I've had something of an eye for predicting change related to quant, fintech and computing trends, e.g. open source, data science, languages, risk/governance, perils and pitfalls of big data, stress testing, etc. Sometimes folk listened, other times not, and I was occasionally encouraged to make some sanitized corporate-friendly predictions public.

For 2025, I’ll attempt some predictions blending Tech & economics ("techonomic"). I won’t predict markets as I’d make a rubbish portfolio manager. However, I consider markets high, with optimistic sentiment hiding some challenging fundamentals underneath, particularly in certain equities categories and speculative crypto assets. For Faang and FinTech listed organizations in particular, opportunities to cut costs mitigate against increasingly competitive tech landscapes for some short term positivity, helping grease some M&A also. However I'll highlight some tricky macro- medium term structural headwinds faced by Fintech providers, before moving into tech trends that could help innovate opportunity and mitigate risk:

Lower Wage Costs Help Mask Revenue Challenges

I’m a fiftysomething. With some college peers at a mini reunion before Christmas, we all bonded over having been laid off. Our common experience was that our wage premium in Financial Services & FinTech for our experience was deemed expendable compared to cheaper, energetic, “all-in” (quoting my former CEO) less experienced staff. Elsewhere, a sixtysomething sales enablement guy I went to my local pantomime with got laid off and, his words, "utterly reasonably replaced" by 8 new Eastern European colleagues.

For youngsters, numbers of applications for jobs are through the roof, as global competition and drives to efficiency kick in. At every dinner table over Christmas, I heard similar stories. Jobs carry much less certainty into 2025. Employers of all types can cast wide geographical and AI nets (no pun intended). Global competition, the need to keep balance sheets tight amidst investor scrutiny, and a new wave of GenAI-enthused automation is dampening wages for many FinTech roles.

For society, that's good. It's quite right, for example, that healthcare professionals wages are accelerating faster than ours given the increased demand for their increasingly scarce services.

Trouble is, downward FinTech wages dampens confidence while also obfuscating difficulties on the revenue side of the balance sheet. Nowhere is this demonstrated so aptly than on the inflated, ram-packed AI marketmap. Not just for those organizations targetting AI and big data applications but also for those in other crowded spaces targeting payments, cybersecurity, process automation, etc, customer retention increasingly is a challenge, while new business sales cycles lengthen. Handfuls of buyers enjoy the luxury of choice when procuring from the many. And here's where "experience" trumps "all-in." Experience, sometimes, knows what channels to pursue, and pursestrings to pull to get the deal. But buying experience comes with risk, and balance sheet expense, which isn't always what investors want.

Beware castles built on sand rather than on stone.

Consumer Demand Dampens Mood

We all pay more. My cup of Costa coffee increased by about 32% over about 18 months. Baristas cost more to import or home-develop, while coffee beans, insurance and store costs are all on the up, and, in the UK, higher national insurance “contributions,” aka taxes, impact too.

We probably have less to spend. Mortgage rates are rocketing, in my case from 1.71% to 4.11%, as we move from a post-GFC low interest regime to a more “normal” post-Covid/Ukraine (and in UK, post-Brexit) regime. Yes, some countries face fewer house price pressures, e.g. seriously long-term deals in countries like Sweden, or simply smaller housing markets, yet disposable incomes are reducing for other reasons. Consumer lending of all types is getting costly, and we’re spending more carefully.

Of course, this isn't FinTech B2B demand we're talking about, but FinTech thrives best when money circulates, investment rockets, and momentum encourages. Mood is publicly optimistic, but behind the scenes, structural consumer demand changes could impact. Less wealth to invest, means less wealth circulating, and diminishing FinTech opportunities. I know that sounds a bit Keynesian, but watch carefully.

Politics: Deregulation v Protectionism

With spending power less, costs on the up, and jobs less secure, people are angry. I've discovered my relatives pretty much all support Reform UK, and immigration debates dominate their dinner tables. I am on the “woke” side of that debate, and my cousins et al blame post-colonial global citizens like me for preferring open markets, rational secularism, multiculturalism/religion, and freedom of movement. It doesn’t help, in my view, that they share memes on social media, possibly of Russian origin, claiming “the return of British values.” Family reunions are less comfortable than they were.

France and Germany both look set to follow Italy, the US and Argentina, in Germany’s case rapidly. The UK Starmer government is unsteady, Trudeau administration in Canada on its last legs, and right-leaning tensions exist across most developed nations. Should Trump/Musk be even partially successful in stabilizing domestic economies, law and order, the hegemony of “progressive” left-leaning internationalist technocrats that has dominated since the 1990s could disappear for generations, or, worse, until after a third world war. 

Financial Services tends to ride waves of political change with relative ease, but it also operates in global markets where time and space have compressed to such a degree that distance and time constraints no longer restrict our ability to do things. If protectionist and political barriers take effect and supercede freedoms from likely pending Trumpist and Musk-enthused deregulation, we'll manage and then some, particularly the high profile crypto and DeFi bros. However, with more barriers put in place to selling, trading and exchange, impacting liquidity and getting money where it needs to be, FinTech, even crypto, may find significant challenges to delivering on its raison d'etre.

A Significant Malware Event Beckons

Remember when cybersecurity was all the rage two or three years ago, in pre ChatGPT olden times, also when the Log4J/Log4Shell vulnerability made headlines? Last year, the Crowdstrike glitch, a bug not malware, had global impact. With the increasingly nationalist, aggressive and defragmented geopolitical backdrop being what it is, while on the software side ever-increasing code collaboration and, with LLMs, propensity for code leakage, I fear conditions are ripe for something bad.

This year, I see an event bringing together bad actors with software vulnerabilities. 2FA won’t solve it. Code maintainers, particularly those working on stretched open-source code bases, will have to work increasingly hard to ensure code safety. Only too late will we empathize with their stresses and challenges, and appreciate the importance of securing our closed and open code repos. FinTech firms are front row participants in this new battleground.

Trickier ESG, Amidst Reconciling Languages of Uncertainty

I follow many climate yay-sayers and some nay-sayers on LinkedIn. It seems that the latters’ time has come. I'm sensitive to arguments of both sides around reliable forecasts and levels of uncertainty. Having spent decades working alongside those in the capital markets, macro-economics and insurance, I see difficulties in predicting days/hours/minutes ahead events in the markets, let alone identifying economic cycles and shocks, and long-term pension/life insurance policy trajectories. Time-series forecasting is hard, with climate forecasting tremendously hard. Climate, environmental monitoring at large, and forecasting and measuring governance and social events are highly subjective and challenging disciplines, and this is increasingly recognized with constant challenge.

This has two implications, one bad and one good

With debate so polarized around climate change, we lose sight of vast environmental degradation elsewhere — the depopulations of species on land and sea, the destruction of vegetation to industrial agriculture and grey-belt, and challenges to our water systems. Some of the great activities I see around Green Finance, including in this Finextra forum are outstanding, but I fear the political climate has gotten so  micro-focussed on the “truth” of climate forecasts and "tangibility" of environmental change that we'll struggle to progress the greater cause this year.

However, we will begin to appreciate that forecasts come with uncertainty. GenAI happily has introduced us to risks of “hallucinations,” which extends far beyond misinformed content (pictures of feet with 4 toes for example) but actually anything model-based, predictive or AI-related relating to very open systems (as opposed to closed systems, like a TV remote controller for example), "trained" on incomplete data and subject to many data factors. We appreciate that AI doesn’t know what it doesn’t know, and increasingly compromise for its uncertainty. This extends to all model types. This year, I see increasing focus on uncertainty language embraced by the intelligent of all sides in a) climate and environmental science; and b) more generally in quantitative modelling.

RAG Goes Mainstream with Agentic AI

In tech, localized LLMs (Large Language Models) will be accompanied by internal data-sets as RAG Agents.

Let’s quickly demystify the tech. RAG (Retrieval Augmented Generation), a topic  I've talked often about on Finextra, essentially couples LLMs with their semantic power and global knowledge alongside internal data-sets. You can kind of think of ChatGPT as a sort of RAG thing, in that it provides an interface to an LLM, specifically GPT-4, and you, through your prompt, give it your context. However, the reality is that we don't give good context in our prompts, and thus may see responses as appropriate as the Old Testament judgement of Solomon.

But let's discuss agents because they tackle enterprise context needs head-on. They're like RPA “chat agents” of old, like Microsoft Clippy, or the never-satistfying Ryanair chat thing. Except they have a broader set of knowledge and understanding to lean on to drive smaller, localized use-case specific instances of managed and perhaps not-quite-so-large LLMs, deployed for more tasks in conjunction with local enterprise data stores.

Through this combination of LLMs and enterprise data-as-knowledge, Generative AI will begin to attain enterprise value and become more useful and increasingly mainstream. 

In this way, Agentic AI will bring Generative AI into similar pipelines as traditional AI, combining local data alongside a general-purpose model. Back to basics, but with greater power. Also, as noted earlier, we're increasingly sensitive to, and willing and able to manage, issues of accuracy, bias, GIGO (Garbage In, Garbage Out), the afore-mentioned uncertainty, etc. Agents are human-managed intelligent orchestrators of global and local knowledge.

Knowledge Gets Appreciated

LLMs brought semantics to the fore. That’s great, because semantic scholars appreciate, mostly, that words are laden with meaning — shades of grey — not simply black and white scrawls or typeset on a page. The maths behind this is computationally and storage-intense linear algebra of encoded stored and searchable vector structures, exciting to many mathematicians.

It’s not, though, just the maths that’s exciting and vibrant. GenAI also shines light on how “facts” need context, and are most useful when deployed with context. In short, facts are more fuzzy than you probably think. Quoting Voltaire, “the more I read, the more I acquire, the more certain I am that I know nothing.” 

I use the word fuzzy deliberately. Lotfi Zadeh came up with the concept of fuzzy logic back in the eighties, helping reinvent the Japanese washing machine where it caught on big time. His theory draws on many other theories - logic that provided rules for control, but implemented with "fuzzy" boundaries, basically allowing room for intelligent manoeuvre and a forerunner of modern AI. In the case of washing machines, the fuzzy washing machine controller intelligently assesses washing load weight, fabric type, water temperature, detergent level, and more to determine how best to run the washing cycle. Some parameters are discrete, others variable. There are many types of washing cycles therefore, managed by your washing machine's fuzzy logic-based controller.

On a different but similar note, French scientific and academic disciplines, meanwhile, pair philosophy with hard science. This means French Quants are, I find, particularly free-thinking about maths but also understanding the world at large, so well placed to understand how closed, say, option pricing strategies get reflected in open markets as volatility conditions change. I also tend to have good discussions when I ask my French quant friends how AI is reflected in philosophical thinking, contemporary literature or cinema, once I've asked ChatGPT or Gemini that is.

Such thinking, or reasoning, typified in both examples brings hard systemic models, stochastic and control in concept, with open universes. In FinTech, this marriage of model and universe allows us to buy, sell, acquire, trade, develop and grow. AI tends toward model automation, but knowledge determines how that AI model augments decisions, when it is applied, and how it gets implemented. Even in the extreme case of the very highest frequency trading, which, like risk management and fraud detection, gets informed by the intelligent selection and tuning of key parameters, tends to draw on human judgement, data availablity and thorough automated back-testing. Such augmented decision-making drives competitive advantage and risk mitigation. 

While knowledge graphs have existed for a long time, underpinning Google search for example, they are fast becoming leading technology vehicles for delivery of knowledge-at-scale in the age of GenAI. They are fast challenging and overhauling brute force vector encoding approaches, or at least working alongside them in GraphRAG pipelines. Graphs manage data about things, entities, and, importantly, relationships between the entities. That's not native to the tabular formats which has dominated the data disciplines for so long, including in Finance and in so-called "relational" databases, which aren't truly relational, in knowledge terms at least. Knowledge graphs bring value therefore to both semantics and general scientific discovery, not just as a one-off static repository of given knowledge but an ever-changing mesh of changing relationships.

For this reason, knowledge graphs will drive FinTech & AI innovation into 2025, though other, simpler means will also come to bear in computing knowledge, reasoning and context in conjunction with new and old AI. Knowledge graphs, like vectors, are still somewhat prescriptive in their application. Data is expressed as "nodes" and "edges", and underneath, like vectors, linear algebra-driven, albeit sparser matrices. However, they bring flexibility, reasoning, and context in a transparent, understandable way. They give us a lots to work with when deploying context with all types of AI.

In this way, maths and STEM will begin to merge with philosophy and other domains of knowledge, as it has done in recent linguistics, NLP (natural language processsing) and computational semantics disciplines. I am excited by this. As the world changes rapidly with AI and grapples with a new political regime, it will help new entrants to our job market and global citizenry create the next generation of positivity and stability with situational awareness amidst the current turbulence. FinTech will be a prime beneficiary.

Thanks for reading. Note that all views reflected in this blog are mine, not my employer's.