A financial institution had sophisticated capabilities for understanding customer behaviour and future value. Yet its advertising platforms were still optimizing acquisition using three broad, static customer-value categories.
The problem wasn’t a lack of analytical expertise. It was the gap between what the organization could understand about its customers and what its advertising systems were being told to optimize for.
Finding the opportunity
Through our work on acquisition, we recognized an opportunity to apply existing predictive intelligence to the signals guiding advertising investment.
An analysis of historical customers revealed substantial differences between the values assigned by the existing system and customers’ modelled future value. Some lower-value relationships were being overvalued, while the potential of higher-value customers was significantly understated.
Working with the bank’s data science team, the approach evolved toward a more differentiated valuation framework based on customer balances, engagement, and expected future outcomes.
But producing a better model wasn’t enough. The values also needed to reach advertising platforms in a form—and at a time—that made them useful.
Designing for how advertising actually works
Automated bidding relies on timely conversion feedback. The bank’s more informative customer engagement scores, however, weren’t immediately available after acquisition.
Waiting for a complete picture meant delaying the information the advertising platforms needed.
- At acquisitionCustomer acquiredAvailable information only
- Right awayProvisional value sentBased on available information and the historical likelihood of reaching each value segment
- LaterEngagement score availableA better-informed signal
- ThenValue updatedReplaced with the value of the customer’s observed segment
We helped develop a two-stage valuation approach. Before the more informative customer score became available, the system would assign a provisional value based on available information and the historical probability of reaching different value segments.
As additional information became available, that estimate could be replaced with the value associated with the customer’s observed segment.
The objective wasn’t simply greater numerical precision. It was to preserve meaningful differences in relative customer value, even while the available information evolved.
We helped translate that logic into advertising requirements, data structures, and a measurement approach suitable for controlled experimentation.
Putting existing intelligence to a new purpose
The bank’s data science team developed the underlying predictive model. Theria recognized its potential application to acquisition and helped define the operational logic needed to make it usable.
The work established a more differentiated customer valuation framework and an approach for testing its impact on acquisition performance.
Rather than building an entirely new analytical capability, the opportunity was to put an existing one to work in a way the organization hadn’t yet pursued.