"Your Pilot Should Fail!": Notes from Day 2 of the Drucker Forum

Amy Edmondson on intelligent failure, Ravi Kant Ji on why AI executions fail, and more from the final day of the 2026 Drucker Forum: why your pilots should fail small, early and cheap.

Vidyotham Reddi

11/6/20265 min read

On Day 1 of the 18th Global Peter Drucker Forum, the lesson I took home was that a leader has to be willing to be wrong first. On Day 2, Amy Edmondson gave that lesson a method: be wrong on purpose, small, early and cheap, so you are not wrong big later.

Human purpose before breakfast

The day started early with The India Way's breakfast dialogue on "Human Purpose in Business". Mark Goyder reflected on a visit to the Viktor Frankl museum here in Vienna and on why leaders who can show vulnerability build more durable companies. Unni Krishnan spoke about an "honorable harvest": taking from a business only what it can give back, rather than running it until it is quietly "bleeding insolvency", as he put it about some well-known names. He framed the choice as a living organisation that runs its machines, or a machine that runs the living.

Cheryl Rodness gave an update on Break the Silence, including a new partnership with YourDOST and a confidential hotline. It was a reminder that purpose is not a slide in the strategy deck; it shows up in who gets support when it matters.

Healthcare: 17,000 small things

In the Next Gen Healthcare deep dive, chaired by James Mountford, Adam Sewell-Jones of the NHS said the next big thing in healthcare isn't one big thing; it's "17,000 small things". He described innovation as "directed creativity implemented", which is as good a definition as I have heard. Ideas are cheap; the implementing is the hard part.

Martin Curley argued that Europe spends about 97% of its health budget on illness and 3% on wellness, and that the system has to "stay left, shift left": keep people healthy longer, and catch problems earlier. He also made the case for treating the patient as part of the care force, not just its customer.

AI and cybersecurity: governance at the speed of innovation

The AI and cybersecurity session had the numbers that made the room go quiet.

Thomas Masicek of T-Systems said that when a PHP vulnerability leaked a few weeks ago, it took only 30 minutes before his team saw the first attacks, and they found about 30,000 instances to patch. His team has automated roughly 80% of its security operations work and cut response times from about two hours to 15 minutes. But he also shared that one AI security pilot was burning roughly €25,000 of credits an hour, or about €220 million a year at that rate. And he worried, rightly, that the junior roles where people learn the craft are disappearing.

Christina Rückert of IBM gave the line I wrote down first: "We only can afford the speed in the innovation if the governance has the same speed." She noted that while over 90% of enterprises invest in AI, only about a quarter of those invest in its security, and only 13% feel ready to govern AI agents.

I shared an experience from my former company, a Fortune 500 food manufacturer, when we brought GenAI to our R&D scientists. We went security-first, and the first question was not about tools at all. It was: who owns an idea that AI helped create? We started with licenses for a few people in each business, not a big-bang rollout. Did we pay a price in speed? Probably. Was it the right call for a business whose recipes and formulas are its crown jewels? I still think so.

Innovation and power

The Innovation and Power panel, with Adrian Wooldridge, Mehran Gul and Ravi Kant, pulled the lens out to companies and countries.

Mehran Gul described a Europe full of good companies that are underfunded, with, as he put it, $800 billion a year flowing from Europe to the US. Citing the Draghi report, he pointed out that Europe has not produced a company worth over $100 billion in the past 50 years. Adrian Wooldridge observed that much of today's innovation comes from founder-controlled companies, and that governments can be innovative too, with Singapore as his example.

Ravi Kant Ji brought it back to people. In his team's study across 23 industries, about 75% of AI executions fail. The 25% that succeed do so because of people, culture, leadership and trust, not technology. What separates them, he said, is "inner resolve": the hunger to see it through.

Amy Edmondson: fail well

Then came the Charles and Elizabeth Handy Lecture, and Amy Edmondson of Harvard Business School.

She started by separating three kinds of failure:

  • Basic failures: a known process, a single error. Her example was the Citibank employee who checked the wrong box in an online form in 2020 and sent $900 million to a group of lenders.

  • Complex failures: many small factors lining up, as in the Notre-Dame fire.

  • Intelligent failures: the undesired results of a thoughtful experiment in new territory. Her example was Eli Lilly's Alimta, a cancer drug that failed its trial. A doctor noticed the patients who did badly lacked folic acid; adding it turned the drug into a success.

Only the third kind is worth celebrating, and only if it meets her criteria: it happens in new territory, in pursuit of a real opportunity, is informed by a hypothesis, and is as small as it can be while still teaching you something. The "half" criterion: you actually learn from it.

Her best story was about a very well-run telecom company that piloted a new high-speed internet service. The pilot was small, carefully staffed and a big success. So they launched it in a major city, and customer satisfaction fell from the low 90s to 13. In her words: "The pilot didn't prevent the fiasco because the pilot was a success."

So, she said, your pilots should fail. Before you trust one, ask:

  1. Is it tested under real conditions, or rigged to succeed?

  2. Is the goal to learn, or to prove to the higher-ups how good the idea is?

  3. Are reviews and pay kept separate from the pilot result?

  4. Did you change anything afterwards? If not, you probably didn't learn much.

As someone who has designed many in-market tests, I would call this the most underrated rule in analytics: a test that cannot fail cannot teach you anything. When a test is set up so that the brand team "wins", the result is not data. It is a press release.

Amy also cleared up a common misreading of her own work. Psychological safety is not comfort, and it is not lowering standards. It is the belief that you can speak up with a concern or a bad idea and not be punished for it. Her simplest tool for leaders: ask "Who has a different perspective?" and mean it. She told us about Ben Berman, a senior United Airlines captain, who opens every briefing with a new crew this way: "I have never flown a perfect flight." He needs to hear from them, because small errors happen on every flight and good crews catch them.

Her recipe for an uncertain world: aim high, team up, fail well, learn fast. Then repeat.

Closing: freedom to perform

Richard Straub closed the Forum with a few numbers of his own: more than 700 participants, about 35 sessions and roughly 140 speakers. Next year's theme is "Freedom to Perform: Organizations that Thrive on Trust", on November 3–4, 2027, and the 20th Forum is planned for November 29–30, 2028.

Three days, one thread

Looking back on three days in Vienna, the thread is clear to me:

  • Day 0, The India Way: suspend judgment long enough to see what is really there.

  • Day 1: then make the call and be someone who is willing to be wrong first.

  • Day 2: make that call testable. Fail small, fail early, and change something because of it.

Trust runs through all three. People will not suspend judgment, name a number, or report a failed pilot unless they trust what happens next. Fitting, then, that trust is next year's theme.

So, when did your last pilot fail, and what did you change because of it?

Vidyotham Reddi is Chief Analytics Officer of RSB Insights & Analytics. Reach him at vidyotham@rsbinsights.com.

Get the newsletter