“Some things may have been tried before their time, but if these things don’t violate the laws of physics, they are likely to prove possible the next time around. Engineering is a series of failures with an occasional success. At least the kind where you are really looking at new technology. You tend to try things. You try things that are extrapolations of what has happened before. A lot of them don’t work. Occasionally, you hit one that does.” Gordon Moore, March 2000 – (1929-2023)
The last twelve months have seen the venture capital industry (and startup ecosystem more broadly) take one hit after the other. None more so than SaaS business models and fintech startups.
In prior periods, we all witnessed anomalies of extreme exuberance and poor investment discipline (e.g., Theranos and WeWork ). There were also perceived outliers within fintech. Little did we know these anomalies were not the cresting point of a great reset, but the early warning signals.
I will discuss the Covid craze only in passing – it happened over twelve months ago anyways. In hindsight, as far as venture investing is concerned, this event will have proven to be less significant than previously thought with a rate of digitalization being accelerated somewhat, then reverting to the mean somewhat. Covid was initially seen as a momentous secular reset for venture investing, and it has turned out as a precursor for where we find ourselves today.
Next came the end of low interest rates. This end has had and will continue to have momentous implications for our industry and for startups. Tech stocks took a tumble in the public markets, mature privately held startups next saw their valuations plummet. Tech assets dove between 20% and 80% (depending on the quality or maturity of the asset). SaaS business models were particularly hit hard given how recurring revenue is valued in direct relation to interest rates.
Private assets valuation woes directly impacted late-stage investors that had rushed into the venture space (hedge funds, sovereign wealth funds, large private equity investors) either with direct investments in startups or investments in venture funds. Many of these investors are unlikely to return to the space with the same wanton abandon as they did before the music stopped. This means median startup valuations will be more grounded going forward.
On the demand side of the equation, the maturity of the Web 2.0 computing paradigm means there are less interesting ideas to invest in overall, although certain verticals such as fintech will still see plenty of opportunities. To be sure, there are still interesting ideas to invest in, but they will be far and few between and will increasingly occur in niche verticals or markets. The potential of the Web 3.0 paradigm is still being tested, with early returns somewhat lackluster. We have seen the implosion of the crypto space with prices plunging, as well as the dramatic demise of large centralized actors (FTX, Silvergate, Signature, Three Arrows, Luna…). The miserly development of the Metaverse (the Facebook experiment being a perfect example), has not provided a counterpoint to the distress in crypto. However, promising use cases (digital and decentralized identities, tokenized assets, decentralized capital markets and asset management value propositions…) still provide a north star, but they are struggling to emerge in the midst of the greater carnage.
To spice up our life, SVB recently exploded. Rising interest rates conspired to blow the bank’s bond portfolio askew, with management hubris, supervisory shortcomings, and a US tech universe that fell prey to a classic tragedy of the commons all led to the first bank run of the digital and social media era. Decisive action from US regulators may have avoided a wider crisis, especially with mid-sized and regional banks.
Astute observers also pointed to another cause for the SVB explosion. Indeed, with a widespread acceptance and use of mobile apps, faster payments, API and cloud technologies, one can near-instantly move one’s monies from one bank to another, from one bank to a money market fund, from one bank to a non-bank entity (see Atomic Vest, Zamp Finance and ModernFi) . This seems to be what happened prior to SVB experiencing terminal liquidity issues. Depositors, faced with banking accounts yielding next to zero yield, naturally moved their cash to instruments yielding up to 4%. Therein lies one very important lesson. Technology change always happens in ways that are difficult to predict and oftentimes in ways that radically change a given business model. In this instance the business model of a bank is now radically changed. A bank’s ability to earn outsized NIM in a relatively high interest rate environment is now an increasingly obsolete business model. Gone are the days where deposits were sticky to a world where one can now move monies almost at will. This is a very profound outcome. Fintech has delivered change to financial services in a way few incumbent banks predicted it. We shall come back to this very important point about the potential impact of technology change.
Finally, I will add to this great reset the advent of generative AI. Some will remember the rise and promise of expert systems in the field of artificial intelligence in the mid-seventies. This promise never materialized from a venture investment or operational point of view– much money was invested and wasted as a result. It seems that generative AI is unlikely to suffer the same fate. Early results are quite astonishing. Even off the limited data sets, the first generations of generative AI engines increase productivity of software engineers and coders by over 50% while increasing productivity of marketing/report writing/copy/presentations… by over 30% (per several studies). These results are quite staggering, and we are only at the very beginning of the learning and innovation curve.
We have to keep in mind that we have lived in a world with diminished productivity gains, even when applying the latest computing power gains to our economies (personal computer, smart phones, edge computing, cloud computing, APIs, social media, aggregation models borne out of Web 2.0, faster and more powerful processing powers, miniaturization). The early results from generative AI may blow the fear of ever more anemic productivity gains out of the water. What generative AI holds is as important as the printing press, the steam engine or electricity. It can re-invent how we learn (education, training) as well as how we organize ourselves at work (from that vein, the Coase Theorem may have to undergo some rejuvenation). What is even more intriguing is that generative AI’s main impact is and will be on mid to high skilled labor where intelligent but repetitive work is concentrated and where pure creativity takes a back seat to skill and knowledge.
Others have also elevated generative AI to the level of the steam engine, electricity, and printing press. Is that a correct assumption? The first two were foundational technologies that were applied to the economy at large and not to one vertical or one application only. Generative AI holds that same profile and, in my view, should be viewed as foundational too. The previous foundational technologies changed profoundly the way we work and organize ourselves and precipitated unemployment for large swathes of the working force. Inevitably, higher value-added employment was created over time. What we are seeing here is a potential acceleration of this dislocation. Take for example the average unit cost of compute power which dropped significantly over a +50 year period to date. Take the average unit cost of computer storage or bandwidth which dropped significantly over shorter periods of time – approximately +30 years. It is not inconceivable that the last input of the digital age, human powered software engineering, will drop significantly, cost wise over an ever-shorter period of time, say +15 years.
This means that we will not need as many software engineers as we do now to produce the same amount of code, or even more code for that matter. Apply this thinking to other mid-level or high-level skilled jobs in the legal profession, consulting, or corporate mid management (manufacturing, logistics, shipping, retail), and you start to understand the momentous changes to our economies.
As for startups, we now have a game changer on our hands. I venture that any and all startups that merely build technology or compete on technology will have a much tougher time being funded and/or breaking through. The reason is simple, the cost of engineering will trend to near zero over time (see cost of bandwidth, storage and compute power above) and the time it takes to iterate a product-market fit and beyond will shrink. The world of startups and venture investing is entering a new era, hence the title of this piece, “The Great Reset”.
For the financial services industry at large, and for its service providers, I suspect momentous changes too. Remember the point I made above on how technology change can upend business models. Most executives miss that at the onset of technology change (e.g., the obsolescence of the NIM-driven banking model was not predicted). Entire ecosystems of financial services and financial services providers models are built on scaling human power (compliance staff for financial institutions, junior lawyers for law firms, consultants for consultancy firms, mid-level bankers tending operational tasks for banks…) and either directly or indirectly bill additional staff to the client or end user. Generative AI invalidates this model. Financial services executives may think they will be able to harness these new technology tools to protect or increase their margins but will not think about business model changes. Let me illustrate this with one example: Generative AI will simplify the funds services industry from fund inception to fund services across transfer agency, custody, depositary, and fund accounting. This will not only reduce staffing needs but also remove the friction currently experienced by fund managers who find it rather difficult to change service providers – which occurs when arcane and bespoke rules are codified and interpreted in an automated fashion within the right contextual understanding. What will happen in this world? The model of the entire funds services industry will be upended. The threat of easier churn will kill the status quo. This is but one example and I suspect there will be many more across the financial services industry.
Now, we live in a world of assets and transactions (at least that is one way to view our economic world). We already believe, and the above analysis does not invalidate such view, in a world of digital assets (more “intelligent” assets than electronic assets). We should also believe in a world of digital transactions (more “intelligent” transactions than electronic transactions). One strong conclusion is that these two worlds (essentially two sides of the same coin so to speak) will be joined by a world of digital actors (more “intelligent” actors than electronic actors). These digital actors may be humans supplemented by whatever avatar generative AI tools may develop for them, or cyborg generative AI tools and platforms that will get to a signal or digital assets, or digital transactions or both in a much swifter and cost-effective way.
This is at least the promise I see for generative AI, a foundational tool that will help accelerate the future digitization of niche markets in financial services (capital markets and asset management to name but two) in a contextual way and with less input and less time to achieve success, product/market fit, product/market/sales fit, scale up….
What does this all mean for fintech startups going forward as the cost of developing will drop to near zero, the cost of trial and error will drop to near zero? Those that apply new technology paradigms while working on network effects will be the winners. These network effects may come from distribution at scale, two-sided marketplaces, data plays or a combination of both. Those network effects will be reinforced at scale by the application of generative AI. Linear businesses, or businesses solely competing on a technology solution or the automation of a process with a technology solution will be much less attractive.
I will end with two additional points.
First, generative AI has to be extremely deflationary for the economy. Demand will decrease and depending on the swiftness of the transition, demand for previously human skill sets may collapse for a certain period of time. This is a significant issue and can not be cast aside in cursory economic language, pointing to a mere transitional economic dislocation for key portions of the workforces. This secular dislocation could very well be severe and long lasting, with implications for the economy and individual workers, and attendant risks to social cohesion. Assuming higher value-added employment cannot be created as swiftly, could UBI (Universal Basic Income) be a solution for our governments? Robot (or generative AI taxes) would be the natural fiscal input to UBI.
Second, and especially as it applies to the financial services industry (and other sectors that bear similar profiles when it comes to the safeguarding of data), generative AI will develop along the lines of a more closed source as opposed to an open source and a more centralized as opposed to decentralized technology. This does not mean there will not be a market for open source and decentralized generative AI tools. Rather, those open-source tools will be used as complements while those closed tools will be foundational. Additionally, whether closed or open, centralized or decentralized, the nature of the algorithms will be scrutinized by both regulators and legislators to ensure as little bias as possible, or to minimize unintended negative consequences. To wit, Italy has moved with a ban against generative AI and the French government is thinking similar steps are warranted. Let’s hope cooler heads will prevail and that sensical compliance will arise instead of outright bans in the EU.
All the above points to new fintech giants in the making over the coming decade. Generative AI is the last positive step in AI coming after many failed AI experiments, and it may not be the last, what with further advances towards “artificial general intelligence”. Moore’s wisdom indeed.