AI Automation ROI: Broken Processes Cost Firms $1.55M

Companies are pouring money into AI, but a new survey from process-orchestration vendor Camunda says the main obstacle is not the models. It is the workflows those models are dropped into. According to Camunda's report, The AI Process Gap, 72% of organizations say process-related problems have caused AI initiatives to fail, at an average cost of $1.55 million per business.2 The findings were released from Berlin and San Francisco on September 9, 2026, and republished two days later.21

What the survey found

The central claim is that enterprises are attaching AI to processes designed before AI existed, instead of redesigning those processes around the outcomes they want.2 Respondents seem to know this. Some 82% say their AI investments will be a bust unless they spend more on process redesign. Another 72% say AI costs will spiral unless they get better control of their business processes.1

Knowing about the problem has not changed behavior. Most organizations, 79%, say bolting AI onto existing processes provokes less internal resistance than rebuilding them.2 Even firms that want to rebuild expect it to take a long time: 82% say adapting all their most important processes for AI could take up to five years.2 Kurt Petersen, Camunda's SVP of Customer Success, said organizations are applying AI to pre-AI processes and then wondering why returns fall short. He also cited a finding that 61% of respondents say process redesign cannot keep pace with AI adoption.1

The report also links process gaps to governance risk. Forty percent of organizations reported an AI-related compliance or governance issue in the past 12 months, and process-related problems contributed to 84% of those incidents.2 Camunda's own framing is that recognizing the risk "has not translated into action."2

The automation boom behind the numbers

These results come during a period of heavy automation spending. Workflow-software maker Kissflow, compiling industry figures, reports that more than 66% of organizations have automated at least one process. It also cites a business process automation market that grew from $8 billion in 2020 to a projected $19.6 billion in 2026, with forecasts of $23.9 billion by 2029.3 The same compilation says about 80% of businesses are accelerating automation, half plan to automate all repetitive tasks, and companies report cost reductions of 10% to 50% after implementation.3

The two pictures fit together, but they are also in tension. Adoption is broad and spending is rising, which explains why failures now cost seven figures. The upbeat ROI figures in automation roundups describe organizations that got it right. Camunda's data suggests a large share have not. Some of the optimism has also aged badly. Kissflow's list repeats an older forecast that about 69% of managerial work would be fully automated by 2024, which did not happen.3 It also cites McKinsey research that half of business leaders believe they can automate up to 30% of their workload.3 Expecting to automate a task and actually extracting value from that automation are different things.

Reading the data with a vendor lens

Both sources come from companies that sell process tools. Camunda markets itself as an enterprise platform for agentic orchestration.2 A report concluding that process redesign is the missing ingredient fits neatly with what it sells. The available material does not describe the sample size or methodology, and the $1.55 million figure is an average of self-reported costs, not audited losses. Those caveats call for some caution, but they do not invalidate the findings.

The core argument also matches a pattern familiar from earlier technology waves. ERP rollouts and robotic process automation both disappointed when companies digitized existing workflows instead of rethinking them. The most telling statistic may be the 79% who choose bolt-on deployments because they cause less friction.2 That is an organizational choice, not a technical limitation. It suggests the ROI gap comes largely from avoiding change management.

Why it matters for workers and buyers

For the labor market, the findings complicate the story that AI will rapidly hollow out roles. If major processes really take up to five years to rework,2 large-scale displacement may move at the speed of redesign, not the speed of model releases. Redesign itself is labor-intensive work: process analysts, compliance staff, and the people who know how operations actually run.

For buyers, the practical message is to budget for process work alongside licenses and compute. The 40% compliance incident rate is a reminder that unclear processes create audit and liability exposure as well as weaker returns.2

On balance, the evidence supports a modest but meaningful conclusion. AI tools are not the bottleneck. The bottleneck is the unglamorous, politically difficult work of deciding how an organization should operate once AI is in the loop. Companies that keep skipping that step should expect results closer to Camunda's failure rates than to the automation brochures.