Transaction categorisation definitely sounds like something Marie Kondo would be into if she were a data scientist. But it’s about much more than keeping banking transactions neat and tidy in your personalisation platform. It creates an all-important base for understanding financial transactions and catering to the needs of the banking customers behind them. A robust categorisation model also helps banks ensure that each transaction is interpreted and handled correctly and consistently across the entire organisation. Let’s see the key aspects of transaction categorisation and how they can drive banks’ personalisation efforts.
1. Transaction IDs
Banking transactions are a treasure trove of information. Besides basic variables, such as transaction type (bank transfer, card payment, cash withdrawal, standing order, bank charges and the like), they come with a great deal of identification data, including card numbers, account numbers and transaction amounts. Plus, banks provide other key input variables like merchant category code (MCC), merchant name, narrative and so on.
In W.UP’s case, the platform’s automated categorisation engine takes all this information and breaks down transactions into categories using mapping or, in more complex cases such as POS transactions, based on keywords. Together, these categorisation methods result in a categorisation coverage ratio of 93-98%, thanks to our extensive library of local and global keywords, plus an online keyword database that we update regularly.
And that’s just the beginning.
2. Personal financial management (PFM) tags
Ideally, a personalisation platform not only categorises but also tags transactions. Using over 20 categories and more than 130 PFM tags, W.UP can catalogue almost any transaction into a main category (e.g. car-related spending) and most of them into several subcategories (e.g. petrol station purchase). If this sounds rather ambitious, that’s because it is. To make sure that the categorisation engine has got everything right, we usually start with an offline categorisation phase long before the go-live date. We cleanse the input narratives, carry out a keyword analysis and train the model using pattern recognition at a speed of 4,000-6,000 transactions per second. Once the engine is up and running, it’s also crucial to monitor existing and new tags to winnow out duplicates and run regular accuracy checks.
3. Brand names
In addition to categories and tags, personalisation platforms can also differentiate between transactions by brand. Why is this important? It allows banks to create customer microsegments and laser-target their campaigns. Let’s say a bank teams up with Tesco for a campaign to boost credit card usage among its customers. Thanks to systematic, consistent categorisation, it can easily identify customers who regularly shop at the retail giant with a debit card – and offer them 2% cashback if they use their credit card instead.
Favourite brand and identifying repeated payment use case examples of W.UP
4. Transaction regularity
How frequently a transaction occurs adds to the complexity of transaction categorisation but also gives a clearer picture of a customer’s needs and preferences. It’s one thing to know if they spend money in cafés, for instance. Knowing if they do so once a month, once a week or every single morning helps banks fine-tune their offers. As a result, some campaigns might reach only a handful of people but will result in a much higher conversion rate and customer loyalty than the good old spray-and-pray approach.
And that’s still nowhere near the end of it. You can go deeper and assign outlier scores to transactions to detect key life events, like buying a first home or a new car, or label deposits according to income type. The bad news is that transaction categorisation is not something you do once and you are done with. The good news is that it’s as high-reward as it is high-maintenance.