ROI of AI: why productivity does not reach the budget

Let’s start with the numbers, because the gap sits entirely within the same research.

Ask people, and 80% say they are more productive and half say they make better decisions. The results are similar at every level, from the C-suite to individual contributors.

Ask the company, and 37% see a contribution to EBIT (operating profit), the same percentage as last year. High performers, who attribute at least 5% of EBIT to AI and describe its value as significant, remain 6% of respondents.

The sample is not made up of laggards. Among large companies (over $1 billion in revenue), 40% are scaling AI agents, up from 27% a year ago. The issue falls on those working in mid-sized and large companies, which are the ones that have invested the most and that will be asked first where the return is.

What companies do today, and why it isn’t enough

Among companies that use AI, 56% are still experimenting (22%) or running pilots (34%). The authors of the study warn against exactly this: piling up tools and pilot projects without turning them into different ways of working.

Only about 25% of respondents who are not high performers say they have substantially redesigned their workflows thanks to AI. In the other cases, AI entered processes that stayed the same as before. If the process doesn’t change, the productivity gain a person experiences stays with that person: work done better, a bit more room to breathe, less effort. These are real benefits, but no income statement records them.

he McKinsey partners who comment on the study are clear on one point: the advantage of high performers lies in the coherence of their approach, and budget alone does not explain it, even though they spend more. Compared with the others, they:

• redesign workflows: almost three quarters say so, up from 55% a year ago
• have leaders who step up personally to drive AI initiatives
• and defined processes to measure impact, twice as often
• define where and when a person must validate AI output
• more often complete a strategic workforce planning exercise
• take in account how AI use impact on performance management.

The authors also point to the main brake: adoption moves more slowly than technology because organizations struggle to absorb change. The direction is the same across all the data.

Redesigning a workflow means deciding which steps AI performs, which stay with a person, where the result is checked and by what criteria. To do that, you need to know the work "from the inside": the exceptions, the shortcuts that work, the errors an expert spots at first glance. That knowledge lies with the people who carry out the activities, particularly in highly skilled professions. Think of an analyst, a lawyer, a process engineer, a senior recruiter. A central project team can draw the flow on paper, but it does not know what matters, which situations are the most critical, and so on.

The same applies to human oversight of AI, which the AI Act also requires and which high performers define more often than the others. To validate an output you must be able to recognize an error that looks right, and that ability belongs to the expert.

This leads to a key point: involving people in the implementation of AI. If we ask someone to explain in detail how they work, meaning what can be automated, while 39% of respondents expect headcount reductions, that person has little reason to be generous with what they know. Before starting a redesign, it must be made clear what happens to the roles, workloads and development of those who free up time.

The redesign starts from the work, but it runs through levers that sit in HR:

• workforce planning
• skills assessment and development
• performance systems
• change management capability

If these levers stay still while processes move, roles, skills and objectives fall out of alignment. High-performing companies, in fact, treat them together with the technology, and the study confirms it.

We work on three questions that an HR leader faces:

• where does AI change the work?
• who is able to work with it?
• how are new internal capabilities built?

Mapping the work with the people who do it. With groups of process experts we break activities down, identify what can be handed to AI and where human judgment is needed. The redesign is born from their knowledge and carries their name.

Measuring how people work with AI. AI Based Challenge© assesses skills through simulated business cases. For HR, it means having data on how each person integrates AI into their own work, instead of opinions and self-assessments.

Building internal capability. In AI Workspace, managers and specialists experiment by building chatbots and agents around their own work. Whoever has built a tool knows where it works and where it doesn’t, and brings that experience into the redesign of processes.