Stop waiting to clean your data before you personalize
BCG's personalization chief says basic customer and purchase records are enough to run a first test, and the results pay for the rest.
The Strategy Skills Podcast: Strategy | Leadership | Critical Thinking | Problem-Solving · 662: Customer Strategy in the Age of AI with Mark Abraham, BCG's Personalization Lead (Strategy Skills classics)Mark Abraham, who leads the personalization practice at consulting firm BCG, has watched the same project stall in company after company. Leaders say they do not have the right data foundation, so they propose spending three years and a lot of money to fix it before they try to tailor offers and messages to individual customers.
Abraham says, "That is always a path to failure." He recently checked back with a large consumer packaged goods company that took that route, and it had made no progress on results. His blunt explanation is that nobody has a perfect data foundation.
His method flips the order. Pick one use case first, meaning one specific way you are trying to help a customer. Then launch a bare-bones first version using the data you already have, which at a minimum is usually customer data and purchase records.
You do not need perfect labels on that data, and you do not need perfect technology. Abraham says much of the early testing can even be done by hand. Once you run that small test with a comparison group that does not get the personalized treatment, you will see what actually lifts sales.
Only then should you add data. That test tells you exactly what is missing and worth tagging or collecting, rather than trying to clean everything at once.
That proof is also what unlocks money. Once you can show what drives growth, you can go to the finance chief and make the case for more investment.
The final step is to make it a companywide job, not just marketing's project. Abraham argues personalization has to sit on the chief executive's agenda, with operations and every other leader clear on their role, because it changes how teams work every day.
Start this week by listing one customer problem you could solve with the names and receipts you already hold.
Questions people ask
What counts as the basic data I already have?
Customer data means names, contact details, and account identifiers, while transaction data means what someone bought, when, and for how much. Together they let you group people by past behavior, send a relevant offer to some and not to others, and compare the sales difference. It is basic, but it is often enough to prove whether personalization pays.
How do I pick the first use case?
A use case is one narrow job, such as reminding shoppers when they are likely to run out of a product or suggesting a useful add-on. Start with one where you can see a sale or repeat visit quickly. A tight test makes it easier to measure results and to learn what extra data you actually need.
What is test-and-control measurement in plain terms?
You split similar customers into two groups, give one group the personalized message or offer and give the other group either nothing or the usual mass promotion. Then you compare purchases over a set period. Because you track individuals, you can tie the difference in behavior directly to the test.