In the second in our three-part series in which we discuss the standout trends that are defining the future of digital banking, we look at automation through artificial intelligence, and accomplishing marketing personalization at scale
Interpretations of automation have advanced with technologies and mindsets, and within banking it has come to be synonymous with artificial intelligence (AI) and machine learning (ML).
It’s anticipated that front office – conversational banking, AI biometrics technology and personalized insights – will save banks up to $199bn, according to Business Insider Intelligence. It is now a part of the mainstream competitive digital banking landscape: In a survey conducted by UBS Evidence Lab, 75% of respondents at banks with over $100bn in assets said they are now implementing AI strategies, compared with 46% at banks with less than $100bn.
Automation and personalization within banking can be segregated into three distinct domains: product, marketing and service. Product personalization is still largely manual, with platforms providing A/B testing and multivariate testing tools. Service personalization and automation is the domain of chatbots like Bank of America’s Erica, or Capital One’s Eno, which are fueled by natural language processing (NLP) AI and service clouds from the likes of Oracle, Adobe, Salesforce or Pegasystems. Within marketing, AI applications tend to be limited to narrow optimizations, such as optimizing email send times, or headline text or hero images. But the next frontier is AI that automates the test-and-learn process that makes predictive personalization possible.
Today, marketing personalization is largely limited to rules or behavioral and time-based logic that “triggers” which messages should be sent once a customer meets certain defined criteria. The challenge, however is that all relevant triggers must be defined in advance, and a system must be put in place whereby the “right” messages are delivered to the appropriate customers who tripped the trigger. This takes a massive amount of analytical and marketing resource to define, test, and engineer – especially to avoid sending tone-deaf communications to edge cases. This means marketing teams quickly hit a point of diminishing returns on their personalization efforts, so what gets implemented are personalization campaigns that stop well before true personalization.
An example can be found in the checking account opening process. A user may have already activated their debit card, but they haven’t downloaded the app. Marketers may then wonder if they should send this group a specific nudge to download the app. To answer the question, they’ll work with the analytics team to conduct an analysis on the size and importance of people who fit that criteria. If it turns out that a large percentage of new customers activate their debit card but don’t download the app – and that those customers are less engaged, have lower deposits, or are more likely to churn – then it makes sense to set up a campaign against that group to see if marketing can drive app downloads.
To execute a campaign, they’ll work with an analytics or engineering partner to define a trigger in data and pipe that data into the email platform (for example, customer tenure < 45 days AND customer debit card activated = true AND customer downloaded app = FALSE) so that the marketing platform knows who to send a message to. Then they’ll create an email message encouraging these customers to download the app. Every customer meeting this criteria will receive the same communication unless the marketer sets up additional targeting criteria and creates new messages or manually sets up an A/B or multivariate test to identify the winning creative out of a set. One can quickly understand the point of diminishing returns that marketers quickly face. Personalization requires them to come up with paths for every possible situation that the customer may face, develop creatives for each situation, and code the logic in the system.
Instead, developments within AI and ML have allowed marketing to advance past the trigger-based approach, removing the resource drain to provide truly personalized messaging to individuals customers.
Amplero, for instance, uses AI to continuously optimize not only the topic (an offer, a brand message, a reminder), but how that topic is wrapped creatively (funny, direct?) and the channel(s) it is delivered in. The platform It even assesses the optimal delivery sequence (push notification followed by email). This is all made possible by the platform’s patented reinforcement learning algorithms, eliminating the need for armies of individuals to execute.
Going back to the checking account onboarding process example, marketers using Amplero simply have to generate creative options they think will drive stickier relationships and load them up into a library. The platform will use that library to automatically generate dynamic onboarding journeys for individuals – watching downstream behavior and adapting its decisions based on which elements are working or not on a daily basis. The downstream behavior it’s using goes far beyond the typical marketing KPIs, such as clicks or opens – instead focusing on KPIs that drive real business value, such as a primacy score or 45-day retention. No analytical hoop jumping or coding required.
“We’ve entered a new age of marketing,” says Sarah Welch, managing director at Curinos. “You need to approach marketing as a scientist would, with the mindset of testing, and learning. The faster you can test and learn and adapt, the better.”
Welch compares the situation to the way humans build relationships, using the example of a new customer opening a new account in a physical branch. To build the relationship and to productively match the customer with products and services offered by the bank, the agent is actually running a series of experiments in which they will throw out some stimuli – a question, a bad joke – then assess the customer’s responses, ultimately using their feedback to adapt the solutions they’ll focus on, and how they talk about them.
“The equivalent in a digital world is the interaction data,” says Welch. “A customer’s digital exhaust provides the feedback you need to drive personalization. But without AI, it takes armies of PhDs to mine that digital exhaust for actionable insights, and then more armies to set up and run ‘relevant’ marketing interactions and still more armies to run and read any experiments designed to identify the lift potential of greater personalization. With an engine like Amplero, you can set up libraries of potential stimuli to learn what works at an individual level –today and tomorrow (as each customer or the environment changes over time).
Amplero matches customer attributes with specific creatives to drive the relationship. Source: Amplero
As AI and automation have developed, consumers have come to expect and rely on AI and automation, which have been encouraged into the mainstream digital experience by the likes of Amazon and Netflix. Indeed, consumers now expect those levels of automation from their financial services providers. Banks in turn are keen to remove the friction of manual task completion, offering smoother and more direct journeys.
As we move into an era of open finance and heightened demand for personalized insights about spending, savings and credit scores, customers expect to be able to make better decisions about their money and to be informed about products and services specific to their needs. Leaders in digital banking recognize this, and are aware that those who master automation will sit ahead in the ecosystem. Personalization is likely to spread across every facet of the ecosystem of digital banking.
“There will be no such thing as a standard user journey,” says Welch. “Journeys will be dynamically generated and personal. Everybody should have a path of their own.”