeeds. The remaining 70% is ownership, change management, and the organizational will to follow through. 

The question facing private equity today is not whether AI creates value in portfolio companies. The evidence is substantial and growing. The question is whether your portfolio is building the organizational conditions required to capture that value, or accumulating the infrastructure of a future case study in what went wrong. 

From AI strategy to value creation 

OMMAX is a leading AI-first management consultancy supporting investors and portfolio companies across AI strategy, business transformation, transaction advisory, and value creation. Our work spans the full transformation journey: identifying value creation opportunities, assessing AI readiness during due diligence, designing operating models and governance structures, building organizational capabilities, and scaling AI initiatives into measurable growth, profitability, and exit outcomes. 

If you would like to exchange perspectives on AI transformation in private equity, please reach out to the authors of this article.