The AI investment surge hasn’t produced the expected results yet. That could change in 2026

Boards that approved artificial intelligence (AI ) budgets two years ago are now asking about returns. And chief financial officers are scrutinising AI spend with the same rigour they apply to any capital programme.

But in many large organizations, the returns are not there – not because the technology has failed, but because the investment has gone to the wrong layer. A recent global study by PwC, for example, shows that 74% of AI’s economic value is captured by just 20% of organizations.

Most enterprise AI to date has operated at the level of the individual – tools that make an employee faster, a copilot sitting alongside someone doing a job. But the value of AI in large organizations does not accumulate in individual tasks. It’s in the processes that connect them. Those processes have barely been touched by AI investment so far.

Consider what drives cost, risk and competitiveness in a major pharmaceutical company, a global bank or a large manufacturer. Rather than email volume or document drafting time – tasks increasingly undertaken by AI tools – it's a company’s operational backbone.

In financial services, it's the investigation of hundreds of thousands of transaction alerts every month, each requiring contextual analysis, structured judgement and a documented audit trail. In pharmaceutical manufacturing, it's root cause analysis of production deviations. This process still takes weeks at many major drug companies and requires a team of specialists for each incident. In life sciences, it’s the end-to-end systems and processes connecting data, decisions and regulatory obligations that determine how fast a company can move.

These are workflows – multi-step, data-intensive sequences spanning organizational boundaries and regulatory obligations. A copilot that makes one analyst in a month-long workflow 20% faster does not compress the workflow, it just relocates the bottleneck.

A McKinsey analysis of generative AI’s economic potential in 2023 identified operations, supply chain and risk management as the largest untapped AI value pools in the economy. They are also, by a wide margin, the least touched by current deployments.

In 2026, three issues have converged that make the case for acting now on AI investment.

Firstly, the technical infrastructure for automating complex, multi-step workflows – coordinated systems of specialised AI agents executing a full process end-to-end – has only recently matured to the point where production deployment in regulated environments is viable.

The 2026 Gartner Hype Cycle for Agentic AI report shows that only 17% of organizations have deployed AI agents to date, but more than 60% expect to do so within the next two years. This is the most aggressive adoption curve of any emerging technology in the survey.

The gap between those two numbers is closing fast, and the organizations that move first will build process-level infrastructure that is genuinely difficult to replicate quickly.

At the same time, the accountability pressure has become impossible to defer. Global AI spending reached $235 billion in 2024 and is forecast to more than double by 2028, according to International Data Corporation's (IDC) Worldwide AI and Generative AI Spending Guide.

Boards and investors are no longer satisfied with capability demonstrations, they want to see AI reflected in operating margins, cycle times and risk metrics. The organizations that provide productivity anecdotes rather than process-level outcomes will face difficult questions about what AI investment has actually bought.

Finally, the competitive stakes have changed. When AI was a productivity tool, falling behind meant being less efficient. Now that it can run the workflows themselves, falling behind means operating on a fundamentally different cost and speed curve.

A pharmaceutical company that has automated its production deviation management runs faster, at lower cost and with fewer regulatory risks than one that has not. This gap does not close easily once it opens.

This moment demands a system of agents that operate within existing workflows – the same end-to-end process an organization would otherwise staff with people, outsource to a vendor or patch together with individual solutions.

These agents are called digital workers. These AI employees plug into existing infrastructure, follow enterprise guardrails and execute complete workflows autonomously. Humans are only in the loop where judgement, sign-off or regulatory accountability genuinely requires it.

Governance must be built into the architecture from the start to make this approach work in regulated environments. Every decision must be logged, every recommendation should be traceable and human escalation pathways need to be embedded by design rather than retrofitted for compliance. In pharmaceutical manufacturing, financial services and similar industries, that is the condition on which deployment is possible.

A recent World Economic Forum report on AI agent governance frames the central challenge of this moment precisely: with 82% of executives planning to adopt AI agents within the next one to three years, the gap between experimentation and mature, governed deployment is widening.

Most organizations are still asking "how do we use AI to help our people work faster?" But the question that actually matters now is "which of our core workflows should AI be running, and what governance framework makes that safe to deploy?"

The organizations asking that question, and building toward an answer, are the ones that will look back on this period as the moment their operational advantage was established through AI investment. The others will be explaining, a few years from now, why they spent the decade catching up.