Mention governance to an innovation team and watch the energy leave the room. It sounds like the opposite of the work: forms, boards, people outside the team judging what is happening within it.

Our experience runs the other way. The teams that keep their freedom longest are the ones with the clearest governance, because freedom does not last without it. It collapses inward into drift, or outward into distrust, and either one will end the work.

What governance is for

Strip away the paperwork and governance does three jobs.

It gives the work legitimacy. Plenty of innovation teams have permission in the weak sense: someone said yes once, a budget exists, nobody has objected yet. That kind evaporates under questioning, from above (a new director, a finance query) and alongside (why does one team get to work differently?). Legitimacy is permission that survives scrutiny: a credible answer ready whether or not the team is in the room.

It settles decision rights: who can start a piece of work, who can change its course, who can kill an idea, and at what moment. Most teams are good at producing ideas and evidence; what stalls them is that nobody holds the right to decide, so promising work waits and weak work drifts on. A team that knows who decides can afford to explore, because someone can always call time.

And it protects the team. Some experiments will fail, and a few will fail visibly. Protection is the agreement, made before that happens, about what failure means and who explains it upward. Without it, the first 'bad' result becomes a referendum on the whole team.

Governance that enables and governance that suffocates

One person hears governance and thinks of good work being killed; another sees the path being cleared. Both are remembering something real. This can come down to what their governance was measuring.

Suffocating governance measures activity. It asks a team to prove that nothing has gone wrong: milestone reports tied to dates, approval queues, slide decks recounting effort. Its underlying question is "are you on plan?", which is the wrong question for work whose purpose is to find out whether the plan is right.

Enabling governance measures learning. Its gates are tied to evidence rather than time: the next tranche of money and attention is released when the team can show what it now knows and what it intends to do about it. Its reviews change decisions rather than record them. That is our test for any governance moment. If the meeting could not have altered what happens next, it was reporting theatre, however senior the room.

This is now the official position. The Treasury's Test, Learn, Grow guidance, published in May 2026 as an annex to the Magenta Book, tells departments that governance may need adapting to support "iterative development, proportionate oversight and timely decision-making", with decisions "based on evidence generated through the cycle rather than fixed plans made at the outset". It names the blockers too: rigid funding models, lengthy approval processes and risk-averse cultures. A team asking for evidence-based gates is not making a special plea; it is asking for what the government's own evaluation guidance recommends.

The OECD's Observatory of Public Sector Innovation, which has studied government innovation labs around the world, reached a blunt version of the same conclusion: the labs that survive are the ones with sustained top-level sponsorship, a clear mandate and stable resourcing. The ones on short funding cycles spend their energy justifying their existence instead of doing the work.

The needs change as the work matures

Governance is not one arrangement held for the duration.

Early on, the scarce commodity is cover. A new team has no results yet, so its legitimacy is borrowed from its sponsor, and the sensible move is to build that cover in peacetime, before the first difficult finding needs it.

The Treasury guidance gives the same advice in quieter language: engage senior leaders and assurance bodies early, agree proportionate controls, and settle in advance what evidence would justify continuing, adapting or stopping. Criteria set before the results arrive protect everyone from the pull of sunk effort, and are easier to agree while the questions are still hypothetical.

In the middle of the work the evidence starts arriving, and some of it will complicate the plan. This is where decision rights earn their keep. A sponsor who can respond to evidence with "stop that, continue this" is worth more than any steering board.

Later, the danger changes shape. Mature teams rarely die of bad governance. They die of illegibility: the work is good but the results are not legible to the people who sign budgets, After that, a leadership change or a spending review is all it takes to bring the work to an end. At that stage the job of governance is translation, keeping a running, evidence-backed account of what the team has changed, in the language of the people who fund it.

How this looks inside a Lab

Our version of the innovation team is the Lab: a team, a method and a platform, held on one problem long enough to change it. The Retrofit Policy Lab, which we run with the University of Leeds with funding from Research England, is where we have learned most of this.

The Lab's platform keeps every assumption, every piece of evidence and every insight in one place, so a governance conversation has something concrete to stand on: a sponsor can see what we believed, what we tested, and how sure we now are. Briefings deal in evidence rather than reassurance. And because confidence is recorded belief by belief, the question "should we keep going?" has a better answer than instinct.

Built before you need it

There is a fair objection to all of this. Governance can suffocate, plenty of teams have been reviewed to death, and a two-person experiment does not need a board. The dose matters, and so does the design.

But the answer to bad governance is rarely none. The teams we see running without agreed legitimacy, decision rights or protection are not freer. They are unprotected, and they find out at the worst possible moment.

Which points at the real constraint: governance cannot be retrofitted in a crisis. The agreements that carry a team through its first visible failure have to exist before the failure. So we treat governance as part of the design of an innovation programme, agreed at the start and revisited as the work matures. Done that way, it is the thing that lets a team take real risks, because everyone already knows what happens when one of them does not pay off.

We can help

Designing governance is part of how we set up innovation work at Lighthouse: sponsorship, decision rights and evidence-based gates, agreed before they are needed. If you are standing up an innovation team, or trying to keep one alive, we would be glad to compare notes. Say hello.

Further reading