Transaction Data Is the Most Powerful Predictor of Future Behavior

It may be a marketing cliché that “most customers are predictable.” But like most sayings with staying power, the phrase is true: Past behavior is, indeed, the best predictor of customers' future behavior. Movie lovers will stay movie lovers, even while media and delivery mechanisms may change (e.g., jumping from theater to Blu-ray to streaming services). Similarly, early adopters will remain early adopters. And bargain-seekers will remain bargain-seekers, unless there are major shifts in their lives. So, in the business of prediction for targeting and personalization, we must cherish behavioral data as a top priority. Out of many types of behavioral data, transaction data is the most powerful predictor. Without a doubt, records of monetary transactions are stronger—whenever available—than other behaviors such as opens, clicks, views, calls and visits, which often serve as backup predictors. Now, transaction data must answer the following questions to be useful: Who bought what product?How much did they spend?When and where was the purchase made?What payment method was used? Let’s dig a little further here, as the devil is in the details in data business. Even if you may not be a hands-on data player, you need to know what elements are critical for your personalization endeavors (beyond responding to opens and clicks, that is). Customer identity “Who” is made of PII (Personally Identifiable Information) such as name, address, email, phone number, customer ID, etc. Something that distinguishes one customer from another. The key word in Customer Data Platform (CDP), CRM, or customer-360 is the “customer.” Even advanced Identity Graph solutions will not be able to connect disparate data without the PII. Just be careful with privacy rules—of different nations and industries—when dealing with it. You may need to encrypt them all. Products and services “What product” gets a little bit tricky, so let’s start with the basics: SKU number, product descriptions and some types of product categories. I’ve seen some messed up product taxonomies, but in the beginning, something would be better than nothing. Most are marred with missing and inconsistent values, but with human perseverance and some help from AI, we can fill in the blanks. Don’t be too scared, as you’d never have to categorize them all—just categorize the most popular items (in terms of sales dollars) and ignore the ones that didn’t sell much. We are in the prediction business, not in the business of creating immaculate data. Dollar amount “How much” is relatively simple, but get ready to deal with inconsistent ways of recording paid amount, net unit price, discount, coupon redemptions, tax, shipping, returns and cancels. If they are too messy or complicated (most systems are, especially for full and partial returns), I’d take any number that resembles the final paid amount. For marketing purposes, the data doesn’t have to match those of accounting system exactly. If you can figure out “percentage discount” for each order, that would be a bonus, as it will help you differentiate bargain seekers from full-price buyers later. You should never treat them the same. Date and time “When” is the easiest one, but make sure you keep the local date and time of the transactions. UTC time stamp is irrelevant to customer analyses unless you sell things only in the United Kingdom. When studying customer behavior in terms of daypart or day-of-the-week, you’d really need the local time (based on location data). Channel, store and site The “where” part is totally on you. How cleanly is the data on the channel, site, store, franchise and device maintained? I’ve seen a large amount of missing location or source information in so-called state-of-the-art retail and digital data infrastructures. If necessary, just go at it again (starting with the significant ones in terms of sales). Payment method “Payment method” is a bonus item, as it is a good predictor of customer behavior. Certain types of credit card indicate “luxury buyers.” Cash and cash on delivery are linked to distinctive demographics, as well. We definitely do not need the whole credit card number for customer analyses, as it is an extra security risk. Just the types of swiped cards and other payment methods such as “financing” or “Apple Pay” would suffice. Non-retail businesses If your business model involves subscription or continuity programs, things get more complicated. As a starter, you’d have to maintain dates of first subscription, renewals, payments, delinquency, cancellation and reactivation, along with current subscription and auto-payment statuses. If you are in the hospitality business, you need to add dates of booking, payment, travel, arrival, check-out and departure. Distances among these dates tell a lot about the traveler. For example, we can sort out business travelers by examining how early they made the reservation, along with their booking method, the site they used and the deals they took. For now, please try your best to keep the basic elements clean and handy for all customers within the past four years at a minimum. I’ve seen so many ambitious targeting and personalization projects stuck in the data collection stages. Would you believe some reputable companies couldn’t pull out reliable product information and related prices easily? Don’t be that kind of a company in this day and age. Not All Data Types Are Equally Useful