Oli Hammond used to work for us at SyndicateRoom. He's now a partner at Fuel Ventures, leading £1 to £2m seed rounds into UK software. So when he tells me 99% of AI investors are going to lose money, it isn't a hot take for the algorithm. It's how he's positioning real money.
I pushed back on him more than once in this episode. But the core of his argument stuck with me, because it points somewhere I didn't expect. Straight at an edge that angels have and institutional VCs simply don't.
Oli is careful about what he means. This isn't a billion going in and half a billion coming out. He thinks the value gets created. It just lands in very few firms, and everyone else is left fighting over scraps.
The part most people miss is where the competition comes from. The threat that kills an AI startup usually isn't the other funded startup doing the same thing. It's the model the product is built on.
Two years ago the market flooded with low-hanging fruit: top-of-funnel sales tools, marketing helpers, recruitment plugins. Thin products, no protection. The trouble is that the model providers now need to make money, and the easiest way to do that is to build agents that swallow exactly those tasks. As Oli puts it, your lunch is eaten, and no one is using you, because they will just use Claude. There is more on the sector in our guide to investing in the UK technology sector.
This is the useful bit for anyone putting money into the space. Oli runs new AI deals through a short list of green and red flags. It's worth stealing.
| Green flags | Red flags |
|---|---|
| The AI is load-bearing: remove it and the product collapses. | A bare GPT wrapper, with no proprietary data or model-improvement loop. |
| A proprietary data flywheel: the product gets measurably smarter as it scales. | Very horizontal: a general tool competing on the models' home turf. |
| It attacks one vertical deeply, rather than doing a bit of everything. | A moat that closes as the next model gets better. |
| The next foundation model (GPT-6, Claude 5) is a tailwind, not a threat. | There is money in wrapper products, just not the kind of money a VC is trying to make. |
| It automates high-wage cognitive labour: competing with payroll, not just a software budget. |
The same discipline that reads a market reads a founder. Oli's simplest red flag has nothing to do with AI: a founder who tells you there is no competition. Because there always is. It says they either haven't done the work, or they can't bring themselves to deliver bad news. In early-stage investing, both are fatal.
If the models can eat horizontal tools, where is safe? Oli's answer is regulated markets. One of the last places, he reckons, where software can still build a real moat.
Regulation is slow, expensive and different in every country. The model providers have little reason to wade through it. And plenty of regulated firms aren't allowed to feed their data into a public AI tool in the first place. That friction, the thing founders normally complain about, is exactly what keeps the models out.
Here's the turn that made the whole conversation for me.
Oli is judged on fund-level returns. That maths only works if he backs companies that can go from zero to £100m. So a genuinely good business, profitable and growing but topping out at £10 to £20m in revenue, is effectively uninvestable for him. Not because it's bad. Because it can't move the needle on a fund.
That's a whole category of solid companies that institutional VCs have to walk past. And it's exactly where an angel can play. You can back that business and do very well from it, especially with EIS relief taking some of the risk off the downside. Oli's one condition is a good one: do your due diligence, assuming the company never raises institutional money again. Can it stand on its own two feet?
So here's where I land. "99% will lose money" isn't really an argument for avoiding AI. It's an argument for giving up on picking the single winner. If value concentrates in a handful of firms, the way to be exposed to those firms isn't conviction in one name. It's coverage across many. Our white paper on the science of startup investing and our 2025 angel investing outlook both make the same case with the numbers.
How you may want to think about your exposure: rather than betting the house on one call, Access EIS builds a diversified portfolio of 30+ UK startups by co-investing alongside the angels with the strongest track records, with EIS tax reliefs on top. See the data behind the Access EIS fund, or explore Access EIS.
Please note: our office hours are weekdays, 9.30am - 5.30pm.