Specialist: Technology & Data

Private equity-backed businesses that can demonstrate a working AI programme, with documented use cases, baseline metrics, and evidence of measurable operational improvement, are beginning to achieve better exit valuations than comparable businesses that cannot. That gap was not visible two years ago, but it is now, and it’s widening.

The shift is from policy document to track record. Buyers are no longer asking whether a target has an AI strategy, they want to know is how AI is integrated into their operating plan (and how they measure ROI). They are asking whether it works, evidenced, measured, and whether what has been built is proprietary or easily replicated.

The following draws on a Mayfair Equity Partners’ Boost! event, where company leaders and operators from across Mayfair’s portfolio examined what AI adoption looks like in practice, and from practitioners who conduct buy-side and sell-side diligence across more than 40 funds.

What private equity buyers now expect to see in AI due diligence

Eight months ago, buyers were broadly satisfied if a target could demonstrate AI awareness. A Copilot rollout, some ChatGPT usage, and a written policy were enough, but that baseline has now shifted. The absence of an executed AI strategy is now flagged as a value drag in formal due diligence reports, and in some cases can even trigger a red flag creating downward pressure on achievable exit multiples.

The current expectation operates across three tiers. Firstly, can the C-suite articulate what the business is doing with AI and why, beyond what IT can tell them? Secondly, are there credible, documented use cases demonstrating AI reducing costs, improving efficiency, or driving revenue? Thirdly, and the one that moves the multiple, is whether what the business has built can be replicated by an AI-native competitor. Businesses that can demonstrate the third tier are the ones attracting a meaningful multiple uplift.

The three root causes of AI programmes that fail to deliver before exit

Research from MIT suggests that around 95% of AI projects fail to deliver expected returns. Reasons identified by practitioners working inside private equity-backed businesses point to three consistent root causes, and none of them is the core AI technology.

The first is data quality. AI systems perform to the standard of the data available to them, and fragmented or inconsistent data creates outputs that management teams cannot trust.

The second is integration. Many AI initiatives remain technically isolated, becoming impressive prototypes rather than operational tools.

The third, and the most significant, is organisational misalignment and lack of capability. AI programmes without a clearly accountable owner above the IT function tend to stall in what practitioners call ‘pilot purgatory’. They are funded enough to continue but are too disconnected from the P&L to deliver or simply do not have the knowledge and capability to do so.

BCG research shows that businesses in the top tier of AI adoption generate 3.5 times greater total shareholder return and 1.6 times higher EBITDA growth than laggards. The difference in outcomes relate to whether these root causes have been addressed, not to which specific model the business chose to deploy.

Why AI has not yet moved the revenue needle in most PE-backed businesses

One of the more candid observations from the Boost! discussions was the asymmetry between where AI is generating results and where it is not. Most AI investment in PE-backed businesses is concentrated in back-office efficiency for faster reporting and reduced manual processing. The front office remains largely underdeveloped.

AI could drive revenue through improved lead conversion, richer customer experience, or accelerated product delivery but, in most cases, it has not done so yet. Even in businesses where AI has roughly doubled development velocity, the improvement has not translated into proportionally faster customer-facing output, because the constraint has just moved downstream to the product management, design, and go-to-market functions.

Acquirers who understand this distinction are asking for evidence of AI in revenue-generating workflows, not just cost-reduction ones. Management teams that can demonstrate both sides of the ledger will be in a materially stronger position.

How to build an AI equity narrative that holds up in diligence

The businesses that perform best in diligence are those that started measuring before they started building. Establishing a baseline is the foundation of any credible ROI claim, whether it’s the cost of a manual process or conversion rate on a sales workflow. Without it, productivity gains are merely asserted rather than demonstrated, and buyers may discount accordingly.

A credible AI equity narrative requires three elements: a named owner for AI programmes at C-suite level; documented before-and-after comparisons on specific workflows rather than generalised assertions about productivity; and an active view on whether what has been built is proprietary or easily replicated.

Businesses that begin this work at, or shortly after, investment generate the track record that holds up in diligence. Those that defer it will be constructing a narrative rather than evidencing one, and buyers are becoming sophisticated enough to distinguish between the two.

The firms that will command the strongest exit valuations in the next two to three years are those that can demonstrate AI as a documented, revenue-generating, and proprietary capability, not those that can show it as a policy on a shelf.

Most management teams say they are pursuing AI transformation. In reality, many are implementing AI productivity tools; simply bolting LLM functionality into existing processes to yield small, incremental gains.

The distinction matters. A business targeting a 30% productivity gain requires a fundamentally different strategy from one pursuing a 10x improvement in how work gets done. They are different investment cases, different organisational changes and different timelines.

Across Mayfair’s portfolio and through Boost! portfolio networking events, a consistent pattern is emerging: businesses often optimise for the first while presenting the second to boards and investors. The first step in building an effective AI strategy is understanding which one you are actually trying to achieve.

What is the difference between AI productivity and AI transformation?

AI productivity means adding tools to existing workflows, meaning employees write emails faster, reports take less time and analyses are produced more efficiently. These gains are real and, in many cases, material. But they are not transformation, which means redesigning those workflows entirely. They are not asking how to make an existing process faster, but whether it should exist in its current form at all.

That distinction surfaced repeatedly at a recent Boost! event attended by operators, executives and technology leaders from across the Mayfair portfolio, and it shapes everything that follows.

This creates two distinct strategies:

Strategy 1: Tool adoption. AI is introduced into existing workflows to help people work faster. Typical gains include shorter reporting cycles, more efficient analysis, faster content creation and reduced administrative burden. Many businesses achieve meaningful results through this approach alone.

Strategy 2: Workflow transformation. The organisation maps how work moves through the business, identifies bottlenecks and decision points, and rebuilds processes around AI-native workflows. The question changes from ‘how do we make this process 30% faster?’ to ‘would we design this process the same way if we were starting today?’ This is where the potential for 10x improvement emerges.

Workflow transformation requires deeper organisational and individual mindset change. Roles, governance structures and operating models all come under scrutiny, which is why relatively few businesses progress beyond productivity gains.

Why do most AI projects fail to deliver value?

The failure mode is rarely the technology. Most AI programmes underperform because organisations automate existing work rather than redesigning workflows.
Research discussed at the event showed that the overwhelming majority of AI initiatives fail to achieve their expected return on investment. Many organisations become trapped in “pilot purgatory”: an endless cycle of experiments and proofs of concept that never reach production scale.

The common causes are structural: no clear ownership, no success metrics, fragmented data and teams that treat AI as an isolated experiment rather than a business transformation programme.

There is also a systems constraint problem. Improving one stage of a process often exposes bottlenecks elsewhere. A software team that doubles development velocity through AI-assisted coding may find that product management or release governance becomes the new constraint.

What do the highest-performing AI adopters do differently?

According to BCG research discussed at the event, businesses leading in AI adoption achieve 3.5x higher total shareholder returns and 1.6x greater EBIT performance than peers. They are not winning because they have chosen better models. They are winning because they have built stronger organisational foundations.

Three characteristics consistently separate leaders from laggards:

Organisational alignment: AI is owned by business leaders, not treated as an IT initiative. There is clear accountability for outcomes and value creation.

Data readiness: AI cannot compensate for poor information foundations. Successful adopters address data quality, accessibility and governance early.

Execution discipline: The best performers move quickly from experimentation to production, with mechanisms to test, learn and scale while maintaining appropriate oversight.

How should management teams get started?

Start by identifying two or three workflows that matter. Map how information moves through each one, the handoffs, bottlenecks and decision points. Then ask a more fundamental question: if this workflow were being designed today, with AI available from the outset, would it look the same?

That question shifts the conversation from tools to transformation. For most management teams, it is the moment the difference between 30% and 10x becomes visible.

Through Mayfair’s Boost! programme, portfolio management and operations teams continue to share practical examples of AI adoption, workflow redesign and organisational change. The most consequential AI decision facing a management team today is not which tool to deploy, it is which outcome to pursue.

An experience-led perspective on turning belief in AI into a live, value-generating platform, without pretending there’s a universal playbook.

For some time now, our view at Mayfair Equity Partners has been that data and AI are not simply incremental tools, but represent a fundamental capability shift in how information is processed, patterns are identified, and competitive advantage is created.

But while our conviction was strong, our starting point was not a detailed roadmap or a neatly packaged business case; it was something far messier, and far more familiar to leadership teams than many like to admit.

“We started with conviction, not clarity.”

The ability to analyse vast volumes of fragmented information, generate insights at scale, and detect signals that humans simply cannot process fast enough is not a future state. It is already reshaping how decisions are made across every industry, and private equity is no exception.

We believed early movers would build compounding advantage, and we believed waiting for a perfect moment to adopt AI was not a viable strategy, because in fast-moving capability shifts, delay is rarely neutral and is often irreversible. Yet despite that conviction, we faced an uncomfortable reality: we didn’t yet know where to apply AI in a way that would genuinely move the needle. The technology opportunity was clear, but the business anchor was not.

Why early exploration didn’t work

Our first instinct was the same one we see across many firms: create a committee, spread the responsibility, and ask smart people to explore possibilities alongside their day jobs. Twelve months later, we had activity, discussion, and interesting ideas, but very little delivery.

Not because the team lacked talent, but because emerging domains do not respond well to part-time ownership. AI is not a bolt-on capability; it demands focus, iteration, and a willingness to make decisions with imperfect information. Committees are good at generating debate, but they are not good at shipping platforms.

The turning point: single-point accountability

The real acceleration came when we changed the model, brought in dedicated experience, and gave it clear ownership. Mayfair appointed a CTO, Josko Grljevic, with a defined mandate: crystallise the problem, define the architecture, build a working solution, and drive adoption. Not as an exploratory exercise, but as a delivery effort with executive sponsorship and momentum behind it. That shift mattered for a simple reason: in unfamiliar domains, learning slowly is expensive.

“In unfamiliar domains, experience shortens the path to value.”

The cost of stalled decisions, repeated false starts, and prolonged experimentation often exceeds the cost of bringing in experience early, and once we had a single accountable owner, decisions moved faster, trade-offs were made quickly, and experimentation became structured rather than open-ended.

Choosing the right problem to solve

AI can be applied across almost every function in a private equity firm, from due diligence to portfolio monitoring, reporting, market intelligence, and internal operations, but not every use case justifies the complexity of building something meaningful.

We needed a problem where even a modest improvement would create disproportionate impact, and that meant focusing on earlier, higher-quality investment opportunities. In a market where many deals arrive through highly intermediated processes, the real advantage comes from identifying opportunities before they become competitive auctions, which led us to a simple exam question: can we identify investable signals earlier than the wider market?

“The goal was not to replace judgement, but to focus it.”

The biggest surprise: AI wasn’t the hard part

At the outset, we assumed the most complex challenge would be the technology itself: machine learning models, tooling, and technical uncertainty. In practice, the opposite was true.
The AI stack is increasingly accessible, strong tools exist, and strong partners exist, and the technology layer can often be implemented faster than people expect. What proved far harder was defining what “good” looks like. AI systems require clarity, shared criteria, and consistent logic, and while investing is built on judgement, it is not always built on standardised definitions of what makes an opportunity compelling, which is precisely what any scalable system demands.

The real workload: data, not code

Data is messy, incomplete, lagging, and inconsistent, which initially feels like a constraint; but, if data is imperfect for everyone, then the firm that can combine, enrich, structure, and interpret it effectively builds advantage. Competitive edge comes not from having more data, but from extracting more meaning from imperfect sources, and in our experience the majority of time was spent on sourcing, cleaning, enriching, and structuring information so that AI could generate something reliable.

Adoption is the real differentiator

Even with a functioning platform, the hardest part is not deployment, it’s behavioural change, because tools only create value when they change how people work. That requires integration into existing workflows, reinforcement through visible wins, and a design philosophy that supports judgement rather than attempting to replace it.

“Technology will keep getting easier. Change will not.”

As AI accelerates, the gap between what is technically possible and what organisations can absorb operationally will widen. Change management is no longer a peripheral concern, it’s a core operating capability. The firms that win won’t necessarily have the most advanced models; they will be the firms that adapt faster, iterate more often, and embed new capabilities with less friction.

No universal playbook

There is no single blueprint for building AI inside a private equity firm, and most organisations can capture meaningful value before they ever reach advanced AI. Documenting and improving processes, connecting systems, and automating repeatable tasks often deliver outsized returns. AI should not be used to compensate for broken workflows; it should amplify well-designed ones.

We began with belief in a capability shift, but not certainty about where to apply it. We made mistakes, went down blind alleys, and challenged assumptions that proved wrong. Today, we’ve moved from conviction to a live, evolving internal capability, and more importantly, we have built the institutional memory and the organisational confidence to keep adapting as the technology evolves.

When people talk about AI, the assumption is often that it belongs to engineers and developers. At Mayfair, we are seeing something different: the most exciting breakthroughs often come from people who simply spot a problem, stay curious, and are willing to experiment.

At Mayfair’s recent Boost! Horizon event for portfolio CEOs and board members, Tessa Remp, an executive assistant in our team, presented the live event portal she had built herself using Replit. Tess is not a developer, and has never written a line of code, yet within weeks, she had produced a professional web app that delegates could access throughout the event.

“I’m an EA, not a developer. I know absolutely no code,” Tessa told the audience. “But I wanted to show that you don’t need to be technical to create something useful.”

From curiosity to a real product

It began when Mayfair hosted an internal hackathon for its portfolio CTOs with Replit, an AI-powered coding environment that allows users to build software through natural language prompts, turning plain English into working code.

The brief was deliberately open. Participants were encouraged to bring ideas rather than technical expertise, and to use natural language to turn those ideas into working software. The focus was not on building something perfect, but on understanding what AI makes possible.

“I realised there were no expectations for me to be technical,” Tessa said. “We brought our ideas to the table and watched them turn into code.”

Solving a real event problem

As part of her role supporting Mayfair’s events programme, Tessa decided to build a live event portal that moved from a simple concept to a fully functioning solution for a high-profile gathering of portfolio CEOs. Using Replit and straightforward prompts, she created an app featuring an instantly updateable agenda, speaker profiles with photos and LinkedIn links, a Citymapper-inspired travel planner to help delegates navigate to the venue, and live local weather updates.

What followed was a rapid build process driven entirely by prompts, experimentation and iteration. Tessa integrated the web app with Transport for London and weather services, despite not knowing what an API was, or how it worked behind the scenes.

“If you don’t know what API stands for, neither do I,” Tessa joked during the presentation. “But I got the gist of what it does.”

She also used Replit’s design tooling to generate a polished one-page agenda, including speaker photos, interactive elements, and embedded LinkedIn links. Pushing it further, she prompted the platform to create a moving image of the Shard to bring the portal to life, inspired by the visual design on other firms’ websites.

Working fast, learning faster

On the day of the event, Tessa realised that registration was being managed through a simple Excel checklist. With minutes to spare, she asked Replit to create a one page check-in tool. She pasted in a rough list of names and refined the output until it did exactly what was needed: tick attendees off, track arrivals and highlight who was missing. The entire tool was built in around five minutes.

Not every feature worked perfectly, but instead of being discouraged, she treated it as part of the process.

“One of the biggest lessons is that AI doesn’t always understand you the first time, or maybe even the fifth,” Tessa said. “It’s a conversation. You have to work with it.”

Key lessons: curiosity beats expertise

Tess approached the tooling with curiosity, she tested ideas quickly, asked the AI to suggest improvements, and treated every version as something that could evolve.

It was a useful reminder that deep technical knowledge can sometimes get in the way. Some of the more technically experienced participants over-engineered their projects, bringing assumptions and complexity that limited what the AI could do. By contrast, starting with plain English and curiosity proved to be an advantage.

“The most remarkable part was that Tessa built the strongest application in a room full of technical people. It was a real reminder that AI is shifting what ‘building’ looks like.”

Investing in our team, and the future

This experience reflects how we think about AI at Mayfair. We see it as a productivity tool for everyone, not just specialists. This is why we actively collaborate with leading AI companies like Replit, and why we create opportunities for our team to explore emerging tools in a practical way.

“I didn’t think I could do this six months ago,” Tessa said. “But once you start exploring, you realise how much is possible.”

A few years ago, the idea that anyone in the organization could build and deploy a live web app for an international event would have seemed unlikely. Six months ago, Tessa would not have expected it herself, but today, it is simply part of how we work.

Reflections from Boost! Horizon

Software creation is changing fast

One of the clearest takeaways from Boost! Horizon was that the biggest impact of AI may not be in analysis, research or content generation, but in how quickly organisations can now turn ideas into working software.

The Replit session in particular landed strongly because it demonstrated something that many leaders instinctively understand but have not yet seen in practice: that prototyping is no longer limited to engineering teams. The gap between a business problem and a functional digital solution is reducing rapidly, with implications for speed, innovation, productivity and competitive advantage.

Why this matters: prototyping has always been a bottleneck

In most organisations, the ability to build software has historically been constrained by engineering capacity. A marketing team might want a campaign microsite, an operations lead might want a workflow tool, or a product team might want to test a new feature, all competing for attention and resource.

At Boost! Horizon, the shift being discussed was not just ‘AI makes engineers more productive’. It was the idea that AI is changing who can build, and how quickly they can do it.

“What we believe is the next wave of software creators really is right here in this room, no matter what technical background you might have.”

This is not about removing the need for engineering, rather it is about changing the front end of innovation. If teams can prototype solutions themselves, engineering effort can be directed towards scaling, security and integration, rather than get tied up in early-stage experimentation.

The rise of “prompt-to-software”

The most significant development highlighted in the Replit demonstration was that natural language is becoming a usable interface for building digital tools. Rather than writing code line by line, users describe what they want and iterate on the output until it works.

The pace of experimentation has changed and along with that the costs of trying something new. Many ideas that would previously have required a formal project, a development roadmap, and multiple approval steps can now be tested much more quickly.

The first version of a product or tool is rarely the final one, so the ability to test and adapt quickly is increasingly becoming part of how digital businesses compete.

Engineering is not going away, but its role is shifting

A key point that came through clearly during the day is that demos are not the same as production systems. Tools can dramatically accelerate prototyping, but businesses still need technical leadership to ensure that what is built is secure, maintainable and scalable.

However, the underlying shift remains: the early-stage work that traditionally sat exclusively with engineers is increasingly becoming accessible to non-technical teams, changing what engineering teams spend time on.

It also changes the nature of internal innovation. Organisations can experiment more widely because the cost of failure is lower. A prototype can be built quickly, tested quickly, and either improved or abandoned without large sunk costs.

The organisational implication: more builders, more momentum

Many internal business processes are still held together by spreadsheets, manual reporting, email chains and repeated admin. These problems often persist not because they are hard to solve, but because they never rise high enough on a product or engineering backlog to justify investment.

“But we’re thinking, what if everyone in your company was a builder?”

AI-enabled prototyping changes that equation. If teams can build small internal tools quickly, a large number of ‘minor’ problems can be solved in ways that compound over time. Over the next few years, the companies that move fastest are likely to be the ones that treat prototyping as a distributed capability rather than a centralised function.

A competitive advantage hiding in plain sight

The most commercially relevant implication is speed. If software becomes easier to prototype, competitors can test ideas faster, launch internal tools faster, and improve customer-facing workflows faster.

For management teams, this raises a strategic question: if a competitor can build and test a tool in days, what happens to businesses still operating on quarterly planning cycles for digital delivery? In digital markets, product differentiation is often driven by user experience, responsiveness, and the ability to adapt to shifting customer behaviour, making this level of productivity highly relevant.

What Boost! Horizon revealed

Boost! Horizon was not framed as a ‘future vision’ conference. It was a practical look at what is already possible today. The Replit session demonstrated that software creation is becoming more accessible, more iterative, and more embedded into day-to-day business problem-solving.

“Unless you’re using AI on a day-to-day basis, it’s really hard to understand how magical it is.”

The larger takeaway is straightforward: AI is not just changing what software can do. It is changing who can create it, and how quickly it can be created, a major shift in how innovation happens.

Mayfair’s Boost! programme exists to strengthen connections and unlock value across portfolio businesses. Boost! Horizon was designed specifically for senior leaders and board members across our portfolio to create space for strategic thinking and be a practical forum for the reality of executive decision-making in a world where technological change is no longer incremental, but structural.