What a lab is for
A lab is not a branded room with sticky notes. It is a contained piece of a system, run under observation, for long enough to learn something that matters.
When a client asks us for a lab, the first thing we do is try to talk them out of it. Not because labs don't work — they do — but because the word gets used for a shape of work it cannot carry.
An innovation lab, in the sense we mean it, is not a room. It is not a month. It is not a sprint with an ambitious name. It is a contained piece of a system, run under observation, for long enough to learn something that matters.
That definition does a lot of quiet work. Let's go through it.
Contained
A lab is bounded. It has a scope, a set of people, a pathway, a budget, a timeline. What is inside is being studied. What is outside is not. The containment is not an administrative nicety — it is the condition under which learning is possible. Without a boundary, a lab becomes indistinguishable from ordinary work, and its findings become indistinguishable from ordinary opinions.
A piece of a system
Labs are not pilots. A pilot tests a single product or service. A lab tests a relationship — often between a service, a policy, a funding stream, a governance structure, and the people moving inside all of them. If the thing being tested can be described in a single noun, it's probably a pilot.
Run under observation
Labs are instrumented. Evidence is gathered continuously — qualitative, quantitative, and structural. The instrumentation is designed before the work begins, not bolted on at the end. Everyone in the lab knows they are being observed, and knows what for. Surveillance culture this is not. It is more like a surgical theatre: the room exists because people are watching.
For long enough
The most common failure mode of a "lab" in the conventional sense is its timescale. Systems don't move in four-week sprints. The interventions that produce genuine shifts — relational trust, new decision rhythms, durable behaviour — reveal themselves on a time horizon of months, not weeks. A lab that ends before the system has had time to respond is a piece of performance, not inquiry.
To learn something that matters
The last condition is the one most often missing. A lab's output is not a recommendation. It is evidence that changes what the organisation now thinks is true. If the findings don't change anyone's mind, the lab didn't do its job — or the organisation wasn't ready to treat it as a lab in the first place.
Why does this matter? Because the word "lab" now does a lot of rhetorical heavy lifting inside institutions that want to signal their willingness to try something different without actually opening themselves to what that would mean. A real lab is generous, serious, and a little dangerous. It is not the safe space. It is the place where the organisation agrees, in advance, to be surprised.
If you're thinking about commissioning one, the first test is simple: can you describe what you are prepared to be wrong about? If the answer is nothing, don't call it a lab. Call it what it is.
Writes in public about the receiving system, systems transformation, and the quiet patterns that decide whether change sticks. More essays in Praxis.
Continue reading
Five Principles for AI in Public Service
AI enters government work through a governance architecture or it does not enter at all. Five principles — advisory, auditable, deliberative, sovereign, secure — written to be defensible at a hearing.
ReadNotes on the first three weeks of a system-change engagement
The most important decisions on any engagement are made before the plan exists. Here is what we try to notice in the weeks before the work is named.
ReadWorking on something like this?