For a leading Peruvian consumer bank, we built a behavioral model that predicts which barriers hold back digitalization and ranks which ones to tackle first.
In the middle of the pandemic, the bank needed to move more customers to its digital channels. The qualitative phase surfaced 28 possible barriers and enablers. The problem: without knowing which ones actually matter, any plan risks investing effort in things that don't move the needle.
The bank wanted to move from intuition to evidence-based prioritization. The question wasn't "what do customers think?" but a more demanding one: which determinants predict that a customer goes digital, and which are just noise that accompanies the behavior without causing it?
We combined a qualitative phase to generate hypotheses with a quantitative one to test them, all organized around the COM-B model (Capability, Opportunity, and Motivation) of behavior change.
In plain terms: a person adopts a behavior only when three things come together: being able to do it (capability), their environment allowing it (opportunity), and wanting to do it (motivation).
In-depth interviews with customers at different levels of digitalization. That surfaced 28 statements about possible barriers and enablers, organized by capability, opportunity, and motivation.
We surveyed 2,220 customers and classified them into three profiles based on their digital usage: Non-Digital, Neo-Digital, and Digital. Each statement was measured on an agreement scale.
Using logistic regression, we separated what actually predicts digitalization (significance p < .01) from what doesn't. The result: a prioritized list of drivers, with their weight and direction.
Who we measured. We classified customers into three profiles based on their use of digital channels:
Most market studies describe what your customers think and do. That's useful, but it stops there. We combine qualitative and quantitative methods with a more ambitious goal than a typical study: modeling what predicts the behavior and ranking what to move first.
Each method does a different job. The qualitative phase generates the hypotheses: in-depth interviews surface beliefs no one would think to ask about, like "the printed receipt is my guarantee." The quantitative phase puts them to the test: it measures which exist at scale and, above all, which predict digitalization.
Most studies stop at describing and comparing. The value for decision-making is in predicting and prioritizing.
We classified each customer as Digital or Non-Digital and used logistic regression (a statistical model that estimates how much each factor weighs) to see which determinants predict that outcome, isolating each one's effect. We required a strict threshold (p < .01) to avoid confusing noise with signal, and expressed each effect as the change in probability that a customer becomes digital. That turns the statistics into an actionable priority list. The model explains 40% of digitalization, a high figure for such a complex behavior; the rest marks where to keep digging. Base: 2,220 customers.
Relative weight of each driver within the predictive model. Client anonymized for confidentiality.
By modeling what predicts digitalization, the bank was able to focus its efforts on the few levers that actually move the behavior, and stop investing in the ones that just tag along.
Non-Digital customers don't reject digital banking: most simply don't know what to expect or how to operate it. The gap with Digital customers is biggest exactly on the three predictors that matter most.
A survey tells you what your customers think. A predictive model tells you in what order to act. The difference between the two is where the budget is won or lost.
The result isn't a report to file away: it's an investment decision. Instead of spreading effort across 28 loose ideas, the bank knew which 3 or 4 levers to concentrate its budget on, and which to discard for not moving the needle. That's money that stops being spent blindly, before investing in campaigns or redesigns.
With the determinants prioritized, the natural next step is to design interventions that break down those barriers and validate them with experiments (testing each idea with real users) before scaling. That's how you go from "we think it works" to "we know how much it works."