Would you consider a 10 percent success rate a win? Not many leaders would, but here’s why Grace Miller does.
As Head of Failure and Experimentation at FlightStory, the media company behind Steven Bartlett’s Diary of a CEO, Miller has built her work around an uncomfortable number. Roughly nine in 10 experiments her team runs will not be a clear success, and this isn’t a flaw, it is the underutilized growth system for today’s market.
In the four years since Miller joined FlightStory, the system she built has helped scale Diary of a CEO from 100,000 subscribers to more than 17 million, with over a billion YouTube views and a multimillion-dollar partnership business.
I interviewed Miller live, in front of a room of senior revenue leaders. From my career inside enterprise revenue at IBM, and now advising Fortune 500 and ASX 200 teams on growth, I went in asking the question every senior corporate leader will ask of FlightStory’s playbook: Can this actually work inside an established business?
Here are the seven counter-intuitive lessons I took from the conversation – and they are for all organizations, not just those with a podcast.
1. Test what most leaders would find absurd
When most companies say they experiment, they mean an A/B test. FlightStory operates at a different order of magnitude and this distinction matters.
Before a Diary of a CEO episode goes live, the team has been known to test more than 100 creative variations, so the version that publishes is already optimized for attention. For Steven Bartlett’s book launch, they ran over 200 title variations. They have A/B-tested details as small as whether a social tile should carry an exclamation mark, the kind of granular variable most marketing teams wouldn’t isolate. They even monitor the carbon dioxide level in the recording studio, because stale air measurably dulls how clearly a guest thinks and speaks.
Each test isolates a single variable. If you change five things at once and get a result, you can’t say which change produced the lift. Miller admits FlightStory made that mistake early on.
“We were running experiments and just going, what can we run and what’s going to work? Let’s try everything. And that was kind of the wrong route,” she acknowledges.
The discipline now is one question, one variable, one measurable outcome. You rarely outgrow a competitor with a single brilliant idea; you ‘outfail’ them, at a volume they are not able or willing to match.
The question for leaders is: How can we adopt an experimentation approach to our work?
2. Make ‘not trying’ the unacceptable behavior
In most organizations, failure isn’t seen as a good thing; it’s what many fear will surface in performance reviews. At FlightStory, the inverse applies: the behavior that is unacceptable is failing to try something new. There is a deliberate bias towards learning and curiosity.
If you change five things at once and get a result, you can’t say which change produced the lift.
Miller’s team has ritualized and rewarded failure. FlightStory keeps a dedicated Slack channel where anyone can post an experiment, whether it succeeded, failed or proved inconclusive. Steven Bartlett actively comments on posts, not just the wins, reinforcing the standard Miller has set across the company. A weekly trophy goes to whoever ran a notable experiment and it isn’t just about the result – the winner might not have seen a positive outcome at all.
The diagnostic question for any leader is simple. Does your culture and your senior leaders actively encourage and reward experimentation and failure?
3. Break the processes that slow learning
When I ask Miller what she rebels against, her answer is process.
“If we followed process, we probably wouldn’t have achieved all the things we have,” she notes.
Her point isn’t that process is bad. Some processes are essential and others are more bendable. The current market rewards speed, which process often blocks, and the art is knowing which to break and which to follow.
As someone who comes from a large corporate background, this one I had to challenge. Miller’s solution is autonomy within clear boundaries.
Jeff Bezos of Amazon drew the same line. The decisions you can reverse – what he called “two-way doors” – move fast. The irreversible decisions – the “one-way doors” – move slowly.
For a corporate team, the practical place to start is internal, where the downside is low. That might mean testing how meetings are run, whether a recurring meeting needs to exist at all, or how internal updates are written. Amazon famously experimented with reading a memo in silence at the start of its meetings and the practice stuck.
4. Expect to be wrong nine times out of 10
FlightStory expects roughly 70 percent of experiments to come back inconclusive, 20 percent to fail outright and only 10 percent to be a clear success. That isn’t a failing team, it is a team testing the edges rather than repeating what it already knows will work.
Naming an owner is not the same as outsourcing the behavior.
It also takes patience. FlightStory tested trailer hooks for almost a year with no clear movement before one finally shifted the outcome, well past the point most organizations would have called the experiment off.
Miller cites Harvard’s Amy Edmondson, whose book Right Kind of Wrong separates basic and complex failures, which are largely preventable, from intelligent failures, where a clear hypothesis is tested in new territory. As Miller puts it, the team doesn’t want to lose money through accidents; it wants to lose it by learning.
5. Give it senior ownership, then make it everyone’s job
Most often there is no clear role or ownership for experimentation. FlightStory created a role most organizations don’t have – Head of Failure and Experimentation – and brought Miller in to build the function from the ground up, reporting directly to Bartlett. He has spoken openly about why he saw the role as critical: he had watched too many companies stop experimenting as they grew and stall because of it.
But naming an owner is not the same as outsourcing the behavior, and Miller is clear the goal is the opposite.
“You don’t need a head of failure and experimentation. It should be embedded in the culture,” she says.
Every role at FlightStory carries an experimentation expectation; what changes is the proportion. The owner exists to build the capability and hold the standard, not to do the experimenting on everyone else’s behalf.
6. Hire for mindset before skill
FlightStory screens for culture before a candidate reaches a formal interview, through a short, structured assessment built around the behaviors the company runs on: high agency, tenacity and comfort with trying something new.
Skills can be taught; a fear-based relationship with failure is far harder to unwind.
Miller hired one of her strongest performers this way. He had no media background at all, having come from real estate, but his culture test scored exceptionally.
Skills can be taught; a fear-based relationship with failure is far harder to unwind.
7. Run AI adoption as a company-wide experiment
When FlightStory moved on AI, it didn’t hand the tools to a pilot group. It ran a company-wide competition and opened it to everyone.
The first round, a twelve-week challenge, had one instruction: build anything. The point wasn’t the output, it was to strip the fear out of the technology and let people see what was possible. Miller herself used it to teach herself to code, something she had never done before. The current round is sharper still: take the part of your own role you would most like to stop doing and automate it.
“We are going to use AI to help our roles, not replace our roles,” she explains.
Accelerate learning
The thread running through all seven principles isn’t failure, it is the speed of learning. The organizations that thrive in uncertain conditions are rarely the ones with the biggest budgets or the most elaborate strategy. They are the ones that observe, test, learn and adjust faster than everyone else.
What makes this work isn’t a single person. Miller built the operating system. Bartlett models it consistently, in public, every week. That pairing of a dedicated builder and a visible sponsor is what most organizations are missing.
I have seen the same pattern outside media, in the corporate revenue functions I advise. Two years ago I started building experimentation into our own weekly scorecard at Sales Redefined, and the commercial benefit was clear inside a quarter. I have since translated the same disciplines into a 30-day execution challenge for client revenue teams. One recent challenge generated US$9 million of new pipeline in 30 business days, alongside live AI use cases in production. The barrier is rarely capability; it is accountability, role modeling and permission.
If you want growth, you need experimentation. If you want experimentation, you need psychological safety.
If you want growth, you need experimentation. If you want experimentation, you need psychological safety. And if you want psychological safety, leaders have to model it first.
The practical move I take leaders through is to follow Adam Grant’s theory of ‘Think like a scientist.’ Set the hypothesis, isolate the variable and remove the bias from the result. Run it for 30 days, watch for the early indicators it is working or needs tweaking, and only then scale it up.
As Miller puts it: “If we want to grow, we need to try new things.” And sometimes that means breaking a few processes along the way.
