AI Investing: What Actually Works for a UK Pension
- Alpesh Patel
- Apr 20
- 7 min read
Updated: May 5
The best piece of AI I use in investing didn't pick a stock. It read a 1,200-page Goldman Sachs year-ahead outlook in seven minutes and told me the three things that actually mattered for a client's pension.
That is the honest answer to what is AI doing in investing. It is not crystal-ball forecasting. It is not a robot trader. It is not a secret algorithm that beats the market. Mostly, it is a very good research assistant that finally lets a private investor see the same evidence a private bank sees, without paying private-bank fees.
I have been running money since 1995. I have watched two dot-com booms, the LTCM blow-up, 2008, 2020 and the 2022 drawdown, and every iteration of the retail-investing gimmick that promised to replace judgement with technology. Most of that dies on contact with the next bear market. The bit that is not dying, and the bit worth your attention, is the narrow, specific, unglamorous work AI is now doing on the long-term side of the business.
This is what I actually use with members on the Great Investments Programme. I'll walk you through it.
Speed is not the edge in AI investing
For twenty years, financial technology has been obsessed with milliseconds. High-frequency trading (HFT) gets the magazine covers, the Michael Lewis books, the Hollywood treatment. It has made fortunes for a handful of quant shops with co-located servers next to the New York Stock Exchange.
None of that matters to you if you are trying to build a pension.
HFT is a closed arms race between institutions, extracting tiny slivers of profit from the bid-ask spread millions of times a day. The entry ticket runs into tens of millions of pounds of infrastructure. A retail investor cannot participate, would not want to, and gains nothing from any of it.
The useful thing happening in AI for retail is the opposite direction. We call it low-frequency investing (LFI). Months and years, not milliseconds. Research, not reaction.
Worth saying out loud, because the marketing around retail AI products is usually trying to borrow the glamour of HFT. If an AI product is pitching you speed, it is selling you the wrong thing.
What the machine actually does
On the research side, the useful work splits into three jobs. All three are things a private bank would pay a team of analysts to do. None of them are glamorous.
Scoring 17,000 stocks on What Matters
Every night, my systems score around 17,000 global stocks on fourteen institutional factors. Not fourteen factors I invented. Fourteen that Nobel laureates, CFA curriculum authors, and the editors of the Journal of Finance have been writing about for decades.
The list includes CROCI, the cash-return-on-capital metric Goldman Sachs's equity research team uses to separate companies generating real cash from companies generating only accounting profit. It includes the Piotroski F-Score, a nine-point check on financial strength developed at the University of Chicago. It includes the Altman Z-Score, a bankruptcy-risk model that's been running since 1968. And it includes eleven others, among them Sortino, Sharpe, Calmar and PEG.
The output is a monthly shortlist of around 200 names. The job the AI is doing is not "picking winners". It is filtering out the 16,800 stocks that fail on quality, solvency, valuation or risk.
My father ran a chemicals company called Ishan Dyes, eventually listed on the Bombay Stock Exchange. His pride was not what he put in. It was what he took out. Purification was the product. The same logic applies to data.
Following Money that Actually Makes Money
Every quarter, every US hedge fund with more than $100m of assets files a 13F disclosure with the SEC listing what it owns. Publicly. For free.
The problem is that there are thousands of them and most hedge funds are mediocre. So we filter. The AI only tracks hedge funds that have delivered at least 40% per annum over the past three years. That is a brutally short list. We then cross-reference what those funds are buying with what Goldman, Morgan Stanley and JP Morgan are simultaneously upgrading in their sector notes.
When three major banks upgrade a sector that high-conviction hedge funds are also accumulating into, that is a signal worth looking at.
This is not tip-following. You are not copying a trade; by the time a 13F is filed, the hedge fund has owned the position for weeks. What it is, is a read on institutional conviction at the sector level, which is durable on an LFI horizon.
Reading the Research Nobody Reads
The largest private banks each publish an annual outlook that runs to 1,200 pages. Morgan Stanley, UBS, Goldman: beautifully written and extravagantly ignored. I know I am not going to read a 1,200-page PDF. Neither are you.
Custom LLM agents now read all of it overnight and pull out the specific disagreements, the sector consensus, and the handful of numbers that actually move a portfolio decision. The output is a three-page note, not a 1,200-page brick. That is AI doing something it is genuinely good at: compression.
10,000 futures, not one forecast
The most dangerous number in retail finance is the "projected return". Your pension statement says 5% a year and you picture a straight line. Markets don't move in straight lines.
This is how Long-Term Capital Management blew up in 1998. Two Nobel Prize winners sat in the LTCM offices, ran models that assumed market returns followed a lognormal bell curve, and were wiped out when Russia defaulted and the tails of that curve turned out to be fatter than their textbooks suggested. Fat tails (extreme events) happen far more often than an A-level statistics course will tell you.
Instead of one projected number, we stress-test every model portfolio through 10,000 simulated market paths. Some paths are historical: 2008, 2020, 2022, the 1973-74 bear market. Some are synthetic scenarios designed to find where a portfolio breaks. Members see the full distribution: the good case, the median, and the 1-in-20 worst case.
That last number is the one that matters for anyone with a pension, because it is the number that tells you whether you will panic-sell in the next drawdown. If you can see upfront that a position would have fallen 21% in 2008 and 11% in the 2020 pandemic shock, you are far less likely to sell it at the bottom. The goal is not to see the future. It is to see enough possible futures to stop pretending there is only one.
Why the Member Never Sees the Machine
There is a real risk in all of this: the black box. An algorithm that spits out a buy signal without showing its working is a liability, not a tool. It is also a regulatory non-starter in the UK.
Our architecture has one non-negotiable rule: AI proposes, humans decide. Every portfolio candidate, every ranked list, every pattern signal goes to an FCA-regulated fund manager for review before a member ever sees it. A separate, fully independent AI system (different memory, different architecture, different training data) audits the primary model's output. Where the two disagree, the signal is thrown out.
On the communication side, the 1,200-page outlooks, the 10,000-path simulations, and the 200-name shortlist all get compressed into three things the member actually uses: a five-minute video, a one-page infographic, and three bullet points. A private bank's work, translated into language a human with a pension can act on.
What this is not
Worth being clear about what this is not, because the AI-investing pitch is crowded with things that deserve to fail.
It is not HFT. We do not compete on speed and have no interest in doing so.
It is not a black-box automated platform handing you an opaque portfolio. Every layer of the logic is explained.
It is not autonomous trading. No algorithm connects to your broker. Members click every buy and sell themselves.
It is not a promise of returns. Markets are uncertain. What we offer is a better-informed decision. Faith-based investing has a poor track record.
The bottom line
Is AI useful for investing? There is a boring answer and an interesting one.
The boring answer: yes, if you use it to read research, filter data, stress-test portfolios and communicate risk. No, if you expect it to predict next week's FTSE close.
The interesting answer: the useful version of AI in investing is the version that slows you down. It gives you the evidence a private bank would give a £10m client, translated for an ordinary pension, so that when the next drawdown arrives you stay in your seat.
If you want to see how the framework works in practice, the Great Investments Programme is where my team and I run it. There is a free introduction you can watch, and if you are trying to decide whether your current pension or ISA strategy is built to survive the next fat tail, that is a reasonable place to start.
Sources
Financial Conduct Authority (FCA), UK conduct of business rules: fca.org.uk
US Securities and Exchange Commission, Form 13F disclosure rules: sec.gov
Piotroski, J.D. (2000). "Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers." Journal of Accounting Research, Vol. 38.
Altman, E.I. (1968). "Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy." Journal of Finance, Vol. 23, No. 4.
Goldman Sachs Global Investment Research, CROCI methodology.
AQR Capital Management, research on fat-tailed return distributions.
Morningstar and SPIVA, active vs passive fund performance data.
Lowenstein, R. (2000). When Genius Failed: The Rise and Fall of Long-Term Capital Management. Random House.
Pensions and Lifetime Savings Association (PLSA), UK retirement income standards.
About the author
Alpesh Patel OBE is a former hedge fund manager, author of 18 books on investing, a Financial Times columnist, a former Bloomberg TV presenter, and a former Visiting Fellow of Business and Industry at Corpus Christi College, Oxford. He founded the Great Investments Programme to give private investors access to the same quantitative rigour that institutions use. He was appointed OBE for services to the UK economy.
Disclaimer
This article is for education and information purposes only. It does not constitute a personal recommendation or regulated financial guidance. Past performance is not a reliable indicator of future results. The value of investments can fall as well as rise, and you may get back less than you put in. If you are uncertain about any investment decision, speak to a qualified professional.
Comments