Three Big Ideas #41

Fiscal folly, field-tested futures, and scaling startups at speed

Welcome to our fortnightly Three Big Ideas roundup, in which we serve up a curated selection of ideas (and our takes on them) in entrepreneurship, innovation, science and technology, handpicked by the team.

---

🕳️ Derin Kocer, Adviser

Since Labour returned to office, every fiscal event has focused on “filling the fiscal black hole.” As Philip wrote last week, the next one will be no different – except, if anything, the black hole only seems to be getting bigger. One obvious way to fill the gap is to raise the rate of VAT, or, more creatively, to broaden the base to which it applies. However, neither approach offers a sustainable long-term strategy to shore up Britain’s public finances.

The Treasury’s real problem isn’t a newly formed deficit — it is an enduring one. Spending has failed to normalise post‑pandemic. According to official estimates, public spending as a share of GDP for 2024-2025 will be at around 44.4%. Although this represents a significant fall from 2020, when the pandemic caused it to spike to 50%, it’s still well above the pre-pandemic levels. The IMF estimates that between 2015 and 2019, government spending averaged 39% of GDP annually. Meanwhile, our debt has also increased dramatically to over 100% of GDP and there is no plan to pay it down quickly. Under Tony Blair, Britain’s debt burden was under 40%.

Our fiscal situation stands in stark contrast to many of our European peers, whose finances returned to pre-pandemic norms more quickly. In Britain, meanwhile, numerous practices designed for the pandemic era were kept in place, which continue to contribute to the unusual and consistent rise in welfare spending. The stealth tax rises Reeves introduced since taking office have paid for these but have also made inflation and high interest rates stickier than elsewhere. We need to get back to normal if economic growth is the core mission.

Rachel Reeves cannot continue to rely on tax grabs to balance the books. She must also stand up to members of her own party and cut spending back to normal levels. It’s worth adding that the Conservatives should give them the ‘political headroom’ to do so, if they too are serious about sustainably fixing the ‘fiscal headroom’ problem every Chancellor has faced in the past decade.

🔭 Anastasia Bektimirova, Head of Science and Technology

I recently took part in two different foresight exercises. One was a simulation game that tasked participants, who role-played as governments, tech companies and scientists, to make constrained choices under uncertainty. Another one was a rapid scenario sprint that asked participants to judge how various shocks would shift the UK’s position across several AI-related strategic fronts. Both were valuable for surfacing trade-offs quickly, exposing coordination gaps and clarifying sequencing. They also shared a familiar limitation of strategic foresight work: generating narratives about the future, but not decision-grade evidence about how people are likely to behave inside it.

In many ways, policymaking, especially in emerging science and technology domains, is the disciplined management of uncertainty – turning incomplete information into choices that aim to protect and create public value. Governments often have to set rules before the behavioural and second-order effects of a technology are observable at scale. On their own, horizon scans and lists of risks and hopes don’t show how people will act within those futures. That requires a different kind of foresight.

A piece published in Nature last week calls this shift “science-fiction science.” The idea is to simulate plausible near-term futures and run controlled experiments inside them to measure attitudes and behaviours before norms and markets harden. Done well, this could offer evidence on how big the effect is, who it helps or harms and which mitigations work. Where this was attempted – most notably around autonomous vehicle ethics – research such as the Moral Machine project helped structure policy debates and, in places, legislation. Where this wasn’t the case – for example, genetically modified foods – public attitudes hardened before behavioural research caught up.

New tools make this approach ripe for experimentation. Take Google DeepMind’s Genie 3, released last week. It’s an example of world-model tooling that can generate interactive 3D environments from simple prompts that can be navigated in real time, with conditions adjustable on the go via a prompt. In theory, this could generate interactive, policy-relevant micro-worlds, lowering the cost of turning priority questions into experiments. That could make it feasible to prototype, for example, a busy high street to examine how delivery robots affect pedestrian behaviour, or a drone delivery corridor to assess noise tolerance and complaint behaviour. The standard of proof would still come from the protocol with clear hypotheses, representative samples, randomisation, incentives and transparent reporting.

Here’s a thought experiment on what an institutionalised version of this could look like. Create a small, cross-government experimental foresight lab that runs or commissions simulations for priority questions and curates the best results as shared benchmarks. Its mandate would be to pre-register hypotheses with policy teams, recruit representative cohorts, and publish effect sizes with uncertainty and distributional impact. It could then endorse high-quality studies as benchmarks, signalling national priorities so departments, regulators and procurement could optimise for them. This would keep foresight close to strategy and delivery, and let evidence travel across portfolios and survive political cycles.

🚀 George Patin, Intern

Over on The Generalist, their team have put together the 2025 edition of The Future 50: the most promising startups valued at or below $200M, as selected by over 200 investors, from all across the world.

One of the most striking things about this grouping is just how lean the teams are. The median team is only 26 people, with many at 15, 10 or even 4 core members. While most are very young companies, many nevertheless boast very impressive multiples of revenue per employee. This broadly tracks with Carta’s earlier report on startup hiring – founding teams are getting very lean indeed.

It is no coincidence that this is happening alongside the rise of vibe coding and better AI in general. Prototyping, early brand design, pitch decks, research — all speed up initial validation and let companies scale faster right away. We’ve already seen Lovable break every speed record to cross $100 million ARR in 8 months. The trend, it seems, is towards companies being spun up practically out of thin air, possibly as fast as a single day.

Policy, too, is gradually adapting to the new speed of startup scaling. The EU’s proposed 28th regime, a harmonised business framework across the entire Union, has now moved into consultation. A project of this magnitude is bound to be challenging, but the goal itself is simple: make company formation faster, compliance easier, and the overall flow smoother.

There is a lot to be said for reducing barriers to innovation. One need only look at Estonia, which Anastasia wrote about here last week, with its recent digital reforms and 10 unicorns to show for it. It isn’t merely removing legislation, either, but making it responsive and closely attuned to the needs of entrepreneurs. If policy is able to match the new pace enabled by emerging tech, we may very well see more Lovable-style stories in the near future.

---