It's a room I'm sure of you are familiar with.
The kick-off session, the round of introductions where everyone shares three facts about themselves. I have a cat called Lola, my middle name is Patrick as I was born on St. Patrick's Day - I usually make a joke about having a brother called Pancake Tuesday. (I do not have a brother.) I never have a third fact to hand. The joke is so good that I usually don't need one.
Sometimes what follows is the two-hour empathy mapping session that concludes, after much sticky-noting, that people are busy and some of them do not much like computers. Or we try assumptions mapping, where we all watch each other's cursors dance around the Miro board as we pull thoughts from thin air. Half of us aggrieved that we lost/won the vote as to whether have background musak.
What I never really walk out of the early sessions knowing is whether anyone has ever tried to solve this problem before, when that was, or what they found when they did. And what stopped them succeeding. What was it really.
No existing personas to pick up and pick apart. No record of what the organisation already knows, or of what it only thinks it knows. What have the great minds of academia got to say about all this?
Sure, such things have been to known to be uncovered deep into the process (often mentioned in passing). Fossils of the best efforts of entirely capable people. Eureka moments and gotchas! covered in cobwebs.
But, alas, a standing start, more often than not. None of this means the people are not good at their jobs. It means we built a way that quietly expects them to begin again every time.
Why always the blank page?
It is right there in the word. Discovery.
"The action of finding out or becoming aware of something for the first time". But how much is being uncovered for the first time? For some of the people on the call, perhaps, but what about the wider organisation, or somewhere else in government?
Someone has sat face to face with these users. Someone has mapped this journey or the journeys the dissect the wicked problems of public design. Past research scatters across shared drives, slide decks and in the heads of people who have moved on, and when a team disbands the memory tends to leave with it.
It's not what's unknown. It's what's unowned.
I've had Senior Policy people describe Discoveries to me as "waiting for your team to catch us all up". We've all had the stakeholder who seems weary from the very first show and tell, who it turns out is on their third go round on the same problem. They are not being difficult. They have just been patient for a very long time.
We are fifteen years (or six prime ministers) into doing all this government digital stuff. We have mature, internationally admired design systems. We have over 120 services hanging off a single sign in, shared notification and pay platforms. That is an enormous amount of hard-won learning.
Imagine if it compounded, each project starting a little further up the mountain than the last.
The system was never designed to remember.
The responsibility hasn't changed
The tools are changing. The expectations are changing. The responsibility has not.
AI tooling can help with the designing itself, and with holding an organisation's research memory (though it's not as simple as stuffing folders of 'stuff' in your fave GPT - more on that in later posts).
The expectations are changing. Speed, and the pressure to release value sooner. We are no longer simply digitising and transforming old systems or work that fits nicely in the business case. 'Lift and shift's are difficult to defend against the backdrop of changed economics and cadence.
What has not moved is the responsibility. Rigour still matters. Evidence still matters. Talking to real people still matters most of all. None of that is up for negotiation, and continuous learning done badly is just guessing with better production values.
The most stubborn-yet-fixable problem, in my opinion, is the research amnesia. Not just 'storing' research, but making years of scattered, messy material findable and usable in the moment it matters.
Old interviews, reports and analytics can be brought together, synthesised and actually interrogated, so a team can ask what do we already know about this and get a real answer.
This is the one place AI clearly earns its keep: not inventing insight, not standing in for the people we should be talking to, but handing an organisation back the memory it always had but could never reach.
This is the work we at Lighthouse are doing now with the organisations we partner with, and it starts well before any platform. It begins with an audit: finding what already exists, making sense of it, separating what is still relevant from what is stale, and being honest about the gaps. Only then is it worth holding in a living record the team can act on and keep current.
That record is what our Platform is for. And there is no real reason for it to stop at a team or a department. The problems government wrestles with, the wicked ones around crime, health and housing, span boundaries. There is no technical reason the insight cannot span just as wide.
It is not only memory
Memory is not the only thing holding the old model in place. (⚠️ Trigger warning: real talk coming up...)
Destiny
The decision is often made before Discovery even starts, so the research becomes a way to confirm a direction rather than question it. White Papers, with ministerial 'lives' on the line become an unstoppable train, with teams hurriedly laying tracks in front of it.
Logistics and risk
Teams are often brought in and the risk of losing the expertise and insight at the phase overlap is so real that often phases 'roll on'. I've been in (I think) 6 Discovery assessments, none of which happened as the original theory intended. I've been two weeks away from wrapping an Alpha and had to sit through the 'formality' of a Discovery checkpoint.
Jukeboxes
And, as hard as it is to hear, we need to consider some of the perverse incentives that would need to be overcome. External consultancies are paid for bums on seats x time. Where is the motivation to radically change that model? It's more lucrative to start afresh, which is at odds with agencies leveraging experience and track record as a differentiator. Is the taxpayer playing to hear the same tune again and again?
Economics
And the economics have quietly flipped. The phases were a good answer to a real problem. They put a stop to the bloated, up-front, requirements-heavy waterfall that came before, and forced teams to understand a problem before committing to a solution. But prototyping is fast and cheap, the cost of being slow has risen, and a fixed ten-to-sixteen-week Discovery and Alpha is no longer obviously the careful option. (More on all that AI design-y stuff in later posts).
What this could unlock
Government still funds and commits in a fundamentally waterfall way. The big suppliers are still working out whether the savings AI brings ever reach the client, which in the public sector means the taxpayer.
Solving the memory problem is a quiet lever on both. When the client owns a compounding memory, the thing worth paying for changes: not the same rediscovery over and over, but the sense-making that turns it into something usable, and the work of keeping it alive.
And the politics can move too. The current system rewards the new, the fresh initiative, the launch, the same bias that sits inside the word discovery. A memory that compounds lets a commitment build openly on what came before, and credit be claimed for progress that adds up over time. The kudos is still there. It is simply earned for stewardship, rather than for another standing start with a new name on it.
None of this is a fringe view any more. The centre now talks openly about a state that learns continuously, and the latest Treasury guidance, the Test and Learn annex to the Magenta Book, tells teams to lean on what is already known and to agree when to stop before they begin.¹.
We cannot learn continuously if we cannot remember. The blank board is a choice, not a law of nature, and for the first time we have a real chance to stop beginning right at the beginning.
Now, without checking, what is my cat called?...
¹ HM Treasury, Test and Learn: Magenta Book Annex (May 2026), authored by the Cabinet Office and the Ministry of Justice Analytical Directorate.