When your AI ROI is flat, it’s time to get back to basics

Written by Ciaran Cosgrave, CEO, Nearform.

With projects stalling and value proving elusive, many executives are blaming the technology and that it simply “isn’t ready”. It’s an understandable conclusion to draw, but it often points to the wrong issue. ​​​​‌ ‍ ​‍​‍‌‍

By now, almost every enterprise has an AI story. The investment’s happening, the pilots are running, and the tools are in place. But turning that activity into measurable business value is proving to be a much bigger challenge. In fact, only 28% of enterprise AI projects are currently meeting ROI expectations, and less than 10% of pilots have made it to production.

With projects stalling and value proving elusive, many executives are blaming the technology and that it simply “isn’t ready”. It’s an understandable conclusion to draw, but it often points to the wrong issue.

Here’s what’s holding your ROI back

Most organisations fall victim to the same thing — launching pilots, setting up innovation teams and giving those teams access to smart tools — leaving everything else exactly as it was. Workflows stay built for human-speed decision-making, with machine-speed systems simply being layered on top. Teams are given access to AI tools, without being upskilled to use them effectively. And then fast forward a few months, everyone’s surprised when results don’t magically appear.

It’s a half-hearted approach.

Abandoning AI projects carries its own cost, though. Businesses lose up to 30% of revenue each year due to exactly the types of structural inefficiencies that AI is designed to — and can — eliminate. This means failing to innovate ultimately means falling behind and sitting back while competitors build compounding advantages with every quarter that passes.

So, the big question remains. How can you build ROI on AI investments?

Avoid shooting yourself in the foot

The gap between AI investment and meaningful business value is almost always an organisational problem, not a technological one. The real cost isn’t failed pilots, it’s half-heartedness. Too many organisations treat AI as optional or experimental and then get frustrated when it behaves that way. It’s like complaining that a gym membership “doesn’t work” when you only show up once a month and never follow a plan. The issue isn’t the equipment but the lack of consistent commitment and integration into your routine.

Now, investing in AI isn’t the same as investing in conventional IT projects, just like you can’t sign up to the gym and expect payback immediately. AI investment needs to be implemented and measured differently. Consider how enterprises adopted ERP systems or email; neither paid back cleanly in a twelve-month window, but very few would argue they weren’t worth the commitment. AI deserves the same grace. It should be viewed more like an R&D investment, with longer horizons and staged milestones, rather than binary pass/fail judgements.

CTOs that make this mindset shift will see the boardroom conversation around AI change, with it moving away from ‘prove it this quarter’ and closer towards ‘build the capability that compounds over time’.

What we learnt from our own experience

We went through this ourselves at Nearform. In the early stages of bringing AI into our delivery model, we saw results that looked promising in isolation: faster improvements in certain areas, useful outputs from specific tools. But it didn’t really stick, and because of that there was no real change in how work moved through the organisation, and no shift in how teams behaved. The gains we expected didn’t compound.

We had to take a step back from the tools and look at the underlying process. We had to get to the heart of the problem. We stopped trying to change everything at once and instead chose a small number of high-priority projects, staffed them with tight cross-functional teams and focused on how AI could work across the full delivery lifecycle, not just the parts that were easiest to automate.

Our lesson wasn’t really about the technology, it was about the operating model around it. When we started treating AI as part of the operating model, instead of a collection of tools, we soon started to see meaningful ROI. We now take the same approach with clients, with faster shipping and fewer people.

Fix the foundations before buying the next shiny tool

In short, if AI isn’t delivering, the bottleneck is usually in one of three places:

• Decision speed: AI operates in seconds, but approvals still take days or weeks
• Data access: critical information is fragmented or locked in systems AI can’t easily use
• Ownership: no one is accountable for end-to-end outcomes

Until these are addressed, adding more tools won’t change the result.

One of the top reasons AI projects stall is because new tools are simply layered on top of operating models that were never built for them. Again, it’s akin to struggling to lift 100 kg as a beginner at the gym, quitting, then going home and still expecting to be fitter. Adding tools to a workflow that wasn’t built for them doesn’t change the workflow. The constraint is underneath: decisions that need to happen in minutes still route through structures built for weeks, and the data those decisions depend on sits across systems that were never designed to talk to each other.

You have to get sweaty in order to see progress and the same thing applies here: the work that actually inspires and gets results isn’t glamorous. Not every core system needs to be replaced, but the processes and assumptions around them may need to change if AI is to operate effectively. It’s about clarifying decision rights, reducing fragmentation in how data is held and accessed, and building teams around outcomes rather than functions. None of that work requires a new model, but instead honesty about what’s really slowing things down.

This is where AI-native engineering delivers a competitive edge, as it enables businesses to stop treating AI as a feature bolted onto existing processes and start treating it as a genuine participant in how the business operates. AI-native engineering means AI is part of how systems are designed and delivered. That means spec-driven development, small cross-functional teams and AI embedded across the full delivery lifecycle, from requirements through to production monitoring.

Every enterprise has access to the same foundation models. The advantage is in the engineering model around them: fewer people, tighter feedback loops and AI handling the repetitive coordination that slows traditional delivery down.

Create a realistic measurement framework

Understandably, you need the entire boardroom to be in agreement — especially if you’re in the unfortunate position of having to prove AI investment after failed pilots. Boards can’t evaluate AI properly if every initiative reports on success differently, so you need to build a standardised AI ROI framework, which tracks utilisation, business outcomes and strategic value. This will enable cross-initiative comparison and portfolio-level decisions.

It’s also important to be realistic about timeframes. Satisfactory ROI on most AI use cases lands somewhere between two and four years, which is a very different benchmark from what technology investments have historically been judged against. That said, faster returns are possible when organisations align people, processes and governance early — especially with AI-native engineering.

Commit fully, or don’t bother

Every organisation wants the outcomes AI promises, but far fewer are willing to do the hard work required to realise them. Leaders have to be prepared to fundamentally rethink how their business operates — not just layer AI on top of existing processes. The fear of failure is real, and caution is understandable. But in practice, it’s not bold attempts that create the greatest risk — it’s hesitation.

AI isn’t something you can trial your way into. It requires commitment at the level of how your organisation operates. The risk isn’t that AI fails. It’s that you never give it the conditions to succeed.

Ready to get more from your AI investment?

At Nearform, we work with enterprise teams to close the gap between AI ambition and real-world results. If your AI investment isn’t delivering the returns you need, we can help you work out why and what to do about it. Get in touch.