The productivity is real. The profit is missing. Here is the management work that closes the gap.
Eight in ten people now say artificial intelligence has made them personally more productive at work. Fewer than four in ten say it has done anything measurable for their organization’s profits, and that share has not moved in a year. The tools arrived. The value stayed personal.
Dr. Erik Strauss has spent two decades on the measurement side of that gap, first as a professor of accounting and now as an adviser to companies working out why their AI programs feel busy and read flat.
His diagnosis is unwelcome, because it is not about the technology. When employers are asked what stands between them and the transformation they want, they do not say capital and they do not say regulation. In the World Economic Forum’s 2025 survey of more than a thousand employers, 63 percent named the skills of the people they already employ as the biggest barrier.
What follows is what he tells executives to do about that, and the two things most of them are still measuring wrongly.
From the Diary of Dr. Erik Strauss
Dr. Erik Strauss is a Jeddah-based professor of accounting and the senior partner of Strauss MindTech. At Strauss MindTech, he and his co-founder design and deliver executive-level workshops and long-term advisory solutions for major companies and institutions. He teaches at the Prince Mohammed Bin Salman College of Business and Entrepreneurship, was professor of management control at ESCP Business School from 2025 to 2026 and visiting professor at Erasmus University Rotterdam from 2019 to 2025, has chaired the supervisory board of the Storch-Ciret Group since 2023, and co-authored a Harvard Business Review article on performance metrics in the AI era. The line he repeats to boards is that “AI transformation is fundamentally a management and organizational design challenge.”
The Twenty Percent That Disappears
Most AI programs start with a reasonable question and end with a disappointing answer. The question is how the existing work can be done faster. The answer, usually, is that it can.
Then nothing else changes.
Strauss has done the arithmetic on this in public, and he is blunt about where the gain goes.
I think this idea deserves to be made much more concrete, because otherwise the default will be predictable: AI makes a task 20% faster. The organization fills the 20% with more tasks. That may improve short-term throughput, but it misses an opportunity: What if companies deliberately reinvested part of every AI-driven productivity gain into human capability development?
He calls the alternative a learning dividend, and the word is chosen carefully. A dividend is declared before it is spent. It is a decision made at the top of the house, not a residue that survives whatever else lands in the calendar.
The reason it matters is competitive rather than sentimental. Speed is available to every rival at roughly the same price. What is not commoditized, he has written, is the ability to identify “what should exist next”.
What Moving Countries Taught Him About Rolling Out AI
He moved to Saudi Arabia in 2026 to take up a professorship, and the move gave him his sharpest teaching device.
Things that had run on autopilot suddenly needed thought. Which routines change. What has to be learned. Which assumptions still hold, and where the old experience actively misleads. He recognized the pattern immediately from the companies he advises, and turned it into a sentence he now uses with executive teams: inserting AI into yesterday’s processes, he wrote, is “a bit like moving to a new country while insisting that everything should work exactly as it did before”.
The instruction that follows is concrete. Before asking which tools to buy, work through which responsibilities should move between humans and machines, which routines no longer make sense, and where the capacity that AI frees up should actually go.
His feel for that last question is old. Strauss grew up in Kassel, where his parents ran a private practice. “I grew up with experiencing what it means to be an entrepreneur, being responsible for your employees, and leading them to success,” he says. His first paid job, at sixteen, was in a computer games store at five euros an hour, and he remembers it fondly because “I was paid for talking about computer games”. He has been translating technology for the people who have to live with it ever since.
The Day With Nothing to Show for It
The second measurement problem is more uncomfortable, because it indicts the metrics that most managers were promoted on.
If one person produces ten AI-assisted outputs in a day, while another spends the same day carefully reviewing one high-stakes analysis and prevents a costly mistake, traditional productivity measures can easily reward the wrong person. In AI-augmented work, human value increasingly lies in quality, judgment, and error prevention, not in the number of visible artifacts produced. That means performance management has to change accordingly.
That argument is the spine of the Harvard Business Review piece he wrote with Randy Bean and Randeep Singh, which contends that judgment, accountability and orchestration are what firms now need to make visible.
He does not exempt himself from the difficulty. “It still feels satisfying to clear a long to-do list,” he has written, and much less satisfying to spend half a day proving that an AI-generated conclusion does not hold. He keeps the habit anyway, and asks managers to build systems that recognize what he calls invisible value: process logs, scenario discussions, reviews of how a conclusion was reached rather than only what it was.
The Governance Layer Software Cannot Supply
Then there is the risk nobody has priced. Autonomous agents are moving from side projects into private life and then, quietly, into the enterprise, and Strauss has been tracking the exposed instances and the malicious extensions that come with them.
Vendors will sell a technical layer that makes AI usage visible, attributable and governable. It cannot do the part that matters.
A control plane can make intelligence consumption observable. It cannot decide what deserves human attention. It can enforce an escalation rule. It cannot determine whether the person receiving that escalation has the judgment to deal with it.
His conclusion is a sequencing rule, and he states it flatly: “Companies must invest in AI literacy before they scale AI deployment.” Not only at board level. Across management and operational teams, so that people understand how agents behave, how permissions create attack surfaces, and who owns the decision when an agent acts.
Underneath sits an idea he has argued at length. Accountability is not only a governance concept but a psychological one, the knowledge that you may have to defend a decision in front of a customer, a board or your own team. AI, as he puts it, “does not experience that weight”. That is why a human in the loop only counts if the human was present in a way that made the responsibility real.
Two Decades of Watching Firms Measure the Wrong Thing
The credentials are unusually well matched to the argument. Strauss took a doctorate and a habilitation in management control at WHU, the Otto Beisheim School of Management, taught the subject at ESCP Business School in Berlin and Erasmus University Rotterdam, and has chaired a supervisory board at the Storch-Ciret Group, a manufacturing group in Wuppertal, since 2023. He has therefore spent his career on both sides of the table where performance gets defined.
A former collaborator who co-led a working group on integrated corporate planning with him describes him as “a knowledgeable and engaging thought partner, always eager to combine academia and practice.”
The practice keeps sharpening the theory. At a masterclass he taught with MBSC Vice Dean Dr. Larissa von Alberti, a student asked him what a university degree is worth when AI can increasingly do the work students are being trained to do. He did not reach for reassurance. He wrote afterwards that the value of a qualification should rest less on certifying accumulated knowledge and more on demonstrating that someone can think, judge and keep learning when the environment changes. In Handelsblatt he has made the same case about junior staff, arguing that firms which look more productive after cutting their juniors are making an expensive mistake.
Where Erik Is Now
He is based in Jeddah and teaches at the Prince Mohammed Bin Salman College in King Abdullah Economic City, while Strauss MindTech, which he runs with Dr. Christina Strauss, delivers executive workshops and longer advisory programs to companies and institutions from Munich.
His advice to anyone starting out is the one he took himself: “There is no free lunch, i.e., if you want to be successful you must work hard and smart.”
The Erik Strauss Playbook: Turning AI Speed Into Advantage
- Declare the learning dividend first. Decide in advance what share of every AI-driven time saving is reinvested in capability, before the calendar absorbs it.
- Redesign the work, not the tool. Ask which routines should stop existing rather than which tasks should run faster.
- Measure decisions, not artifacts. Build evaluation formats that examine reasoning, so the person who prevented the costly mistake is not outranked by the person who shipped ten drafts.
- Teach literacy before granting access. Make sure managers and operational teams understand agent behavior, permissions and data boundaries before agents touch enterprise systems.
- Name the human who carries it. Assign every AI-supported decision to a person who will have to defend it, because a system cannot feel accountable.
Erik Strauss designs advisory programs at Strauss MindTech and is a professor of accounting at the Prince Mohammed Bin Salman College of Business and Entrepreneurship. Connect with him on LinkedIn.
