At AI Horizons in Lisbon on October 15, 2026, Bharath Rao, founder of Sittin' Pretty, asked for a show of hands from people who work with AI late into the night. Then came the follow-up: put your hand down if your sleep has got better as a result.
The room laughed. "No correlation, obviously," Rao said.
That moment summed up the day. Nobody at AI Horizons argued that AI doesn't work. Speakers described tools that write code, draft proposals and process documents in minutes. Yet talk after talk returned to the same question: if the machines are doing so much more, why are the people around them not doing less?
Researchers now call this the AI productivity paradox: people save time on individual tasks, but the business doesn't see it. Teams are busier than ever, and in most companies revenue per employee hasn't moved.
For a COO, that is the real question about AI in 2026. Not whether your people use it, but whether the hours it saves ever reach the P&L. Here is what we heard in Lisbon, what the latest research shows, and five steps that turn AI time savings into revenue per employee.
What We Heard on Stage at AI Horizons
The paradox wasn't a side note. It came up in keynotes, panels and audience questions alike.
"If this is liberation, why is it so exhausting?" Bharath Rao, founder of Sittin' Pretty, opened his talk with this, describing people who work until 2 or 3 a.m. on weekdays because there is always "just one more message" to send to an AI tool. The same talk put it bluntly: "A token is rent."
"Burning tokens again and again." Rao works with small and mid-size businesses, and said they give one of two answers when asked about AI. Either they don't know where to start, or they already use it, with no real idea of how it's being used or what it returns.
"Don't automate dysfunction." Rao argued that most AI fails for "wonderfully boring" reasons: brittle workflows, tools that don't remember, and systems that don't fit how people actually work. The advice: fix the plumbing and document one workflow first. And a line every operations leader should keep: "The revolution won't have a shiny dashboard."
"We didn't fire people. We started hiring people back." Anton Voroniuk, CEO of Skillsbooster Academy, whose team builds AI systems for its own education business, explained that AI created new work: setting permissions and boundaries, testing outputs, catching hallucinations. Humans remain the orchestrators.
"All of us in this room can build an amazing demo in two hours." João Costa Fernandes, founder and CEO of Vesster, said this on the "AI agents in the real world" panel. Building is now cheap. Making it reliable, governed and accountable is where the work went.
What the 2026 Research Says
What speakers described from experience, researchers have now measured. Three studies published this year point the same way.
- AI intensifies work rather than reducing it. UC Berkeley Haas researchers Xingqi Maggie Ye and Aruna Ranganathan spent eight months inside a 200-person US tech company. They found generative AI didn't free up time: "Employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day," often without being asked. Their study was featured in Harvard Business Review in February 2026.
- "AI brain fry" is now measurable. BCG surveyed 1,488 US workers at large companies for a study published in Harvard Business Review in March 2026. Workers who oversee AI heavily reported 12% more mental fatigue and 19% more information overload, and self-reported productivity fell once people used four or more AI tools. Among those with brain fry, 34% intended to leave their company, against 25% of those without.
- Many employees say their workload grew. In a survey of 1,250 US office workers, reported by the Journal of Accountancy in February 2026, 31% said their workload rose after AI was introduced, against 16% who said it fell. 59% said AI raised the pressure to perform.
The pattern is consistent: individual output goes up, and so does the load on the people producing it.
Why AI Time Savings Don't Reach the P&L
The paradox has a simple cause. Most companies adopted AI one person and one task at a time, while the process around those tasks stayed the same.
A sales manager drafts a proposal in ten minutes instead of two hours. But the proposal still waits three days for legal review, still gets re-keyed into the CRM, and still goes through the same four approvals. The time saved doesn't reach the customer or the P&L. It goes into more drafts, more versions and more checking.
Three things absorb the gains:
- More output to review. When drafting gets cheap, people produce more drafts, and someone still has to read, correct and approve each one.
- More tools to supervise. Every new assistant, copilot or agent needs prompting, checking and fixing. BCG's research found that heavy oversight of AI tools adds to mental fatigue.
- Higher expectations. As the Berkeley researchers found, faster output raises what managers expect, which pushes people to take on still more.
The result is a team that feels more productive and more tired at the same time, while the business metrics that matter, such as cycle time, cost per transaction and revenue per employee, stay flat.
Five Steps to Turn AI Time Savings into Revenue per Employee
Revenue per employee rises in only two ways: the same team produces more revenue, or the same revenue needs fewer hours. AI can do both, but only if time savings are captured on purpose. These five steps reflect what speakers at AI Horizons described and what the 2026 research supports.
1. Pick one workflow where time is money
Don't start with "AI for everyone." Start with one process that has volume and a clear cost: proposals, order-to-cash, client onboarding, claims, reporting. One redesigned workflow shows results faster than ten tools spread across the company. As Bharath Rao put it in Lisbon: "The first engine shows the second."
2. Measure the baseline before you touch it
Record cycle time, cost per transaction, error rate and the people-hours the process takes today. Without a baseline, saved time can't be found later, and nobody can prove the project worked.
3. Redesign the whole flow, not one task
If AI drafts a proposal in ten minutes but it still waits three days for review and gets re-keyed into the CRM, nothing changes for the customer. Remove the hand-offs, duplicate entry and approvals around the AI step. "Don't automate dysfunction," as Rao put it.
4. Decide where the saved hours go, before you save them
This is the step most companies skip, and why the Berkeley researchers saw time savings turn into more work. Decide upfront what freed capacity is for: serving more customers with the same team, faster delivery, or moving people to revenue-generating work. Dun & Bradstreet's CTO told Fortune that work that took eight hours now takes two, and the company moved developers to new projects rather than cutting them. That is how revenue per employee grows: more output from the same team, not just fewer people.
5. Give it one owner and limit tool sprawl
BCG's research links heavy oversight of AI to more mental fatigue, and found productivity drops once people use four or more AI tools. Give the redesigned workflow one accountable owner, keep human review where the stakes are high, and retire tools that duplicate each other. Then report the workflow's numbers against the baseline every month.
What It Looks Like in Practice
- A 14-day process down to one hour. Fortune reported that energy company AES cut an auditing and data-entry process from 14 days to one hour by changing the whole process, not by handing auditors a chatbot.
- Twice the work with the same team. At AI Horizons, Bharath Rao described an executive search firm that automated how it builds candidate presentations, from hours of PowerPoint to about five minutes. The firm went from 25 to more than 50 live searches and had its first million-dollar quarter.
- At Cody Solutions, increasing revenue per employee through business process automation is the core of what we do. Our Forward Deployed Engineers work inside the client's team, map the workflow as it really runs, redesign it end to end and stay until the result is measured. For example, an international cosmetics distributor with around $50M in annual revenue was building B2B offers by hand across spreadsheets, email and catalogs, and each offer took days. Cody automated the full offer cycle: AI-generated offers synced with SAP, automated approval routing and an AI check for pricing errors. Preparation time fell from days to hours, approvals became 52% faster and offer errors dropped by 43%, giving the same sales team more capacity to sell.
Frequently Asked Questions
What Is the AI Productivity Paradox?
It's the gap between faster individual work and flat business results. AI speeds up tasks, but when the surrounding process stays the same, the saved time turns into more output, more review and more oversight instead of lower costs or higher revenue per employee.
What Is AI Brain Fry?
BCG's term for mental fatigue caused by excessive use or oversight of AI tools. In its 2026 study of 1,488 US workers, brain fry was linked to more errors, information overload and a higher intent to quit.
Does AI Increase Workload?
For many teams, yes. UC Berkeley Haas researchers found that generative AI led employees to work faster, take on more tasks and work longer hours, often without being asked.
How Can AI Increase Revenue per Employee?
By redesigning whole workflows rather than speeding up single tasks, and by deciding in advance where the saved hours go, such as serving more customers with the same team. Without that decision, the time saved is usually absorbed as extra work.
How Should COOs Measure AI ROI?
Against a baseline taken before the change: cycle time, cost per transaction, error rate and people-hours for one workflow. Usage figures and licence counts show adoption, not return.
What Was AI Horizons 2026?
An AI conference held in Lisbon on October 15, 2026, with talks and panels on AI in business, AI agents and the future of work.
The Takeaway from Lisbon
The most honest moment at AI Horizons was a room full of people laughing at how tired they were. AI will keep saving time. The companies that win will be the ones that decide where that time goes, and redesign the work so it reaches the customer and the P&L.
If your team saves hours with AI but revenue per employee hasn't moved, talk to us: [email protected]. Our Forward Deployed Engineers will start with one workflow, with your people and your live data, and measure the result against your baseline.
