AI ROI: Why 'How Many Roles Can We Cut' Is the Wrong Question

The question driving most enterprise AI projects is the wrong one: how many roles can we eliminate with AI? The question that actually determines ROI is: how much more output can one person produce with AI support? By 2026, enterprise AI costs more than most business cases assumed — compute, integration, orchestration, security, governance, data and oversight add up to a cost structure that has little to do with the license price on the vendor's slide. Chasing ROI through headcount reduction alone means missing the real lever: value created per person.

The expensive truth behind every AI business case

According to Presenc AI Research (May 2026), governance now accounts for 8–12% of AI budgets, up from just 3–5% in 2024. A typical enterprise budget split today looks like this: 30–40% software/SaaS, 20–25% cloud, 15–20% talent, 10–15% implementation/consulting, 8–12% data, 8–12% governance, and 3–6% training. GS Consulting (June 2026) found that in one regulated project, platform, model and cloud together made up only 16% of Year-1 TCO — the remainder was the work required to make the system production-ready: data cleanup, architecture, identity, security, compliance, integration, monitoring, training and people.

Companies that ignore these cost blocks tend to land exactly where cost-overrun data suggests: Futurum's 1H 2026 research found that 41% of organisations report higher AI costs than planned — a gap that traces directly back to governance, integration and data work that never made it into the original budget.

Why 'how many roles can we cut' is the wrong question

Gartner (May 2026) examined the relationship between workforce reduction and ROI directly, and the finding is unambiguous: 80% of AI-driven headcount reductions showed no measurable ROI connection. Cutting roles saves short-term wage cost, but it is not a reliable AI ROI lever — a distinction that matters considerably more than most restructuring plans acknowledge.

The right question: what does one person multiply?

AI agents don't eliminate the value of humans — you're going to give existing workers force multipliers.

Deloitte's Global Human Capital Trends 2026 (March 2026) puts a number behind this shift in perspective: organisations that prioritise work design over simple tool rollout are twice as likely to exceed their ROI expectations. A telecom case from the study illustrates the scale: a tool rollout with no workflow change delivered +5% productivity. A redesign that put 90% of the budget into human-AI interaction — new workflows, trust thresholds, escalation paths, training — delivered +30%.

• Redesign workflows — don't just deploy tools on top of old processes
• Define trust thresholds: when does AI decide, and when does it escalate?
• Build escalation paths and control points directly into the process
• Target training at the roles that sit closest to the lever

Training: the smallest cost line, the strongest lever

Training accounts for just 3–6% of the typical AI budget — the smallest line item — yet BCG and Deloitte analyses show it carries the strongest correlation with ROI of any cost category. Gartner (May 2026) quantifies the gap: employees with proficient, diverse AI use are twice as productive, produce 2.3 times higher quality, and generate 3.2 times more process improvements than average users. At the same time, 19% of employees report no time savings from AI at all — a clear signal that tool access without redesign and training simply doesn't translate into value.

Deloitte's State of AI 2026 confirms the pattern: 84% of companies have not adapted jobs to AI despite high automation expectations. The main obstacle is worker skills — yet fewer than half of companies are changing their talent strategy at all. Gartner adds a sharper warning: by 2027, 50% of companies without a people-centric AI strategy will lose their top AI talent.

What this means for decision-makers

McKinsey Global Institute (November 2025) puts a figure on the upside: AI-powered agents and robotics could unlock roughly $2.9 trillion in value by 2030 in the midpoint scenario — but only if organisations prepare their people and redesign workflows, rather than simply automating individual tasks. That is not an automatic outcome. It is an investment decision.

The practical first step

Before signing off on the next AI budget round, a simple exercise helps: take the three to five roles that spend the most time on research, documentation or reporting, and ask not 'do we still need them' but 'what could they produce if AI removed the routine work'. Calculating that value properly — including full TCO rather than just license cost — is where most business cases either hold up or fall apart.