BBVA

Solution

BBVA built an in-house AI analytics workflow to measure the causal impact of its digital financial health tools on customer savings. The process uses anonymized customer profile and behavior data from Spain and Mexico, including age, income, spending level, tool usage, and resulting account balances. Because the tools were already available to all customers on the BBVA app and website, the data team could not use a conventional A/B test. Instead, it applied causal inference to identify confounders and create matched 'twin' socioeconomic profiles that differ mainly in whether the tools were used. BBVA then ran a double machine learning approach with two AI models: one estimated how confounders affect the likelihood of using the tools, and the other predicted expected balance changes for each pair. By comparing predicted and actual savings across users and non-users of features such as Movimientos Previstos, Mis Presupuestos, Objetivos de Gastos, Mi día a día, BBVA Plan, and Apartados, BBVA isolated the savings attributable to the tools themselves and fed that evidence into digital strategy decisions.

Data flow

BBVA takes anonymized customer data from its app and website, including age, income, spending level, feature usage, and account balances, and groups similar customers into matched 'twins.' A first AI model estimates how confounders affect tool usage, a second predicts expected balance changes, and BBVA compares predicted versus actual savings to isolate the effect of each financial health tool.