*Forward the Original Title ‘#81 - Beyond the Hype: Why Gen AI Is Banking’s Next True Platform Shift (Free to Read)’

Sponsored by Triage

Built by Executives

In Africa’s financial services sector, niche expertise in areas like risk, credit, compliance, and technology can make or break a company’s performance. Navigating complex regulations and integrating new technologies demand leaders who understand the nuances of financial products, stakeholder expectations, and market realities. Without this depth of knowledge, even well-funded institutions risk costly missteps that unsettle investors, stall growth, and undermine customer trust.

Triage brings a mix of hands-on experience as operators in banking and financial services, and global experience working with some of the largest financial services businesses in the world. Our team has worked with senior leaders in over 35 countries across Africa supporting a range of growth and change strategies across a spectrum of clients, from early-stage ventures to scale ups, to digital transformations and turnarounds. This broad experience enables us to spot the difference between passing expertise and genuine capability, ensuring you engage leaders who truly understand what it takes to succeed in the quickly evolving world of financial services.

Introduction

Skepticism is what it takes to look behind a balance sheet, the latest miracle of financial engineering or the can’t-miss story. . . . Only a skeptic can separate the things that sound good and are from the things that sound good and aren’t. The best investors I know exemplify this trait. It’s an absolute necessity. - Howard Marks

As a young person in finance, it’s important to develop a healthy dose of Skepticism. In banking specifically, skepticism pays off because the most successful banks are those that avoid loss rather than those that chase wins. It’s a negative art. Nonetheless, skepticism is not the same as pessimism. It simply means having discernment about what is hype and what isn’t. The challenge with many people in finance is that they fall into the trap of being skeptical for the value of socially signalling that you’re smart.

As John Collison or was it Naval Ravikant said, “Pessimists sound smart, Optimists make money”. A skeptical mind is valuable. However for it to be valuable, it must be matched with analytical rigour and importantly, the ability to change your mind when facts change.

This is a useful context for the current generative AI discussion particularly in banking and finance. It’s important to specifically refer to generative AI as opposed to the machine learning that’s been around. for some time particularly in the banking sector. Generative AI is the type of intelligence that can create new things such as text, images, audio or video from learning on vast troves of data. Lazy skepticism is leading many people to prematurely call AI hype, whilst unbridled optimism could lead to premature investments. For smart decisions around AI to be made, it’s important to put AI in context, particularly its economic context. That means analysing AI as a platform shift and putting it in its historical place compared to other platform shifts. AI in its historical context should make bankers and the finance industry in general make the right decisions.

In today’s article, we’ll understand what a platform shift is, review past platform shifts and their impact on the financial services industry, place AI in its context as a platform shift, look at global initiatives by banks and Fintechs around AI and evaluate the key lessons for leaders in the financial services industry.

Platform Shifts and Financial Services

What’s a Platform Shift

Finance like any other industry is subject to the vagaries that technology brings about. Whether it’s the telegram and its impact on branch based banking or the mini-computer and its impact on ATMs. Finance has always adapted to platform shifts. In technology, a platform shift refers to a fundamental shift in the underlying technological architecture that enables new capabilities due to a step change in the underlying cost structure. Often, this enables new business models and ways of creating value. The key thing is that there must be a fundamental change in the cost structure of something i.e. the cost of doing X reduces by a factor of 10x+ for something to be truly considered a platform change. The key characteristics can be described as;

• Foundational Architecture Change: Platform shifts involve radical changes in how technology is structured and accessed, not just improvements to existing systems.
• Exponential Cost-Performance Improvements: They typically deliver order-of-magnitude (10x or greater) improvements in cost, performance, or capabilities, not just incremental gains.
• New Value Creation Models: Platform shifts enable entirely new business and value creation models that weren’t previously viable.
• Ecosystem Formation: They spawn rich ecosystems of complementary products, services, and businesses. This is often an effect rather than a defining trait.
• Market Disruption: Platform shifts often disrupt existing industries and create entirely new markets.
• Democratization: They typically make technology accessible to broader audiences, enabling new participants.

We’ll look at some historical platform shifts and importantly analyse;

1. What the platform shift was all about;
2. Its impact on cost structure;
3. Winners from this shift and how they leveraged the technology;

1. Mainframes: Centralising Computation (1950s–early 1970s)

Historical context and key characteristics
Before the 1950s, banks kept ledgers by hand or with electro‑mechanical tabulators. Processing a cheque meant a clerk typing a line item, filing paper, and reconciling totals at day‑end. Mainframes such as IBM’s System / 360 introduced stored‑program computing, magnetic‑ink character recognition, and batch processing. For the first time a single machine could read tens of thousands of cheques an hour, apply account rules automatically, and post results overnight.

Cost curve
The capital bill was steep, several million dollars, but the marginal cost of posting a transaction fell by about one hundred‑to‑one versus manual entry. Error rates collapsed, cut‑off windows tightened, and scale became a software problem rather than a staffing problem.

Winner’s story
In post World War 2 America, the US middle class was booming and demand for banking services and cheques in particular was growing. In Bank of America, the number of checking accounts was growing at a rate of 23,000 accounts per month and the bank had to close by 2pm just to process cheques. Bank of America implemented the Electronic Recording Machine Accounting (ERMA) system in 1959. It processed about 36 000 cheques per hour, (about 10 per second) versus ~245 checks/hour by a human bookkeeper. It handled three quarters of a billion postings a year, and freed the bank to expand beyond California without hiring thousands of clerks. For Bank of America, by dramatically improving throughput (over 100× faster), it drastically reduced the cost per check processed and scaled to serve more customers. Automating back-office tasks gave early adopters like BofA a cost advantage, fueling their growth into national leaders.

2. Minicomputers: Departmental Automation (1970s–mid 1980s)

Historical context and key characteristics
The advent of minicomputers – smaller and much cheaper than mainframes – democratized computing beyond the Fortune 500. Banks, brokerages, and service providers could deploy mini and mid-range systems (from vendors like DEC, Data General, IBM’s AS/400 line, etc.) at the department or branch level. This era saw the birth of electronic networks and fintech services that could run on less expensive infrastructure, enabling new specialized players

Cost curve
A branch could now have its own compute power for a fraction of the cost of a mainframe computer. Interactive sessions replaced batch reports, and new channels such as ATMs became economical. Minicomputers slashed the price of computing. A mid-1970s mini could cost in the tens of thousands, bringing per-unit compute costs down by an order of magnitude compared to 1960s mainframes. This affordability broadened IT adoption in finance. As a result, by the 1980s, even mid-tier financial firms were computerizing operations, leading to faster service and lower unit costs

Winner’s story

A DEC Mini Computer - Source DEC

Citibank bought hundreds of Tandem NonStop and DEC mini computers, networked them to ATMs, and launched its “Citi Never Sleeps” marketing in 1977. When a blizzard shut New York in 1978, Citi’s ATMs kept serving customers, transaction volumes jumped twenty percent, and deposit share in the city doubled within three years. Teller costs, roughly a dollar per visit, dropped to about thirty cents on an ATM.

3. Client‑Server and Relational Databases: Distributed Processing (late 1980s–1990s)

Historical context and key characteristics
Prior to the client-server era, the database sat within the computer combining both the data and the interface. The client-server era brought about a separation between the data layer and the interface layer. There was a client (PC) and a server. A Windows or Mac PC handled the presentation, a mid‑range server stored data, and SQL spoke between them over a local network. Off‑the‑shelf relational databases meant new insight: millions of rows could be queried in seconds, enabling statistical marketing and risk models.

Cost curve
Sub‑$2 000 PCs plus sub‑$100 k Unix boxes let banks query millions of rows in seconds.

Winner’s story
Capital One, spun out of Signet Bank in 1994, used a client‑server grid running Oracle to test thousands of credit‑card offers in parallel. It priced risk at the individual level and grew customers forty per cent in 1997 while incumbents relied on broad FICO tiers. Return on equity consistently exceeded twenty per cent because analytics replaced blanket pricing. Other winners included Charles Schwab who figured out that the client server era could democratise stock brokerage.

In Africa although there was a slight delay, the winners included;

1. Equity Bank who harnessed client-server architecture and mini-computers to scale their transactional capabilities by upgrading to Finacle (A Client-Server based Core-Banking System) whilst also scaling their ATM network using the same systems. This eventually turned into Agency Banking. They moved from being a non-entity in the 90s to East Africa’s largest bank by market cap;
2. GT Bank - Used client-server architecture to improve transaction processing at the branch enabling a better customer service. Prior to this, clients had to wait hours to process a simple deposit or withdrawal.

4. Cloud 1.0: Infrastructure as a Service (2006–early 2010s)

Source: Business and Finance Magazine - The Collison Brothers

Historical context and key characteristics

The web still required companies to own servers. Amazon Web Services turned compute, storage, and databases into metered utilities. An application could scale from ten users to ten million without a hardware purchase order.

Cost curve
Instead of millions in cap‑ex, a developer needed a credit card and could pay pennies per hour for compute. Elastic capacity meant cost scales roughly in proportion to usage, eliminating big step‑ups.. This was a far cry from the relational database era where you needed to size up your growth in advance leading to significant upfront capex.

Winner’s story
Stripe launched in 2010, four years after AWS launched in 2006, with a payment API that went live in minutes. Its seven‑line code sample abstracted away merchant underwriting, settlement, and compliance. By 2024 Stripe handled about US $1.4 trillion in payments, volumes that previously sat with bank acquirers and legacy processors, and its onboarding cost remained a rounding error thanks to usage‑based cloud billing. APIs become a new form of value creation validating the cloud as a true platform shift.

5. Mobile & Cloud‑Native Era (2010s – 2020s)

Source: [itweb.co.za]

Historical context & key characteristics
Smartphones put an internet computer, biometric sensor, and secure element in every pocket, turning “distribution” into an app‑store listing. On top of that, Public‑cloud platforms (AWS, GCP, Azure) delivered bank‑grade infrastructure as a utility; micro‑services and CI/CD pipelines enabled weekly—even daily—feature releases. Mobile networks doubled as payment rails; QR codes and virtual accounts displaced dedicated POS hardware and branch networks.

Cost curve
In this new framework, customers supplied the terminal, bandwidth, and authentication; incremental onboarding cost collapsed to a fraction of the costs of onboarding either a branch based client or merchant. Transaction fees on app‑based rails drop below 1 %, opening profitable access to low‑ticket payments and fee‑free accounts.

Winners & their playbooks

• Nubank (Brazil, founded 2013)
  • Mobile‑only acquisition via viral wait‑list; 2 000+ AWS micro‑services for real‑time credit analysis.
  • Deployed code dozens of times per day, rolling out features faster than regulators could approve legacy banks’ price increases.
  • By 2023: 100 million customers across Latin America; US $11 billion in legacy fees saved for users; double‑digit share of Brazil’s card market while maintaining one of the lowest cost‑income ratios in global banking.
  • Their revenue is growing at twice the rate of their costs and are on track to be the most profitable bank in Brazil by 2028.
• TymeBank (South Africa, founded 2019)
  • Cloud core on AWS; instant, paperless KYC via biometric kiosks inside grocery chains (Pick n Pay, Boxer).
  • No owned branches, lean tech team; cross‑sell savings, credit, and insurance in‑app.
  • By 2024: 8 million customers; reached profitability in under five years, demonstrating that physical distribution can be outsourced while the core remains purely digital.
• Other examples include companies like Flutterwave, Stitch, MNT Halan, Paystack and Paymob.

Why they won

• Zero‑infrastructure distribution ;
• Elastic economics 
• Developer‑first surfaces;
• Rapid iteration cycles;

Taken together, these players illustrate how customer‑owned devices plus cloud‑native architecture create a structural cost advantage—and made speed, not legacy scale, the decisive weapon in African and global banking alike.

Key Lessons from Past Platform Shifts

Some key lessons from past platform shifts

1. All the platform shifts enabled financial services to be done in a different way. The fundamental idea was the cost structure of a specific issue either the cost to transact or the cost to distribute.
2. Beneficiaries of platform shifts were either banks that quickly adopted technology (BoFA, Citi) or new entrants that understood what the shift enabled (Stripe, Nubank);

The Context for Generative AI

For me, past platform shifts have focused on cost and distribution given that these domains were really software specific i.e. deterministic. Gen AI may not necessarily be a cost and distribution issue. My view is that Gen AI will be a 10,000x reduction in the cost of delivering a bespoke relationship. Currently, banks and Fintechs have distributed transactions via tech and this is a trend that will continue. Almost everyone transacts on their phone with very little transactions happening at the branch. This applies for individuals as well as corporate clients. Nonetheless, the remaining bottleneck for further distributing financial services is enabling relationship banking at scale. This is because this remains the work of human beings given relationship management is high-context and requires judgement.

Gen AI can deliver premium “relationship banking” for pennies per customer. Today a top African RM costs roughly $6 000 a month to serve ~30 clients, about $300 each after overhead. Shift that work to AI and the cost could drop to mere cents, unlocking high‑touch advice for the mass market and transforming financial access across the continent. This in my view is the next frontier given that transactional Fintech has already been solved.

Relationships will still matter in banking—but they’ll shift from human‑to‑human to human‑to‑AI. Money conversations often carry shame; many customers hide basic questions from a banker’s gaze. An inanimate, tireless AI lowers that social barrier, inviting candour and endless “dumb” questions. Greater honesty plus 24/7 guidance makes AI a powerful, scalable relationship manager.

Global AI Initiatives

How the World’s Banks Are Actually Using Generative AI

If you strip away the headlines and hype, the question remains: what are the world’s biggest banks really doing with generative AI? Not the future potential. Not what the vendors are pitching. What’s actually been deployed, and where?

Over the past two years, the global financial sector has quietly entered the era of generative AI. But the picture that emerges is not uniform. It’s a mix of quiet internal tooling, cautious client-facing experiments, and a few genuinely bold moves that hint at how banking may be restructured from the inside out. I give an overview below;

Internal First, Client Later

If there’s one consistent theme, it’s this: AI starts inside.

The bulk of generative AI adoption has focused on internal productivity—tools that help staff do more with less. From JPMorgan’s analyst assistant that parses equity research, to Morgan Stanley’s GPT-powered tool for wealth managers, the early bet is on empowering bankers, not replacing them.

Goldman Sachs is building copilots for developers. Citi has AI summarizers helping staff handle memos and draft emails. Standard Chartered’s “SC GPT” is live across 70,000 employees, helping with everything from proposal writing to HR queries.

Given that we live in a high-touch regulatory environment, internal tooling makes sense because banks can experiment and sharpen their AI chops without coming across any regulatory infractions. If the recent CBN action against Zap is anything to go by, then better safe than sorry.

Segment by Segment: Where the Value Is

Different divisions are moving at different speeds. Retail banking leads in terms of volume. In this regard, Wells Fargo’s Fargo or Bank of America’s Erica, chatbots powered by generative AI are now handling hundreds of millions of interactions annually. In Europe, Commerzbank recently launched Ava, its own chatbot.

The issue nonetheless is that some of these are not actually using generative AI and are in fact relying on machine learning. This article gives a good breakdown of how Erica by Bank of America works, it’s in fact a mechanical turk. Nonetheless, it’s the experimentation that matters.

In corporate and investment banking, the shift is subtler. JPMorgan’s internal tools support research and sales teams, not clients. Deutsche Bank is using AI to analyze client communication logs. This isn’t customer service—it’s data leverage, helping bankers understand and serve clients better, faster.

Wealth management sits somewhere in between. Morgan Stanley’s AI doesn’t talk to clients directly, but it ensures advisors never walk into a meeting unprepared. Deutsche Bank and First Abu Dhabi Bank are piloting client-facing assistants for their top-tier clients, designed to answer nuanced investment questions in real time.

Regional Differences: Who’s Moving Fast?

Source: Evident AI Index

North America is leading, as expected. The US banks; JPMorgan, Capital One, Wells Fargo, Citi and RBC have turned AI into a productivity engine. And thanks to partnerships with OpenAI and Microsoft, they’ve had early access to cutting-edge models.

Europe’s more cautious. BBVA, Deutsche, and HSBC are testing tools internally, often with more guardrails. GDPR casts a long shadow. As always, Europe is focusing on regulation rather than progress and this may cost them.

Africa and Latin America are at an earlier stage, but moving fast. Nubank in Brazil is a standout, partnering with OpenAI to deploy tools internally and eventually to clients. In South Africa, banks like Standard Bank and Nedbank are running internal AI pilots across risk, support, and development.

China: Building Its Own AI Stack

China’s banks aren’t just using AI—they’re building the stack.

• ICBC launched Zhiyong, a 100-billion parameter large language model built in-house. It’s been called over a billion times, powering use cases from document analysis to marketing automation across 200 business domains. It’s not just internal tooling, it’s a foundational shift in how the bank runs.
• Ant Group launched two financial LLMs - Zhixiaobao 2.0 (for retail customers) and Zhixiaozhu 1.0 (for financial professionals). The former is designed to explain financial products to everyday users inside Alipay. The latter helps wealth managers summarize market reports and generate portfolio insights.
• Ping An Group, one of my favourite Fintechs, a hybrid of insurance, banking, and tech, is pushing even further. It built AskBob, a generative AI assistant for both clients and relationship managers. For customers, AskBob can answer investment and insurance questions in natural Chinese. For advisors, it pulls and summarizes customer history, product data, and marketing materials, turning every agent into a digitally-augmented financial expert. Ping An’s ambition is to redefine financial advisory through AI, not just answer questions, but preempt them.

In China, where regulatory frameworks strongly encourage data localization and model transparency, these institutions are taking the long road: building custom-trained AI that can thrive in domestic regulatory, linguistic, and market environments. Moreover, China has sufficient talent density to enable banks to build their own foundational models, a feat that may not be repeated anywhere else in the world.

Who’s Powering This?

A few big names show up everywhere: Microsoft (via Azure OpenAI) is by far the most common platform. Everyone from Morgan Stanley to Standard Chartered is running their models in Microsoft’s secure sandbox.

Google’s LLMs are in play too, Wells Fargo uses Flan to power Fargo. And in China, it’s mostly homegrown: DeepSeek, Hunyuan, and others.

Some banks such as; JPMorgan, ICBC and PingAn are training their own models. But most are fine-tuning existing ones. It’s not about owning the model. It’s about owning the data layer and the orchestration.

Overview of Different AI Initiatives Globally

So what?

In a highly regulated industry, it’s important to be cautious and that’s why banks are keeping AI in the loop, not on the frontlines. Nonetheless, like we’ve observed in other platform shifts, it’s critical to be decisive and experiment rapidly. Regulation will never be ahead of execution and it’s not smart to stall AI experimentation with the idea that you must wait for regulations. I remember building agency banking over a decade ago in a country that didn’t have such regulations. Once we built it, we were the ones who explained it to the Central Bank. If I were on the board of a bank, my question would be “how many experiments are we running and how many insights are we generating?”

To really measure progress, you must go back to the fundamentals of a platform shift. Your AI strategy must answer:

“Does our AI strategy rebuild the core architecture, slash costs by 100×, unlock new value models, spark ecosystem link‑ups, disrupt markets, and democratise access?”

The logic is clear, it’s important to be skeptical but the logic and facts point towards AI being a new platform shift. Moreover, the logic and facts show that past platform shifts have proverbially moved the cheese in financial markets. Citi’s work with technology in the 70s and 80s significantly expanded its retail business. Capital One came out of nowhere to be a top 10 bank in the market and a significant player in adjacent industries such as auto loans and mortgages. In Africa, Equity Bank rode the client-server wave to be East Africa’s largest bank by market cap. The same wave was ridden by Access Bank, GT Bank and Capitec in their respective markets.

The AI platform era is here and it will create winners. The idea is not to focus on losers because what happens is that the winners take significant market share in a specific vector e.g. Stripe in Payments. These initial wedges lead to market share gains in adjacent areas like how Nubank used credit cards to become a serious player in SME and retail banking.

My view is that the winners in the AI era will focus on the cost of the relationship. It’s no longer a transactional game. That’s already happened. It’s a customer experience and relationship game. This is the core insight that financial service leaders should take. How can you create a 100x improvement in customer experience and relationship banking at a fraction of the cost? How can we harness intelligence as a bank to help you better manage your finances, your business and your life? The players that answer these questions and execute will be the winners.

Disclaimer：

1. This article is reprinted from [Frontier Fintech Newsletter]. Forward the Original Title‘#81 - Beyond the Hype: Why Gen AI Is Banking’s Next True Platform Shift (Free to Read)’. All copyrights belong to the original author [Mary Mogoi]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Unless mentioned, copying, distributing, or plagiarizing the translated articles is prohibited.