Dateline: Woking, 8th August 2026.
Rory Sutherland is a great guy. A former president of the Institute of Practitioners in Advertising (IPA), he founded the behavioural science practice within the Ogilvy Group. The group’s goal is to develop marketing techniques inspired by the fields of psychology and economics, rather than through conventional advertising. His excellent book “Alchemy: The Surprising Power of Ideas That Don’t Make Sense” is a classic in the field and a must-read even you don’t much care for the Mad Men.
Rory’s advice on looking for new products and services is straightforward. He says that marketers should not only test the things that make sense but the things that do not make sense because if you find something that works that way then you have learned something mega valuable. That is, you know something that nobody else knows, because the odds are nobody else “has been wacko enough to test it”. To what extent this applies to digital financial services, I couldn’t say. Presumably allowing a multi-currency account was dismissed as wacko at Barclays twenty years ago, which is why I use Wise every day, but on the other hand I can imagine that people are rather conservative about things involving money so that must constrain possibilities.
Anyway, people being conservative probably doesn’t matter very much because neither Ogilvy nor anyone else will be working hard to persuade people about anything in financial services for much longer. It will be robots, not people, making the buying decisions. McKinsey’s April 2026 report on “How gen AI agents threaten retail banks’ customer relationships” shows around a third of consumers already using AI for financial advice at least monthly, primarily for understanding products, getting investment advice and comparing providers.
Now, it is fair to observe that this use of AI is for investigating and not yet for buying. The eMarketer analysis of Stripe’s annual letter notes that in 2026, 95% of e-commerce sales driven by AI platforms will still be completed off the AI platform and on the merchant site. Agents manage context and drive discovery., but when it comes down to it, humans still click buy and type in their card numbers. At the time of writing, three-quarters of agentic activity (well, 70%) touched product and search routes: browsing product listings, reading articles, and running searches. The remaining activity splits across authentication (9%), user account routes (9%), miscellaneous (5%), content engagement (5%), and checkout and payment (3.%).
This balance, between discovery, enabling and transacting is changing though, and it might change quite quickly as standards coalesce and customers become more comfortable handing over power. It’s a small step from AIs comparing offers to pressing the button.
This is going to mean significant disruption in financial services because customers will have access to AI as powerful as the banks themselves because Google, Facebook, Apple and Amazon (and companies like them) will be giving it to them and few organisations have a strategy for dealing with customers who become literally one thousand times smarter overnight. And this projection is hardly science fiction. It’s already happening. Look at the recent announcement from Perplexity and Plaid about allowing AIs access to customer accounts. Right now the integration is read-only: Plaid shares data you’ve explicitly permissioned, but Perplexity does not move money or initiate transactions (yet). This is in line with Plaid’s model, whereby consumers can see which apps they have connected to which accounts and can revoke access at any time via Plaid’s own Permissions Manager and consumer controls.
(This is not all about financial services, of course. The advent of agentic AI also threatens to start “unpicking an online retail ecosystem“ that cost billions of dollars of investment to build over the years. Control over customer relationships and decades-wroth of data might not be the moat against competition that they are now.)
These are not new perspectives, by the way. Banks and their advertising agencies have had plenty of time to prepare for this tectonic shift.
Back in 2019, I wrote in Wired magazine that the real revolution in banking would be when customers got AI, not when banks got AI. I stand by that prediction and when I see the energy going into the agentic commerce, and therefore agentic payments, space I begin to think that, if anything, I was underhyping the impact.
In 2020, Cathy Hackl wrote that “marketing is about to change, and marketeers will needed to add business-to-robot-to-consumer (B2R2C) to their duties”. Indeed they will and probably should have already done so.
By 2021 researchers were calling for more work to be done on “AI-driven customer journeys” and on rethinking relationship management when human and machine agents jointly shape financial decisions.
In 2022, Gartner surveyed executives and found that they believed that by 2030, a quarter of consumer purchases would be “substantially delegated” to bots, suggesting a market shift roughly twice as large and twice as fast as the historical arrival of e-commerce. That too may have seemed a trifle hyperbolic at the time but now seems pretty reasonable.
The Alan Turing Institute in the UK published a paper in 2023 on “The AI Revolution: Opportunities and Challenges for the Finance Sector”, which highlighted the information asymmetry between financial institutions and individuals and noted that it can be attacked using “AI to enable individuals to access credit, save money, make deposits, withdraw, transfer, and pay for goods and services using a mobile device with AI intelligence”. This would be a much more cost-effective route to financial health for the population than trying to teach them about compound interest.
In 2024, I wrote on Forbes that the interplay between custbots, roboregulators and banks will create new dynamics that we cannot predict using traditional strategic planning primarily because as these bots interact, they will reshape our concepts of value, trust and identity. We have already seen experiments in which bots left to communicate with one another have developed their own language, their own groups and indeed even their own religions.
Last year Kirsty Rutter of Lloyds Banking Group and I published a paper on “Digital banking in the artificial intelligence era: Strategies for serving nonhuman customers” in the Journal of Digital Banking vol. 10, no.1 (2025). In that paper we wrote that while AIs were unlikely to care about which soccer team a bank sponsors or how cool the TV ads of an insurance company are or what the Super Bowl show featuring plugs for a crypto exchange looks look, there are presumably some things that bots would care about. But what?
Who knows what bots will be attracted by. If it’s only price, then we’re in a race to the bottom.
The FCAs recent “Mills Review” found that many UK financial journeys may soon begin with agents, so the marketing boffins need to get into the lab to begin their experiments in an environment where the buyer cannot be persuaded by a television advertisement or a clever slogan in the conventional sense. It will evaluate rates, fees, eligibility, exclusions, service quality, complaint outcomes, security controls, portability and the reliability of the underlying API.
Hence one plausible role for future marketing will be to ensure that the institution’s proposition is legible, comparable and (crucially) provable. This means building, for want of a better phrase, “machine-readable product truthiness”. Financial services providers will need to provide agent-friendly descriptions of pricing, terms and condition, eligibility, service level, redress processes a countless other things that I don’t know about.
There will of course also need to be a solid trust infrastructure so that marketplace participants know who and what they are dealing with. Agents will need to prove who they represent, what they are authorised to do and such like. For marketing, this makes control and explainability (and perhaps privacy) up-front product attributes, not back-office compliance matters.
(Safety and security will be table stakes. In their excellent IMF paper “How Agentic AI Will Reshape Payments”, Sonja Davidovic and Hervé Tourpe rightly warn that agents may optimise commercial incentives rather than consumer welfare, which suggests to me at least that agents will need to have some kind of “duty of care” certification before being allowed access to consumer financial services.)
Marketing departments need to learn who to measure whether agents can find, interpret and recommend their products. They need to learn how to assess the success of these measures. And, if they are to gain public trust, they must grow their markets without trying to manipulate these measurement through backdoor commercial deals.
They must, in short, find out how to influence in a world in which, according to noted entrepreneur Elon Musk, 99% of compute long term will be for inference.
My colleague Jean Luc Di Manno, the Innovation Lead in Payments over at Fime came up with a very nice way to help us think about such things. He said that today’s commercial system was built on human emotions (such as desire, aspiration and impulse) and then asks a key question: How do you seduce an agent?
How indeed.
Perhaps it is time to embrace the wacko again, because there is a new science of behavioural economics for bots waiting to be born.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.