AI is everywhere, but ROI is missing

You have given your team access to AI tools. Months have passed. The time savings have not appeared, the costs have not fallen, and nobody can point to a single process that runs measurably better. The conclusion feels obvious: AI does not work for businesses like yours.

That conclusion is reasonable. It is also, the evidence suggests, wrong.

A recent BBC News segment asked exactly this question, interviewing UK firms that felt AI had delivered nothing. The expert featured pointed to three linked barriers: staff not trusting AI outputs, integrations that proved harder than expected, and a pace of change that outran the skills of the people expected to use it. What the segment did not find was a fundamental problem with the technology itself.

This article is for decision-makers and HR managers who are fielding that same frustration internally. It explains where the gap actually sits, what the evidence says about fixing it, and what a realistic next step looks like.

Why the "AI does not work" myth took hold

The myth is understandable. AI tools were sold, in many cases, as plug-and-play. Vendors promised that access would translate directly into productivity. Businesses bought licences, switched on the tools, and waited.

The waiting continues for many of them. One MIT study, cited in the BBC segment, found that up to 95% of AI projects delivered nothing measurable. That figure has circulated widely, and it is not wrong. What it does not mean is that AI is inherently incapable of delivering value. It means that most implementations were not set up to succeed.

The gap between access and return is a human and organisational gap. It is not a technology gap.

What the evidence says about why returns are missing

Three patterns show up consistently in the research on AI adoption across UK businesses.

First, use cases are too broad. When staff are told to "use AI to be more productive", they do not know where to start. The tools feel abstract. Experimentation stalls. Analysis of UK AI adoption points to execution, not innovation, as the core problem. Businesses that see returns narrow the brief sharply: two or three clearly defined process problems, with a measurable baseline to compare against.

Second, training is insufficient. The BBC segment noted that most employees are not receiving the training needed to make the most of AI tools, despite those tools being widely available. Access without understanding creates frustration, not productivity. Staff need to know not only how to use a tool but also when to trust its outputs and when to check them. That is a skill that has to be taught.

Third, experimentation time is not protected. Research on UK implementations is clear that employees need structured time to experiment if AI projects are to succeed. When AI is added on top of an already full workload, with no protected time to learn and test, adoption does not happen. The tool sits open in a browser tab and closes again at the end of the day.

The real cost of treating AI as a tool rollout

For decision-makers, the cost of the myth is not just a wasted licence fee. It is the opportunity cost of a team that believes AI cannot help them.

Once that belief takes hold, it is difficult to shift. A second attempt at AI adoption faces a workforce that is already sceptical. HR managers see this clearly: staff who were asked to use a tool without support, who found it confusing or unreliable, and who drew their own conclusions about whether it was worth the effort.

The organisational confidence gap that forms after a failed rollout can take longer to repair than the original implementation took to run. That is the real risk of treating AI like a software upgrade rather than a change programme.

There is also a governance dimension worth noting. Designing AI projects with clear accountability and documented processes is not only good practice for adoption. It also positions your organisation well for the UK's evolving AI governance expectations, where documented decision-making and clear human oversight are increasingly expected.

What a structured approach actually looks like

Organisations that report measurable AI returns share a common pattern. They do not start with tools. They start with problems.

The sequence that works runs roughly as follows.

1. Identify two or three specific process problems where the inputs and outputs are well understood.

2. Assess whether the data those processes rely on is clean, accessible, and consistent.

3. Provide targeted training so that the staff involved understand both what the AI can do and where its outputs need human review.

4. Protect time for structured experimentation, with a simple way to record what worked and what did not.

5. Measure against a baseline KPI before drawing conclusions about whether the approach is working.

None of this is technically complex. All of it requires organisational intention. The businesses that report the best returns are not necessarily using more advanced tools. They are using ordinary tools in a more disciplined way.

Considerations and limitations

It is worth being honest about what this approach does not fix.

Narrowing AI to defined problems works well when those problems are genuinely suitable for AI assistance. Not every process is. Some workflows rely on contextual judgement, relationship knowledge, or regulatory precision that current AI tools handle poorly. Starting with a realistic audit of where AI is actually a good fit, before any tool selection, prevents the frustration of pushing a tool into a context where it will consistently underperform.

There is also a leadership dimension. Structured AI adoption requires sustained attention from decision-makers. If the initial enthusiasm fades after the first quarter, the programme typically stalls. Accountability for the adoption effort needs to sit with a named person, not with a vendor or a pilot team acting in isolation.

For HR managers specifically, the workforce readiness question matters as much as the technology question. Staff who feel that AI is being introduced to monitor or reduce headcount will disengage from it regardless of how good the training is. Transparency about the purpose of AI tools, and genuine involvement of staff in identifying use cases, makes a material difference to adoption rates.

If your organisation wants structured support designing that kind of programme, gecco's Training and consultancy offering works with your team to define use cases, build capability, and create a roadmap for sustainable returns.

Your next step if the tools are not working yet

If you are a decision-maker or HR manager who recognises this pattern in your own organisation, the AI Readiness survey is a practical starting point. It is designed to surface exactly the kind of adoption and capability gaps described here, not just to assess which tools you have access to.

Take the AI Readiness survey. You will get access to 65+ free resources and a custom AI Readiness report. We then offer a free 45-minute AI Readiness call to walk through your results.

gecco's Training and consultancy work is built on the view that the 95% failure rate reflects a training and adoption gap, not a tool failure. SMEs that unlock AI returns treat adoption like a structured change programme, with defined roles, experimentation time, and ongoing capability building. If you want help building that kind of programme, that is exactly what we do.

Website · LinkedIn · Case Studies · Newsletter