What do you get if you throw together a pioneering machine learning researcher and a top former Goldman Sachs quant? The short answer: the world’s first platform that lets investors easily create and trade their own AI-powered portfolios, with an interface facilitated by natural language processing (NLP). The duo’s brainchild is called SkopeAI and The Stack is making it our sixth “one to watch” – our regular feature on early stage startups that excite us with technology-led innovation that we think could become widely known.
Former Goldman Sachs Managing Director Dr Erkko Etula left the investment bank – where he ran strategic and quantitative asset allocation research – in May 2021 after a decade to co-found SkopeAI. He had crossed paths with former NYU adjunct professor Dr John Nay when Nay was building what he describes as “data feeds and portfolios for investment managers to power systematic trading strategies”; a process that involved meticulously labelling millions of SEC filings and other public data from listed companies for machine reading.
Follow The Stack on LinkedIn
Joining forces this year, the two set about building an investment platform powered by institutional-grade AI and quantitative investing technologies. SkopeAI, they say, will help investors avoid the fees they might otherwise pay to off-the-shelf ETFs, mutual funds, or fund managers, and instead use AI to build portfolios that can “express your values, interests or goals” via a completely customized AI-Traded Fund, or an “AITF”.
(Better yet, investors using it will directly own the stocks they’re trading, unlike in an ETF or a mutual fund, to further increase tax efficiency and to lower fees…)
The platform, they say, will scan millions of data points to construct an optimal portfolio based on users’ search terms, employ stock-level tax loss harvesting traditionally reserved for much larger and more sophisticated investors, and even trade on your behalf. The two say their team has refined the platform while powering billions of dollars of traded assets for corporate partners. We sat down to find out more.
What’s SkopeAI’s raison d’être in a nutshell?
Erkko: “I come from the investment management industry. John comes from a background where artificial intelligence and natural language processing (NLP) are traditionally being applied to institutional needs.
“The big realisation for me was that in today’s world of fractional share trading and near-zero transaction costs, we can use technology to give nearly everyone the personalised and tax-optimised investment experience enjoyed currently only by the wealthy. The one piece missing from this equation was how to translate the individual goals, values and interests of millions of investors to fully customised portfolios at a massive scale — and that’s where John’s expertise in natural language processing (NLP) is so crucial.
“Put differently, we can offer bespoke investment advice to practically everyone because our technology understands natural language. That puts SkopeAI in a unique position to be the investment advisor of choice to millenials and gen-z who are independent, curious, expressive, and expect intelligent self-service. So, we founded SkopeAI on the guiding principle of empowering the investor. By disintermediating middlemen, we can free up value and deliver that directly to the end client.”
Read this: Swiss stock market operator SIX’s Head of Markets Thomas Zeeb on the digital future of exchanges.
John: “We’ve seen a few really important trends here. One is broader participation in markets in the US, by the end investor. Another is trust in AI and machine learning, especially with regard to managing money. So we’re at this inflection point where you don’t necessarily need a lot of intermediaries. There’s a proliferation of thematic ETFs; there’s trillions flowing into thematic ETFs or other non-market cap-based indexes. There is little need for an intermediary like the ETF provider or the fund manager. Our platform can do what they do, but better and cheaper because we can use AI to do the stock selection, optimize portfolios and cut out ETF fees. People often miss the fact that even if they’re using a robo-advisor, that robo-advisor is putting them into ETFs and they’re paying the ETF fees in addition to the robo-advisor’s management fee, with little to no added personalisation.”
Do you think the retail investor market is sophisticated enough to want to be doing this? ETF fees may be a pain, but there’s a simplicity to investing in a product like that, which is attractive to many.
Erkko: “Certainly in the US there is a great benefit to owning stocks directly. Even putting the ETF-type fees aside, you can actually increase the after-tax return of the strategy by doing active tax loss harvesting, and doing that at the individual stock level. So in essence, you would be matching the return of the ETF on a pre-tax basis, but actually outperforming the ETF on an after-tax basis; this is one of the things that family offices and high net worth investors have been doing for decades. Through technology, SkopeAI can give that exact same experience — helping increase the after-tax return – to the smaller investor as well.
“It’s not what you make, but what you keep. I also think that personalising your investments makes you more connected to your portfolio, which in turn gives you more staying power through turbulent markets.
John: “Another key point here is ESG. We’ve seen a lot of products released with an ESG focus, like a low carbon ETF [e.g., the Carbon Transition Readiness ETF], and investors going ‘what’s the deal with this, it looks just like the S&P 500?’ Because in terms of the stocks in there and their weights, the thing is pretty close to the S&P 500. And that’s because investment organisations are really worried about tracking error; about how far this might stray from the exact return path you see of the S&P 500. And a lot of people don’t like that! They want to move their investments further towards companies that have low emissions, or make more surgical changes — for example, oil companies may be popping up in certain ESG ETFs because the stock selection may be driven by financial risks, not the actual ESG impact of the company. That’s one example of where we see real demand for that kind of customisation. Some individual investors don’t care about benchmarks like the S&P 500, they want to see impact.”
Let’s lift the hood a little. It sounds like you will need a tonne of compute and harmonised data. Tell me what the stack looks like here?
John: “That starts around five years ago as I was finishing my PhD and some of the research I was doing around natural language processing, and machine learning. We were building models of company data, like 10Ks (the annual filing in the US for a public company); earnings call transcripts and other textual information, and linking them to events in the world like legislation and regulation.
“We were applying NLP to challenges like ‘what does this company really do?’ ‘What are its risks and opportunities?’ ‘What might materially impact that company. That obviously led to potentially actionable data and predictions related to financial markets. This was helpful to investment managers and investment banks: identifying which stocks would be impacted by an election, or by regulation that was being proposed, for example.
Dr Louise Beaumont: Financial services has long been a “fat complacent oligopoly with zero innovation”
“We now provide data-driven historical simulations of investment performance to our users. The only way you can do that is if you have point-in-time historical data across everything because you need to be able to model how investors saw things in the past. That’s taken years to curate, refine and make sure it’s all validated; that serves as the backbone to then build this AI platform on top. Yeah, it involves a lot of compute. We use the cloud to analyse the data and have it ready in real-time, so as a user interacts with it using NLP, it’s querying millions of data points to determine what’s relevant to them. We primarily use AWS for computing and for data storage.
“In terms of the data, we’ve done years of validation and post-processing and use a variety of database solutions depending on the type of data, whether it’s text data or numerical, and whether speed is a bottleneck.
“Over the past five years, we’ve developed proprietary machine learning models that remove unhelpful sentences from public filings, for example. We also have data that’s proprietary; data that we’ve paid people to label. So for example, “this specific company is extremely relevant to this particular theme” – we have millions of data points of that nature.”
Take us from that to portfolio optimisation?
Erkko: “On the optimisation side, the big challenge out there has been how to combine the expected returns and risks of your investments with other preferences of the investor. Typically, it has just been those first two parts. Your traditional ways of constructing a portfolio start from estimating the expected returns and the risks — and you try to optimise the amount of expected return per unit of risk, however defined. Now, what we are bringing in is what you could call the third leg of the stool: interests, values and all of the other preferences of the individual investor. Back at Goldman Sachs, we would have extensive conversations with each client to understand what those additional preferences are, and then incorporate them in their portfolio.
“What we built with John is the ability to translate the values or interests expressed by each individual directly into the objective function that we’re maximising. We use a technique called ‘robust optimisation,’ which helps us incorporate the expected returns, the risks, uncertainties, interests, and values into the portfolio in a systematic and consistent way.
“Someone might care about companies that promote gender diversity, while another person could have a thematic interest or bet, like, ‘I want to own manufacturers of patio furniture!’ You can imagine the spectrum that these interests or values may take. To enable that systematically across our client base is huge. In addition to optimization, the technology stack also has to accommodate risk management algorithms; we’re using institutional-grade systems to manage risk and to dynamically adjust the portfolio over time as markets change.
Let’s just pivot to the front-end quickly. We’ve seen a lot of sophisticated projects fail simply because their UI/UX was sp awful that nobody would adopt them. Have you… hired designers?
John: “Great question! We’ve abstracted away a lot of details – while making them available to people who want to see them – but we’re really making it as simple as a Google search, illustrating the results and the optimal portfolio in a really understandable way, and then clicking through to invest.
“We’ve worked with a designer who was previously at Facebook. And we’ve worked for months designing it just to make sure it’s user friendly and then spending the past few months converting that into code. We are also going through tests with a lot of different users of different backgrounds. We’re hyper-tuned into building an intuitive and beautiful user experience.
How big is the SkopeAI team – and when are you launching?
John: “The SkopeAI team right now is about 15 people, mostly product people, but we have started to build out compliance and operations too. Tests are expanding throughout the fall and then we’re planning on launching more publicly in 2022.”
Erkko: “We have also been working with select family offices over time to refine our investment process, calibrate our algorithms, and to battle test our technology with real capital. This is where our roots are so we will continue serving this clientele even as we begin to gradually expand the SkopeAI membership.”