The Decades Long Club
It takes decades to save for retirement.? People eke out their savings for decades thereafter.?Likewise, it takes decades to buy a home.?Consumer facing businesses like life insurers, pension providers, and mortgage banks have customer communication opportunities that?pet insurers and credit card providers do not share.?
Cross border trade: Banks offer credit cards and mortgages (monthly payments going into a future asset). Insurers offer pet coverage and pension savings (monthly payments going into a future asset). Two of those four activities are basically the same, but for trivial differences in a thicket of regulations.
Some think artificial intelligence will kill us all. Tosh. "Abundant Intelligence" is the opportunity dangling in front of those with the courage to grasp it. Hopping across borders from pension savings to mortgage lending requires institutions to do two things: grasp the challenge and deal with complexity. The cost of the latter is falling fast.
Members of the Decades Long Club have always had data collected over years of interactions with customers where material amounts of money change hands monthly. But deploying that data is challenging when the underlying contract is complex and the customer's circumstances are bespoke.? It's the edge cases that kill you. Running issues up the chain when the line staff hits a speed bump is expensive and ad hoc. How do you recruit, remunerate, and retain scores of intelligent, empathetic people to work the night shift and respond to customer inquiries in a highly regulated environment?
Quantum of Solace:? Everything needs to be rebuilt for the age of Abundant Intelligence. The quantum of?data that can be thought through and taken into account in a fluid conversation between customer and service provider has changed and will continue to change.? The cost of making sense of that data is heading to zero. Hiding customer care phone numbers, keeping people on hold, and keeping calls short no longer makes sense.? If your customer wants to shoot the breeze, then shoot.
To appreciate the possibilities, it’s worth exploring how large language models have overcome the technical limits of their early iterations and how these breakthroughs can benefit members of the Decades Long Club.
Token effort: A token is the basic unit of language processing for a large language model. It might be as short as a single letter or as long as a word, depending on the complexity of the text. A context window, meanwhile, is the span of tokens that the model can "see" at one time. Early models struggled with tiny context windows, typically processing only a few hundred tokens at once – the equivalent of reading a couple of paragraphs before forgetting everything.
Break on Through to the Other Side:?Three years of expanding context windows by several orders of magnitude required innovation on multiple fronts.
Optimized attention mechanisms allow models to weigh the importance of different tokens across a sequence. This shift, coupled with innovations in sparse attention and memory-efficient architectures, has reduced the computational cost of processing large amounts of tokens.
Hardware improvements, such as the development of Nvidia's graphics processing units, tailored to Abundant Intelligence workloads, have enabled faster computation on larger datasets.
New training paradigms leverage incremental learning and help models process extended histories without bloating the underlying parameters.
Of Needles and Haystacks:??Together, these breakthroughs make it feasible to store and retrieve customer data across long periods. However, one challenge facing large-scale data retrieval is finding the “interesting” or relevant nugget in a sea of noise – the proverbial needle in a haystack. Models are increasingly adept at mapping out information into multidimensional vectors where similar items cluster. These techniques allow systems to locate the most contextually significant details without sifting through irrelevant records manually.
Microsoft AI CEO, Mustafa Suleyman, predicts?that 2025 will be the year of near-infinite memory.? Here's why this matters. You are a member of the Decades Long Club with a million customers' unique needs to track for the next 15,000 days.? That's four decades in the old money. Armed with one key piece of data, the customer's date of birth, you know how long they are likely to live (life insurance, pension savings, retirement income) and how long it takes to buy a home, decades in advance.?OK.?Now what?
Comprehensive Knowledge Retention: Infinite memory might allow you to remember every interaction with your customer, enabling you to track an individual's progress, interests, and challenges. The core life events that are common to the Decades Long Club allow you to take these learnings (looking backward) and use them to offer services with a decades long predictive basis.
Evolving Understanding: As your platform learns and adapts continuously based on past interactions, it can offer increasingly personalised advice and support tailored to the individual's future needs and (as discussed below) inform the pricing for similar services for the next cohort of customers. This cohort effect dwarfs cross-selling for balance sheet businesses.
The Killer App:? You could offer what Suleyman calls a personalised curriculum that would enable customers to receive tailored educational support throughout their lives.? Let's call this curriculum financial advice.
领英推荐
Permanent Persistent Memory:?The prospect of permanent memory could revolutionize customer service for industries rooted in decades-long relationships. But the real revolution is not cost saving (asymptotic to zero).? It's revenue potential from winning market share in a mature economy.? The top eight mortgage banks in the UK control 80% of the market. In Australia, the top four control more than 80%.?The winners will be those who don't hide their customer care phone numbers, don't keep people on hold, and are delighted when customers want to shoot the breeze.
We are not talking about cross-selling (so 1990s). The Decades Long Club are cohort businesses with big balance sheets. You deploy a billion dollars in 2025 and lock in a mortgage yield for decades – unless, for some bizarre reason you think it make sense to offer a two-year fixed rate. ??
During 2025, you test concepts with consumers and monitor price elasticity that informs product pricing for the cohort of 2026. Per billion invested, one basis point of incremental yield is $100,000 per year locked in for decades. It takes data, and product design, courage, and intelligence to figure out what next year's cohort is willing to pay for your core product. Tacking on somebody else's product is a bonus.
Here's a heretical thought. If your pension savings provider knew everything about you and "paid" you (by offering lavish customer service) for your mortgage data, they would be well-placed to advise on your strategy for paying down that mortgage early to save money. The pension provider could translate those savings into tangible forward looking strategies and fold meaningful numbers into their advice:
Playing for all the marbles: Let's break this down. Company A in industry A, says to customer X: [fill in the blank] once a day with Company B's decades long product and I will sell you more of my product 9,250 days from now. If this sounds fanciful, look at Sprive App. A plucky FinTech start-up advises mortgage borrowers about how to pay down loans with multiple giant mortgage banks. In other words, customers of deep-pocketed Bank X, pay Sprive to read their impenetrable loan documents.
Why would anyone want to trust their data to, or receive this sort of advice from, multiple vendors? The winner will be the innovator in either industry A or industry B who obsesses about delivering what today is seen as hideously complex, expensive, and highly regulated advice in the other institution's industry. In the age of Abundant Intelligence, the cost and complexity of such "cross border" innovation falls away and the power of brand, loyalty, trust, and the strategic use of data in context rises.
Pretty soon, one member of the Decades Long Club will be playing for all the marbles.
People affected by these issues should feel free to reach out.
Ike Udechuku | Cofounder | Pathway
#FinTech #FinTechinnovation #AIInnovation #CustomerExperience #LargeLanguageModels #PersistentMemory #FinancialServices #AIinBanking #CustomerRetention #LongTermRelationships #DigitalTransformation #ContextWindows #AIandFinance #TechInnovation #FinancialAdvice #FutureofFinance #AIinCustomerService #DataDrivenDecisions #SmartTechnology #PersonalizedExperience #DecadesLongClub