Does your data accurately reflect your customers’ behavior or does it tell a limited story about when and why they bought your product?

Your customer data is a moving target. If you use machine learning to predict how your customers behave now, based on how they behaved at the time they bought your product, then you cannot expect to be accurate. Put simply, if you sell lawn mowers, and a portion of your customers move to the city, then associating their current address with that purchase is a disaster for your predictive modeling. It may sound counter-intuitive, but past data is often more predictive than current data.

At CTI, this is why we build data curation strategies that embed necessary point-in-time data within existing structures, giving you the best of both worlds, and making relevant data readily available/useful for analytics and decision-making (

Why does this matter? Look at the basic ways your life can change over time, and the implications for what and how you buy. Consider the story and buying habits of young Mary, starting when she moved out of her parents’ house and became an independent consumer:

  • Moves in with roommates:
    • Buys whatever is cheapest, cannot afford favorite brands
    • Avoids discretionary items
    • Vacations are local
    • Uses Uber or shared vehicle
  • Is still single, but now living alone:
    • Can afford to buy favorite brands on sale
    • Stays at cheap hotels
    • Buys fashionable items on a budget
    • Frequent take-out food
    • Buys economy car
  • Gets married:
    • Dual income, buys some discretionary products
    • Eats at home more
    • Buys favorite brands (finally!)
    • Cuts back on fashion
  • Has a baby:
    • Buys products for babies
    • Eats home delivery
    • Buys bigger car
    • Lives on credit
  • Changes jobs, career on the rise!:
    • Takes more expensive vacations to beaches
    • Buys second car
    • Buys new furniture
    • Impulsively spends with credit cards
  • Buys house in the suburbs:
    • Needs lawn mower, garden supplies
    • Solar panels for roof
    • More furniture, artwork
  • Has another baby:
    • Buys minivan
    • Kids clothes, afterschool activity supplies
    • Saves money
  • Empty Nest – Sends kids to college:
    • Travels abroad and stay at nicer hotels
    • Buys sports car
    • Pays off loans
  • Gets divorced:
    • Downsize, changes spending patterns
    • Moves back to city
    • Eats out again
    • Dating, but this time it’s theater and nicer restaurants
    • Vacations to cultural attractions rather than beach
  • Remarry, retire:
    • Buys a house at a lake
    • Spends less on favorite brands due to changing tastes
    • Buys comfort items for elderly living
    • Buys sedan with big trunk for new golf hobby
    • Vacation to Vegas

Now let’s examine how Mary’s changing buying habits impact your marketing. When she bought your product, she was likely in the sweet spot of what describes your best customer persona. Along the way, she shifted into new phases, making her less likely to repeat buy. Moreover, her current profile is misaligned with her sweet spot when she was a buyer. This misleads your predictive analytics because you may be pairing more recent data with her purchase history. Your regression models make it appear that more recent characteristics are associated with purchase behavior, and your lists are off-target. You may even start to think your product appeals to a slightly off-target population, which can affect your ad placement and messaging.

How do you properly capture and curate all of this customer information over time?

One common mistake that marketers make is applying current profiles to previous behaviors, and not differentiating data that is out-of-date from intentionally collected and managed historical data. Point-in-time database construction captures both aspects — historical data that is up-to-date as of the time the behavior occurred. Think of this as a series of activity ‘freeze files’ that coincide with purchases.

This information is vital for marketing attribution as well i.e., marketing information must also be included in our ‘freeze file’. And what about price, offer and product details? Back-testing requires these historical details to be in sync as well. If prices are different in the current offer, this must be considered.

Another common data curation mistake is thinking that keeping a transaction date and timestamp is the same thing as keeping point-in-time data. While it is true that recording the transaction dates is important, they do not contain information about the circumstances of the purchase. What was the person’s life stage at the time (remember Mary’s journeys)? Where were they living? What marketing communications had they received leading up to the purchase?

It’s also important to note that your company’s business is moving through life-stages of its own, and data tables about changes in product life cycles must be a part of the ‘freeze file’ mentality.

How is this different from simply archiving our database?  It’s different in two important ways. The data curation structures we are describing bring forward critical point-in-time information into the current database.  So, your archive would have historical data embedded in it.  Also, data archives are mostly repetitive in that most of the information stays the same from archive to archive while certain values change at different times. A better data curation structure compresses this information while capturing/reflecting different changes at different times.

It’s exciting for today’s marketers to leverage powerful innovations in data science, ML/AI, and advanced analytics and improve their understanding of how customers behave. However, positive outcomes will be limited if we don’t consider data curation strategies that accurately capture and tell the story of your customers, how they behave today, and how we want them to behave tomorrow.

For more information on our consulting services contact us at

© Corporate Technologies, Inc.   |  Privacy & Legal