Why measuring ROI on tech investment matters more than ever in the age of Agentic Engineering

Measuring the ROI on technology investment has for a long time been a bit of a black box. Investment goes in, activity happens, but the connection to tangible business value is opaque and untrusted. Attributing spend to value is hard: in complex organisations with multiple teams, systems and investment initiatives, untraceable lines between business objectives and the work teams are doing, coupled with poorly defined methods to measure value, it becomes even harder. Reporting on genuine ROI becomes an afterthought.

For the last two years, I’ve seen AI adoption inside most organisations follow the same pattern: give people access to the tools, let them experiment, and worry about the numbers later. Token usage has been largely unconstrained, licenses have been cheap relative to their real cost, and "we're using AI" has often been treated as an outcome in itself.

That era is ending. As clear winners emerge from the underlying model race and providers move to recoup years of heavy investment, the free-wheeling phase of AI adoption is giving way to something closer to normal technology economics - where cost, value and accountability all have to line up. And they aren’t lining up. Recent market data emphasises this difficulty: a 2026 Gartner survey revealed that only 28% of AI use cases fully succeed and meet ROI expectations, often because organisations expect too much, too fast.

Whatever the precise figure, without governance in place to monitor and manage AI spend, it becomes just another uncontrolled budget line - and for many technology leaders, it already has. That's why I believe return on investment needs more focus now than ever before. Here are three reasons why:

1. The cost of doing nothing about cost is rising

The pricing models behind large language models are volatile and, so far, one-directional: up. As the market consolidates and providers stop subsidising usage, the gap between "AI is basically free" and "AI is a serious line item" is closing fast. Teams that scale from a handful of users to an enterprise footprint can see their bills jump by an order of magnitude almost overnight, simply by crossing a licensing threshold.

Technology leaders who don't yet have a way to monitor, forecast and attribute this spend are flying blind. Observability and FinOps-style tooling - platforms that sit between your teams and the underlying model providers - are quickly moving from a nice-to-have to standard practice, both to control cost and to prevent being locked into a single provider as pricing shifts.

2. Individual productivity gains don't automatically become organisational value

Most organisations are currently stuck between two states: pockets of genuine individual productivity, where people have quietly worked out how to get real value from AI tools in their own workflows, and the much harder job of turning that into organisation-wide benefit. The two are not the same thing, and the gap between them is where most of the real investment sits.

Embedding AI properly into an operating model isn't a software upgrade; it's a transformation on a par with any other major change programme. It requires investment in data foundations, orchestration layers and security, alongside the people and process change needed to actually shift how teams work. If you skip that investment you're left with a collection of individual habits rather than a system that compounds. Understanding what that investment needs to return is what makes the difference between a useful internal tool and an actual capability shift.

David Jensen, CTO at Storio Group, states “You need to be clear what you are optimising for, communicate why effectively and ensure that it’s measurable. At Storio we are optimising for throughput towards strategic outcomes. This has helped elevate the conversation and created focus in areas that cut across teams due to wait times being our biggest opportunity.“

3. Getting it wrong is expensive in ways that don't show up immediately

The organisations most at risk right now aren't the ones moving too slowly - they're the ones scaling AI-generated output faster than their engineering practices, governance and architecture can support it. Weak data platforms, thin security and privacy controls, and inconsistent engineering standards don't show their cost on day one. They show up later, as technical debt, rework, or worse. A proper ROI lens forces the question that's easy to skip in the rush to ship: is this actually well-built, secure, and maintainable, and what's the total cost of ownership once the novelty wears off?

The measurement problem underneath all of this

None of this works if organisations can't actually measure what they're getting. For years, plenty of technology leaders have got away with fairly loose measurement - some proxy metrics, some story points, a lot of gut feel. AI removes that luxury. The best CTOs already measure everything and can point to exactly what a given investment is delivering; for everyone else, rising AI costs are about to force a level of rigour that was previously optional.

That doesn't mean every pound of AI spend should be judged against a hard ROI from day one. Experimentation still matters, and treating every trial as a failed investment just because it doesn't immediately show a return will kill off the exploration that finds the next real opportunity. The more useful approach is to separate the two: ring-fence an innovation budget for experimentation, and hold the rest to account against your actual strategy and roadmap - not simply "can we do this faster now." Where the full outcome won't be visible for months, leading and lagging indicators can offer an early enough signal that you're heading in the right direction, without pretending you have certainty.

It's also worth treating headline productivity claims with a healthy dose of scepticism. AI is a genuine accelerator for producing output, but producing output and creating value are not the same thing - a point proven by recent surveys questioning just how much measurable productivity gain organisations are actually seeing, despite near-universal adoption. Speed without direction just means you get to the wrong place faster.

None of this is an argument for slowing down. It's an argument for treating AI investment with the same discipline as any other significant capital allocation decision: know what you're spending, know what you expect back, and know how you'll tell the difference between real value and expensive noise. The organisations that build that discipline in now - rather than retrofitting it once the bill arrives - are the ones that will still be experimenting confidently in three years' time, rather than explaining to their board why the AI budget got away from them.

If any of this rings true for you - whether you are pausing to consider the forecasted return on your technology investments, trying to understand how to get better visibility of your AI spend, or even looking at how you can get cohesive, enterprise-wide value from AI - then we can help. We’re working with organisations doing just this, day in, day out.