You want innovation. You built a system that quietly forbids it.

I was in a meeting yesterday when a leader walked the room through the importance of KPIs inside their goals. Sensible, clearly told, the kind of thing that gets nods. So I asked a few questions, because the questions are where the truth usually hides.

Were the targets a step-change, or the next dot on the plot-line? Were they expected to be hit? And what were the real and the perceived consequences of missing one?

The answer came back clean and confident. Goals should be achieved. So targets are calibrated to be hit.

I let that sit for a second, then asked about innovation. About discovering genuinely new approaches, the kind that create value long after the quarter they were found in. The leader lit up. Oh, we have loads of examples of that. Innovations that changed our performance for years.

And there it was. The same person, two minutes apart, describing two things that do not really get on with each other, and each said with total sincerity.

The thing you designed for is the thing you get

Put those two answers side by side and they start to argue.

We track KPIs. We report KPIs. We reward hitting KPIs. And to make sure they get hit, we calibrate the targets so they can be hit. That is a system, and it is a good one, at doing exactly what it was built to do: produce reliable, predictable delivery of things we already know how to deliver.

Then, in the next breath, we say we also want innovation. New thinking. New approaches. Value that did not exist before.

But innovation, by its nature, means doing something you have not done before, whose outcome you cannot calibrate in advance, that might not work. It requires thinking differently, acting differently, and being genuinely comfortable with risk and failure as part of a balanced portfolio of bets. Every honest definition of innovation has uncertainty at its core.

So we have built a machine that rewards certainty, and then asked it, nicely, to hand us some uncertainty. We want one thing and design for another, and then wonder why the innovation feels so rare and so hard to come by.

Where your best innovations actually came from

This is the part I always find quietly revealing. Ask a team where their proudest innovations came from, the ones that changed performance for years, and then look honestly at how they actually happened.

Almost always, they escaped the process. They happened in the gaps. A skunkworks nobody was tracking. A stubborn individual who ignored their KPIs for a fortnight. A crisis that suspended the normal rules long enough for someone to try the thing. A side project that was never on a goal sheet, because if it had been on a goal sheet it would have had a calibrated target, and a calibrated target is the enemy of the leap.

Your innovation did not come from your goal system. It survived it. That is not a story about how well your system works. It is a story about how much talent you have despite it.

The science is not ambiguous about this

This is not a matter of taste, and I am not the first to notice it. The research is unusually clear.

Locke and Latham gave us the most replicated finding in organisational psychology: specific, hard goals beat vague ones. But their own students found the crucial exception. Seijts and Latham showed that on complex, novel work, the kind where you do not yet know the method, a specific performance target actively hurts. People grind on the number before they have discovered how, and the pressure to hit it crowds out the exploration that would have found the answer.

Which is why, on that kind of work, we do not dress the uncertainty up as a goal at all. You cannot promise a number you have no method for yet, so we stop pretending you can. Instead the work becomes structured discovery: name the assumptions the idea rests on, test the riskiest one first, and build real confidence in the belief before anyone commits a target to it. It is not “hit 34 percent.” It is “find out whether 34 percent is even the right thing to chase, and how we would get there.” Discovery is a process with its own rhythm, run alongside delivery, never squeezed into a key result and graded as if it were one.

And the dark side has a name too. “Goals Gone Wild” catalogued what happens when you bolt hard targets to consequences: tunnel vision, gaming, the quiet erosion of the intrinsic motivation that creativity runs on. When missing the number is dangerous, people stop reaching. They sandbag. They pick the target they know they can clear, because in the system you built that is simply the sensible thing to do. You did not hire cowards. You built a machine that quietly rewards caution, and it is doing its job beautifully.

The magic is getting all of it, by design

The good news, and the reason I still enjoy this work, is that this is a false choice. You can have reliable delivery and real innovation, engagement and discretionary effort, ambition and safety. You just cannot leave it to chance and hope. You have to design for it on purpose. Four moves do most of the work.

Separate the kinds of work. Delivery of the known and discovery of the unknown are different games with different rules. The known work gets outcome targets. The uncertain work runs as experiments, with learning goals, on its own track, judged on what it taught, not on whether it hit a number it was never in a position to promise. Stop grading a bet as if it were a delivery.

Judge the portfolio, not the bet. No investor evaluates a fund by whether every single position went up. They judge the spread. A culture that punishes every failed experiment individually will, rationally, stop running experiments. A culture that expects a portfolio of bets, most small, some larger, a known fraction of which will fail, keeps betting. The failures are priced in. That is what a balanced risk portfolio actually means, applied to goals.

Grade, do not score. When success is a decimal you have to hit, ambition is a threat to be managed down. When you grade the quarter honestly, Excellent to Bad, a bold miss can outscore a safe win, because you are judging the quality of the attempt and the learning, not just the arithmetic. This single change is what makes ambition safe rather than reckless.

Make the reward system tell the truth. If you reward only hit KPIs, you will get only safe KPIs, whatever your posters say. Reward the behaviours you actually want: the honest experiment, the smart pivot, the killed project that freed up resource, the insight that will pay off two quarters from now.

Do those four, and the change is one you can feel in the room. People stop defending safe targets and start reaching for real ones, because reaching is finally safe. Creativity has somewhere to live. Engagement rises, because the progress principle is real: people are at their most motivated when they are making meaningful progress on work that matters, and quietly clearing a safe target is not that. The discretionary effort, the part you cannot mandate and cannot buy, turns up on its own. Teams lean in.

I call this safe ambition, and it does not happen by accident. It is a design choice.

Why AI has raised the stakes

I could have written most of this essay five years ago. What has changed is that the cost of getting it wrong has quietly gone up, and AI is the reason.

For most of business history, advantage came from having more: bigger teams, more skills, more capital, more reach. That era is closing. When everyone has access to the same capable models, the average is free and universal, and having more people doing average work is not an advantage, it is an overhead.

The advantage now comes from the things the average cannot give you: genuine innovation, real creativity, and the speed of your learning loops. How fast can you form a hypothesis, test it cheaply, read the result honestly, and go again? That loop is the game, and AI has collapsed the cost of running it to almost nothing. You can test in an afternoon what used to take a quarter.

Which is what makes the contradiction I sat in yesterday quietly costly. A system calibrated to hit safe targets does not only fail to innovate. It wastes the one advantage AI actually hands you. You have been given a machine that makes learning loops nearly free, and then pointed it at a culture that punishes the very failures those loops run on. The tool is sitting there, ready. The system politely will not let anyone use it.

This is where our work and this argument meet. The same operating system that remembers what a company learns is the one that makes safe ambition operational: it separates delivery from discovery, holds the portfolio view, grades honestly, captures what every experiment taught so the next loop starts smarter, and keeps humans in the loop on the calls that carry stakes. Innovation stops being the thing that escaped the process and becomes the thing the process is for.

The one line to leave you with

You cannot reward certainty and expect uncertainty. If you want innovation, creativity and progress, you have to design a system where ambition is safe, because in the age of AI, intentional safe ambition has never been more needed, or more possible.

Stop wanting one thing and designing for another. Build for the thing you actually want.