This post is an adaptation from my on-line talk to the Chartered Governance Institute West Midlands branch on 8 October 2026, on creativity, culture and decision-making.
A fashion retailer losing more than €100 million a year
In 2014 I spent a few months inside a large Italian fashion group. Sales had been falling at about 6% a year, and in 2013 the group had lost more than a hundred million euros. That is the situation: a national network of several hundred stores, a family company, and a board asking the only question that matters at that point. What can we do?
What happened next is what usually happens. Management consultants were brought in for the turnaround, the whole of it: cost base, brand portfolio, network, organisation. Pricing was one lever among several, and it is the one this story is about, because pricing in a retail network raises a hard question.
There are dozens of things that might move what a customer spends: the price point itself, the promotion mechanic, where the store is, how big the window is, how many mannequins are in it, whether the shop faces the piazza. And a dozen more that have nothing to do with you and change the numbers anyway: the season, the weather, the town fair, the patron saint’s day, a bank holiday, the shopping centre that opened next door. Change one thing in one store and watch sales go up, and you have learned almost nothing. Run the tactics one at a time, A against B, and you will need years you do not have.
To their credit, the consultants knew this, and they knew it is a problem engineers solve for a living. That is why they brought us in. We set it up as a design of experiments across the network: many factors varied deliberately and simultaneously across many stores, in a pattern that lets you separate their effects. We modelled the confounders rather than hoping they would average out. We measured daily: footfall, purchases, basket size, conversion. Regression across the whole dataset then told us which levers moved which numbers, by how much, and with what confidence. Only those were rolled out.
Some of the favourites did nothing. Some things nobody was excited about moved the numbers. The point is not that anyone was wrong. The point is what would have happened without the experiment: a national roll-out on the strength of a well-argued plan, with no way of knowing which parts of it were working, and no sense that anything was missing, because everyone was confident. The business avoided that, not because its people were cleverer than most, but because somebody made testing part of the process rather than an act of courage.
That is the whole argument of this piece. The rest is what it takes to make the question “has anyone tested it?” an ordinary thing to ask in the room where decisions are made. Not brave. Ordinary.
New, and useful
The word creativity does a lot of damage in boardrooms, so let me define it. Teresa Amabile at Harvard has studied creativity in organisations for decades, and her definition is the one I use: in business, an idea has to be new and it has to be useful. Appropriate, actionable, something that changes how the work gets done. Originality alone is not enough for business innovation. We need both halves: an idea worth exploring and a way to find out whether it works.
The first half, the new idea, runs on curiosity. Amabile finds creativity has three ingredients, expertise, creative-thinking skill and motivation, and she places particular emphasis on intrinsic motivation: the interest someone takes in the work itself. Managers can create conditions that encourage it, or conditions that extinguish it. She quotes the Nobel physicist Arthur Schawlow: the most successful scientists are often not the most talented, but the ones impelled by curiosity. Curiosity has a side effect that matters here. Curious people do not stay in their own field. They wander into other worlds (Darwin read an economist, Malthus, and came back with natural selection) and because they have seen several worlds, they can see when two of them share a shape. I will come back to that.
Amabile also asks managers where in their company they would not want creativity. About eighty per cent say “accounting”. In a room of governance professionals, I suspect the other answer is governance. Both answers come from the same place: anywhere with systematic process and legal exposure, creativity feels like risk.
I want to argue the opposite. Governance is exactly where the second half belongs, and it is the half an engineer recognises. Not the flash of inspiration; that is curiosity’s job. The hypothesis, the test, the result you can defend to a sceptic. That is the discipline that turns a curious idea into a useful one.1
Why iterate at all
The instruments a board normally reads describe what has happened. The decision it needs to make concerns what happens next. Nokia and Kodak show two different sides of that problem.
Nokia shows why current profitability is not enough. In 2010 it still reported an operating profit of more than €2 billion, even as its estimated share of the mobile-device market fell. The warning signs were there: its own chief executive acknowledged significant competitive challenges. The lesson is not that the accounts concealed everything. It is that a profitable business can already need a different future.2
In 2013 Nokia agreed to sell substantially all of its Devices & Services business to Microsoft; the sale completed in April 2014. A board cannot wait for losses to settle every argument about whether to change.3
Kodak shows the other side: seeing the technology early does not settle what to do with it. Its engineer Steven Sasson built a digital camera in 1975, and Kodak introduced a professional digital camera system in 1991. It did act on digital. By 2005, digital products and services accounted for more than half its revenue.4
The governance question I take from Kodak is harder than “have we spotted the invention?” It is “how will this become a business that can replace what we depend on today?” Technical capability, customer demand and the economics of a transition are different questions. A working camera answers only some of them.
Financial statements report results. An invention opens a possibility. Neither tells you, on its own, when or how to commit. A small test can supply another kind of evidence: is there demand now, at this price, from these people? Which assumption would change our decision if it turned out to be wrong?
No experiment guarantees that a company will navigate a transition successfully. It gives the board something better to work with while there is still time to change course.
As shown in Figure 1, the talk contrasted the limits of financial hindsight with the uncertainty of a technological opportunity.
The scale of the uncertainty matters. Improving a familiar process may need a short test within management’s existing authority. A new product or business model may need several tests and staged commitments, with board oversight proportionate to the money and risk involved. Calling something “incremental” or “breakthrough” does not settle who should approve it.
What should change is what gets recorded:
| In the proposal | In the decision record |
|---|---|
| What we believe will happen | The hypothesis being tested |
| What evidence would support or challenge it | The measures and review date |
| What we propose to commit now | The resources and limits authorised |
| What we would do with the result | The conditions for stopping, changing or proceeding |
This is where governance makes the difference. If the minutes only show “the board approved the project”, the board has approved an answer without recording the question, and in two years nobody will be able to tell whether it worked, because nobody wrote down what working would look like.
The engineering approach, in four moves
Strip the jargon and there are four moves. None of them needs a laboratory.
The approach is summarised in Figure 2: state a hypothesis, test it, agree how failure will be recognised, and iterate within agreed limits.
Let me run one ordinary decision through all four: a business thinking about opening a second site.
One: state a hypothesis. Not “we should open a second site” but “we believe a second site in this town will reach break-even inside eighteen months, because the first site is turning customers away and a third of them come from there.” Said that way, it can be wrong, and that is the point. An idea you cannot be wrong about cannot be tested. This is what “love the problem, not the solution” means in practice: the hypothesis is about the customers being turned away, not about the building. The moment you are attached to the building you have stopped testing and started defending, which is a different activity with the same meetings.
Two: run a controlled test. Before signing a ten-year lease, run a six-month pop-up and compare its results with the existing site over the same period. Where suitable data are available, look at trading trends in a comparable town too. Those comparisons will not remove every difference between locations, but they help distinguish demand for your offer from a good trading period everywhere. The test informs the eighteen-month forecast; it cannot prove it.
Three: agree the stopping criteria before you start. Written down before the pop-up opens: we will review footfall, conversion and the economics at month three, and decide at month six whether the evidence justifies a lease. Set the thresholds now, including any early-stop conditions. This is the move organisations skip, because it is uncomfortable to write down the conditions of your own defeat, and because by month three somebody has already fallen in love with the town.
Four: iterate within tolerance. The first month may reflect launch effects rather than steady demand. So the tolerance says: no decision on one bad month. But tolerance runs both ways: no extending the trial past the six months you agreed because the manager likes it there. The first version exists to produce the second. It does not exist to be defended.
Two refinements an engineer would add.
- The first is a practical trick for getting a hypothesis out of people’s heads and onto paper, inspired by Amazon’s working-backwards approach. Write a one-page announcement of the future result, in the customer’s words. Include a measure that would demonstrate success. Writing the announcement forces everyone to say what success looks like, in a form that can be wrong, before money is spent, and it does it together, so the definition of success is shared rather than assumed. If the room cannot agree what the press release would say, it has not agreed the hypothesis. The announcement makes the ambition concrete. The next step is to agree the evidence, thresholds and date that will determine whether to proceed. A company secretary could use a twenty-minute version of that exercise to open the discussion.5
- The second is a pair of words engineers keep separate and everyone else merges. Verification: did we build it right, does it do what the specification said? Validation: did we build the right thing, does anyone want it? Most post-mortems I have read describe a verified solution to an unvalidated problem. The product worked. Nobody needed it.
And for the sceptics who think this only works in factories: I have used precisely these tools, capacity models, scenario runs and thresholds agreed in advance, to plan the bars, the transport and the egress of a music festival before the gates opened. The method does not care what industry it is in. It cares whether you wrote the question down.
The organisation that can run it
The method is simple. The organisation that can run it is not, and this is where most attempts die.
The frame I use is adapted from Nadler and Tushman’s congruence model, which examines the fit between work, people, formal organisation and informal organisation. For this discussion, I group the questions under four headings. Strategy is how we go about making money—or, in a public-service or charitable organisation, how we deliver our purpose and sustain it. Work is our internal processes: how things actually get done. People is who we hire and the skills and motivations they bring. And organisation is the arrangement around them: culture, the values we put into practice; structure, how we organise teams, responsibilities and authority; and rewards, the incentives that tell people what really matters.6 The useful insight is that a problem may lie between those elements: two otherwise sound arrangements pulling in opposite directions. A strategy that says “we experiment” alongside a reward system that punishes anyone whose test comes back negative is incongruent, and the reward system is likely to win. The experiments stop, and the strategy document carries on saying “we experiment”.
As shown in Figure 3, the fit between strategy, work, people and organisation matters as much as the individual elements.
This is also my answer to the question people always ask about culture: should it be a high-pressure environment, or yoga before the day starts? Either can be congruent. I have seen both work and both fail. The test is fit with the other three components, not fashion.
The prize is worth naming. A single innovation can be copied. A superior process for producing innovations, one where hypotheses get stated, tested and killed without drama, as routine, is very hard to copy, because it lives in all four boxes at once. And the process comes with tools: the design-of-experiments methodology the engineers brought to the retailer was exactly that, a tool that brought a different discipline to the decision. A stage-gate with kill criteria built in, a press-release template, a hypothesis register, a model library reused from one decision to the next. These accumulate, and a competitor cannot copy the machine that made your product by copying the product.
Congruence tells you where to look. Amabile offers six managerial levers for examining how everyday decisions can encourage or suppress creativity. Here they are with a governance reading.
| Lever | How creativity gets killed | The iterative version |
|---|---|---|
| Challenge | The most available person gets the most urgent project | Match the question to the person who cannot leave it alone |
| Freedom | Goals move every quarter; “empowerment” in name only | Fix which mountain; let them choose how to climb it |
| Resources | Fake deadlines; budgets so tight the creativity goes into finding budget | Enough to run a credible test; justify additions |
| Team | Homogeneous: faster, less friction, high morale, same mind | Diverse enough to disagree, aligned enough to finish |
| Encouragement | Critics look smarter, so every idea meets layers of evaluation | Recognise the well-run test, including the one that failed |
| Support | Failure ends careers, so nobody reports one | A failed test with a clean result is a deliverable |
The extreme case: the startup
Take the organisation to its extreme, because the extreme makes the rules visible. I ran a university spin-out for seven years, with institutional investors and a board, so this is not theoretical for me.
In an early-stage venture, much of the business plan is a stack of untested beliefs with a spreadsheet on top. Who needs the product? What will they pay? Can we make and deliver it at that price? The uncertainty is the operating condition.
Eric Ries’s The Lean Startup gives that condition a useful unit of progress: validated learning. Before building the finished product, build something that tests a critical assumption. A hand-made batch of ten. A pilot customer who pays. Something small enough to change, but real enough to teach you something.7
The governance question is concrete: what should the board expect between funding decisions? Progress includes commercial results, but it also includes evidence that changes what the company should do next.
That has consequences for rewards. If an innovation team’s job is to explore an unproven opportunity, assessing it only on this year’s revenue can push it towards selling what already exists. The strategy says test; the reward says sell. People follow the reward.
Learning needs scrutiny too. A busy programme of experiments is not automatically progress. The question is which important uncertainty has been reduced, and what decision that now allows us to make.
Where the ideas come from
A method for testing ideas is useless without ideas, and I said curiosity was the engine. This is the mechanism.
Gary Klein, who studies how experts get insights, finds that the most common trigger is connection: two things nobody had put side by side. They stay apart for a dull reason. Each is only ever described in its own trade’s vocabulary.8
Here is the example from my own trade:
- A conjoint study in pricing can vary product attributes to estimate customers’ preferences.
- A factorial trial can examine more than one intervention.
- A designed experiment on a production line can vary temperature, speed and feed.
The applications and assumptions differ, but they share a principle: vary factors deliberately so you can distinguish their effects.
Recognising that family resemblance is useful. It lets you ask whether a method developed in another field could help with your problem, and what would need to change before you could trust it. That is what we did in the fashion retailer.
So the habit I would commend is to ask, of anything in front of you, what is this a member of? Strip the vocabulary, name the class, and the other members of the class, and whoever has already solved the problem under a different name, become visible.
AI can make that search across fields faster. In my own work, it has substantially reduced the effort involved in finding connections worth investigating. But a plausible analogy is a starting point, not evidence that a method transfers. Somebody still needs to understand the assumptions and check the result.
The question belongs in the board paper whether or not AI helps answer it: who else has already solved this, under a different name?
What it costs
A bounded test can cost far less than an untested roll-out, especially once you count the time an organisation spends defending a decision it could have challenged earlier. The comparison is between the cost of learning now and the cost of discovering the same mistake at full scale.
Boards like a discounted cash flow because it produces a number, and I have built plenty. But a DCF on a breakthrough is an opinion with decimal places: the input the model needs most, what demand will be for a thing that does not exist yet, is the one input nobody has. The precision is borrowed from the arithmetic, not earned from the evidence.
The engineering answer, and the governance instrument, is to stage the commitment. Fund the test, not the programme. Each gate sits with the appropriate decision-maker and arrives with criteria agreed at the previous stage. The conversation can then use results to update the forecast. Financial appraisal still matters at the start; the test should improve the assumptions before the commitment grows.
One warning from having lived on the other side of this. Funding the test is the easy half. The hard half is that when the test passes, there needs to be a clear route to the next tranche: its size, conditions and release process agreed in advance. A team that passes its gate in March and gets its money in September has lost the people, the momentum and usually the window. The board has killed the thing it approved, by delay, and the minutes will show a yes. So the gate decision is really two decisions taken together: here is the test we fund now, and here is the funding authorised to follow if it passes and the agreed conditions still hold.
How much? Enough to make the result interpretable. An under-resourced trial may tell you more about its inadequate staffing or duration than about the idea. Define what a credible test requires, fund that, and make any additional spending justify itself.
Six ways it goes wrong
Engineers do a thing called failure mode analysis: before a system goes live, you list the ways it could fail. Here are the six I see most often when organisations try to become iterative. The first three are failures of judgement; the second three are the method being gamed.
- Paralysis by analysis. The belief that all the answers are in the data beforehand, if only we commission one more report. They are not; that is what the test is for. A board that insists on certainty before a small test is demanding the output of the experiment as a condition of running it.
- Love of the solution. The product works and the market shrugs. Verified, never validated. The remedy is upstream: state the problem, in whose life and costing what, before anyone in the room is allowed to name a solution. Once a solution has a sponsor it acquires an immune system.
- The echo chamber. Congruence chased too far becomes agreement: the homogeneous team that is fast, friction-free, high-morale and wrong together. Diversity chased too far becomes noise. The line is thin; the test is enough difference to disagree and enough alignment to finish, and a board can apply it to its own composition.
- Testing until it says yes. The first reading is bad, so the sample is re-cut, the period extended, the measure redefined, until the result agrees with the sponsor. That is why the criteria are agreed before the first reading. If new evidence genuinely requires a different test, record the change and its reasons; do not quietly redefine the original trial as a success.
- Measuring the proxy. A target can encourage people to improve the number without improving the outcome it is meant to represent. Footfall is not revenue; a closed audit action does not by itself demonstrate a safer organisation. School attendance is a useful measure if the hypothesis concerns attendance. It does not, on its own, prove that children learned more. Same number, different claim.
- Innovation theatre. Labs, hackathons, an innovation day with sticky notes, and not one decision anywhere in the organisation changes as a result. The test of whether innovation is real is embarrassingly simple: what does the board now do differently?
Underneath all six is a frame worth keeping. Governance, as a discipline, is built to reduce error; Klein calls that the down arrow, and it is essential. The experimental board does not dismantle it. It adds the up arrow, the deliberate search for the better outcome, and gives it the same seriousness in the minutes.
Monday morning
Nothing above requires a strategy department. It requires the papers, the agenda and the minutes, and those already belong to the governance professional. Three changes a governance professional can put forward now.
Put the question in the paper, and the answer in the minutes. Three headings in the standard board-paper template: what we believe, how we will measure it, and the point at which we stop. A proposal that cannot fill them in is a preference, and the template will say so before anyone has to. Then the minutes capture those points, or refer to the approved paper that records them. In two years, when someone asks whether it worked, the answer is in the minutes instead of in somebody’s memory.
Draft an experiment envelope. Propose clear delegated authority for bounded tests: limits on cost, duration and risk, any exclusions, and a deadline for reporting the result. Once approved, the envelope lets management run qualifying tests without returning to the board for each one. It puts the distinction between agreed goals and freedom over methods into governance practice. The board keeps the mountain; management gets the freedom to find the path; the board gets a steady stream of real results instead of a yearly deck of forecasts.
Add one standing item. “What did we learn from what did not work.” Ten minutes, every meeting. The first few will be awkward and thin. Over time, the standing question gives people a reason to bring evidence of learning, including results they might otherwise have kept quiet. What the agenda asks for helps shape behaviour.
When those three have stuck, two more. The risk register is a broad instrument, most of it known hazards with controls against them, and none of that is what I mean. But scan it for the strategic entries, and some of them are untested assumptions in a risk’s clothing: “customers may not adopt the new service”, “the market may move to a subscription model”. Those are hypotheses. Some can be investigated through a pilot, alongside whatever controls are still needed. The useful question is whether a test could reduce uncertainty enough to change the response. And add one question to the annual board evaluation: how many of this year’s decisions were tested before they were rolled out? Boards measure what they are asked about.
The question I left the room with, and leave you with, is the practical one: what would an experimental board actually minute differently, and what in the current rulebook stops it?
Five sentences for the boardroom
The original closing slide brings the argument together in Figure 4. The wording below includes the clarification about agreed destinations and limits.
- Has anyone tested it? Make the question ordinary, not brave.
- Love the problem, not the solution.
- Write the kill criterion before the first reading.
- Agree the destination and the limits; give people room to find the route.
- Minute the hypothesis, not just the decision.
If you try any of the three Monday-morning things and it works, or it does not, I would like to hear about it.
Further reading: Amy Edmondson, Right Kind of Wrong (2023).
References
- Teresa M. Amabile, “How to Kill Creativity”, Harvard Business Review, September–October 1998. Article. Also informs the six managerial levers discussed below; their governance applications are my interpretation. ↩
- Nokia Corporation, fourth-quarter and full-year 2010 results, 27 January 2011. Results report. ↩
- Nokia, “Nokia completes sale of substantially all of its Devices & Services business to Microsoft”, 25 April 2014. Announcement. ↩
- Eastman Kodak Company, company milestones and 2005 annual report. The interpretation of the governance lesson is my own. ↩
- Amazon, “6 things that make Amazon a ‘Best Workplace for Innovators’”. Description of the PR/FAQ approach. The shortened boardroom exercise is my adaptation. ↩
- David A. Nadler and Michael L. Tushman, “A Model for Diagnosing Organizational Behavior”, Organizational Dynamics (1980). The grouping used here is an adaptation. See also Penn State’s explanation of the model. ↩
- Eric Ries, The Lean Startup (2011). ↩
- Gary Klein, Seeing What Others Don’t: The Remarkable Ways We Gain Insights (2013). See also Klein’s overview of his insight research. His distinction between reducing errors and increasing insights also informs the discussion below. ↩
Leave a Reply