#22 - Personalisation in Consumer Finance
How AI and ML is driving personalisation in consumer finance and some of the considerations around this.
Hi all - This is the 22nd edition of Frontier Fintech. A big thanks to my regular readers and subscribers. To those who are yet to subscribe, hit the subscribe button below and share with your colleagues and friends. 🚀
Summary
This week I take a look at personalisation in consumer finance. Looking at what Netflix, Amazon and Spotify have done in other industries, it’s only a matter of time before such levels of personalisation are witnessed in consumer finance. Nonetheless, firms will need to understand their north star as regards personalisation and the application of AI and ML in consumer finance. In my view, the main aims should include solving for bounded rationality and cognitive bias in personal financial management.
Some of the areas in which personalisation can add value include recommendations, choice architecture, peer analytics and automated finance. Some of the challenges that face firms are data, data architectures, culture and expertise particularly in Africa. Lastly I review some of the key players in this space include Personetics, Kasisto, Strands and Yodlee. Scale is one of the most interesting to watch over the long term.
Introduction
In February of this year (2021), Spotify launched its services across Africa and I was one of the first to download the app. Prior to this, I had an Apple Music account and at the time I thought it was a good service and it had playlists that I liked. However, downloading Spotify was a mind blowing experience as to what exactly a music streaming service should do. Since I downloaded Spotify, I’d say the amount of time I spend listening to music has shot up by magnitudes of time. I now listen to music on the app all the time including even now as I write these words.
The memes above capture the thoughts of everyone who has listened to music on both Spotify and Apple Music. The reason is simple. Spotify’s Artificial Intelligence and Machine Learning capabilities are significantly more advanced than those of Apple Music. Interestingly, Spotify has a bigger data pool than Netflix. Some of the interesting techniques that Spotify uses to improve its algorithm are; collaborative filtering and natural language processing. In collaborative filtering, Spotify provides recommendations to users based on the preferences of other users with similar tastes. In Natural Language Processing, the company analyses songs, blog posts, articles and song metadata to generate tags for each song. It then compares these tags with those for other songs to create linkages between songs which at first brush may not seem related. All these factors contribute greatly and it’s why you can have Wiz Kid on the same playlist as Adele.
Why does all this matter? Just like Netflix, Spotify has revolutionised the way people consume media. If music makes us feel good and uplifts our spirits, then Spotify is the biggest “happiness” company in the world. What’s worth noting is that Spotify can compete in this category with Apple despite Apple’s significant resources based solely on its focus on its algorithms.
As customers live more of their lives within the digital realm, this sort of personalisation will be expected across all their experiences. Amazon led the charge with personalising the shopping experience particularly through its rich recommendation engine. It has gone further with Alexa. Customers will come to expect the same personalisation in financial services. Financial service providers must consider how AI and ML will drive their product strategies in the future particularly as regards to personalisation.
Personalisation Strategy
The discussion around personalisation and the role of AI and ML in consumer finance is subject currently to a lot of noise. Additionally, previous attempts by banks have been focused on edge events that have little to no fundamental impact on the financial lives of their clients. These have included features such as customising the colours of the bank app, remembering your birthday and countless of ‘personalised’ notifications that don’t add much value.
Without the north-star of what you’re trying to accomplish with personalisation, then most investments into personalisation may be wasted. This will likely lead to a feature-based product strategy that doesn’t solve any fundamental problems.
What then is the problem to be solved? The following passage from the Stanford Encyclopedia of Philosophy is useful;
The perfect rationality of homo economicus imagines a hypothetical agent who has complete information about the options available for choice, perfect foresight of the consequences from choosing those options, and the wherewithal to solve an optimization problem (typically of considerable complexity) that identifies an option which maximizes the agent’s personal utility… Most formal models of judgment and decision making entail logical omniscience—complete knowledge of all that logically follows from one’s current commitments combined with any set of options considered for choice—which is as psychologically unrealistic as it is difficult, technically, to avoid (Stalnaker 1991).
In my view, investments into personalisation should solve for bounded rationality captured in the above passage. Human beings suffer from the following constraints when managing their finances and making decisions in general;
Limited processing power when analysing multiple choices - this often leads to heuristics such as saving 25% of your salary and not spending more than 35% on rent. In worse outcomes, people just give up completely and spend wantonly;
Imperfect information - this is simply a data problem and compounded by the siloed nature of data. Additionally, there is the issue of information overload that makes it difficult for people to separate useless information from valuable information;
Cognitive biases particularly in the realm of personal finances and investments. These include;
Confirmation bias - where we are more likely to seek or emphasise information that concurs with an existing belief or hypothesis;
Loss aversion/endowment effect - this happens when someone places a higher value on a good that they own compared to a similar good that they don’t own. It leads to a situation where people prefer avoiding losses than making gains;
Group think - where people gain comfort in something because many other people are doing the same;
Restraint bias - the tendency to think that we can resist temptations. We are wired to be greedy - in investments and finances we tend to place all our bets with a “sure winner”. Risk controls are designed to cater to restraint bias;
Anchoring bias - this is the tendency to anchor or reference one recent piece of information when making decisions. For example, we base future interest rates based on the most recent rate even if future rates may have no linkage to the current rate;
Nudge by Richard Thaler and Cass Sunstein as well as Thinking Fast and Slow by Daniel Kahneman are all useful reads. They both discuss cognitive biases, heuristics and how the theory of economic man breaks down in real life.
Thus the north-star of AI-driven personalisation should seek to solve the above issues. Indeed, big data solves for imperfect information whilst algorithms can be designed to cater for the cognitive biases. In the broader perspective of democratisation; personalisation can democratise sophisticated financial expertise. A private banker on your phone for everyone.
The north-star should thus be; how can a financial service provider use AI and ML to improve the financial lives of their clients by solving some of the fundamental issues in personal financial management. Personal financial management just like corporate finance for business is very important for long term wealth creation and financial resilience.
How Personalisation can add value to both banks and clients;
There are a number of interesting features financial service providers can provide to their clients within their overall personalisation roadmap and add value to both the client and the institution. Note that I don’t use the term “banks” largely because technology is completely changing the landscape of financial service provision and to deliver world class financial services particularly in Africa, the consumer facing layer may not necessarily have to be a bank.
Choice Architecture
Customers are faced with numerous financial decisions to make and therefore financial service providers will have to develop intuitive choice architectures. Currently, a number of efforts within the personalisation space have been driven by primitive product recommendation strategies with logic such as “sell a mortgage to someone in their mid 30s”. Products or apps should be built with choice architectures that are likely to improve a clients long-term financial health. For instance, most customers find it hard to opt-out of something. Savings products as well as life insurance products can be built into a product from the start with the customer having to choose whether to opt-out of the product. Currently, most products are designed with “opt-in” architecture that is less sticky. Of course regulatory considerations will have to be reviewed. Nonetheless, it’s easy to design products and auto-enrol clients whilst asking them to opt-out.
Peer Analytics;
Our brains are wired for social proof and this is one of the issues I mentioned above as “group-think”, it’s just an inescapable part of humanity. AI driven personalisation should lean into this trend and design customer experiences around this. Such a product could for instance include notifications such as “the average spend this month on groceries for customers such as you were 30% lower than yours” and “fellow customers are saving 40% more on vehicle insurance by taking the following actions…”.
Clients want relevant insights that are personalised and actionable. Using modern technologies, firms can add value to their clients by embedding peer analytics. This will likely drive significant customer engagement within your platforms and increase retention rates.
Self-Driven Finance
In a recent post on consumer finance apps, I noted how the previous vintage of apps that included “Personal Financial Management” apps failed to take off. Of course, we have already analysed this in part. Giving people tools to manage their finances doesn’t solve the issues mentioned above such as bounded rationality and imperfect information. Additionally, the obvious negative feedback loops that arise from PFM tools don’t work well in our dopamine fuelled world where people seek out positive feedback loops from apps such as TikTok and Instagram.
Automated finance or self-driven finance involves a financial service provider making the decisions on behalf of the client based on numerous data points and then asking for authorisation from the client. For instance, a typical experience could include “your health insurance is due for renewal within the next two months, you can save x% by shifting to this provider, should we set aside x per month to provision for this expense and acquire insurance from provider y?”. This adds real value to a client. Remember, we have bounded rationality and processing capabilities and thus most people can’t make such calculations on the fly.
Financial Well-Being and Long-Term Financial Outcomes
It seems pretty logical that if a financial service provider has financially stable/resilient clients, then in the long-term this is good for business as it will manifest in lower defaults and growing deposits or AUM. Thus investing in AI driven personalisation guided by a sensible north-star is good for long term business with the key benefits being both long-term financial well-being and increased customer loyalty.
Additional benefits;
Some additional benefits include;
Integrating intent with external content - you could for instance attach an intention to save for a holiday with content about where to buy cheap tickets, cheap and affordable hotels, how much other customers spent on their holidays and other such features. Tinkoff is doing this well with their super-app;
Managing subscriptions by reviewing your subscription history and cancelling subscriptions that are not adding value to your life;
Help with explaining complex legal matters for instance a mortgage agreement. AI can help in explaining legal agreements by explaining what is standard and can’t change and focusing on what matters. Some companies such as Ocrolus are working on this
Understanding a customers interaction channels e.g. ATM, mobile, web and branch and delivering content and insights at the point of interaction;
Ultimately personalisation is becoming a critical differentiator and over time, the Spotify effect could render some financial institutions irrelevant. These will be meme’d away in the same way as Apple Music.
Challenges around personalisation
The case for personalisation is very clear and when viewed together with the macro-trends in Africa such as young and growing population together with increasing connectivity and smartphone penetration, then it’s clear that more investment needs to be done in this regard towards. Nonetheless significant challenges abound. These include;
Data and Data Architectures
There are significant issues around data and the architectures around this. This in fact should be the topic of a stand-alone blog post. Data is emerging as the fourth factor of production in the modern economy. Nonetheless, existing institutions are not built with this reality in mind.The issues around data include siloed data, identity, data mapping and IT architectures that add value to this data.
As regards siloed data, this exists across a clients life. For instance, how many kids you have, their ages, your previous medical records, your driving habits, your spending habits and personal preferences such as enjoying to travel should all have a role to play in designing a personalised financial experience. Nonetheless all this data sits in separate databases with organisations that don’t want to share it. If indeed it is aggregated, it will be aggregated by a non-bank. In China, we have seen this with Ant and Wechat with the government now forcing Ant Group to share its data with the government.
Once the data is solved for, the next issue that emerges is that most bank architectures are not designed for big data. Concepts such as data lakes, data streams and off-line and on-line deployments are yet to be executed. Nonetheless, most banks are slowly working towards this. Here’s an interesting post about how Netflix built their data architecture for their recommendation engine. McKinsey have a good article on how to design data architectures for personalisation.
Expertise
Source: Statista
The graph above shows the number of AI publications per country from 1997 to 2017. It’s a useful proxy of the depth of AI expertise which can be extrapolated to data science and analytics expertise per country. Of note is the dominance of China and the USA with the two countries having almost the same number of publications as the next 12 countries combined. Additionally, the lack of expertise in Africa is clear from the above diagram. What this means is that financial service providers in Africa will have to make the decision as to whether their AI and ML strategy will be proprietary or vendor-based.
Guiding this strategy should be the same “FUBU” consideration that I keep referring to. There’s no TikTok for Africa and neither is there an African Whatsapp. What this means is that sub-optimal African products will lose out to globally competitive products. The choice should be to partner with AI as a service providers. When one considers the “h-index” in computer science, a metric that compares the productivity of an academic author; South Africa has a h-index of 2 compared to 2,799 in the USA. This should also inform government policy, investments in AI and data science education has very high ROI.
Culture and Approach
Ultimately the culture and approach towards personalisation by top leadership teams within financial service providers will be the key differentiator. Feature based products will lose out to products that are guided by a customer-centric north-star. In my experience, the issue with “features” is that you’re always faced with the “what next” question i.e. once you’ve done this, what’s the next logical step within this roadmap. Sometimes this is not clear. Stripe is the perfect example, with “increasing the GDP of the internet” as their north star, verticals such as Identity, Terminals, Tax and Radar are all logical steps within this trajectory.
Source: The Generalist
The diagram above from thegeneralist shows how Stripe’s product has evolved over time with each additional product adding value to the existing suite of products. Of course leadership should also provide guidance when the obvious push-back from IT departments emerges.
Some of the Players within this space
A number of interesting players have emerged within the space to help banks drive their personalisation journeys. These are;
Personetics - Personetics helps banks to enable personalisation within an omni-channel framework. They do this by leveraging their expertise with big data, AI, ML and analytics and integrating this to existing bank workflows. Interestingly, they have also received investment from Warburg Pincus which was featured in my previous article on Vodeno and Aion Bank both also investees of Warburg Pincus.
Abe.ai - Abe falls under the category of conversational AI vendors who largely build virtual financial assistants that integrate to apps;
Clinc ai - Clinc is also a conversational AI provider that uses natural language processing to help banks engage better with their clients.
Yodlee - Yodlee is a comprehensive platform that leverages open banking capabilities and AI to enable customised financial service provision. This includes conversational AIs, actionable consumer insights and personalised offers;
Kasisto - Kasisto seems to be similar to personetics with a personalisation offering that includes virtual assistants, personalised offering and actionable insights. Some of Kasisto’s customers in Africa include Nedbank, Absa and Standard Bank. Globally Kasisto has worked with DBS Bank.
Strands - Strands seems to work similarly to Personetics and Kasisto as well - some of their customers include Santander, Metro Bank and HSBC. NCBA in Kenya partnered with Strands for their Loop product;
W.UP - Similar to Kasisto and Personetics by offering a personalisation platform that banks can integrate to;
One of the players I am really watching is a company called Scale, Scale is a data platform for AI. They started by labelling data for ML algorithms for firms such as Google and Tesla particularly in the domain of autonomous driving. Nonetheless, they’ve built up capabilities across the data and AI value chain from data labelling to detection, natural language processing, video analysis and document reading. This deep dive by the ever mercurial Packy M is a must read. It seems that financial service companies will have to integrate their capabilities in the long-term to help with customer service, personalisation, recommendations and most services within the consumer lifecycle.
It’s clear that personalisation will be a key strategic imperative for financial service providers. It will be interesting to see how this space evolves. In my view, fintech apps have an advantage in this regard. Personalisation could be one of the factors that drive the continued commoditisation of banks.
As always thanks for reading and drop the comments below and let’s drive this conversation.
If you want a more detailed conversation on the above, kindly get in touch on samora.kariuki@gmail.com;
Great read as always