Work Itaú Private Bank
01 · Relationships
Eleven clusters, and then one client at a time.
I ran the scope of a clustering project and directed the data science team, so that offers, service and communication could stop treating a highly differentiated base as one audience.
- Role
- Project scope and direction of the data science team
- Organization
- Itaú Private Bank
- Inputs
- Behavioral and consumption data
- Output
- 11 clusters, and individual profiles inside them
01 Context
A base that was never one audience.
Across nearly three years at Itaú Private Bank I worked on strategic experience challenges spanning digital banking, investments, credit, onboarding, personalization and relationship models.
The clients of this segment have almost nothing in common except wealth. They differ in how they earn, how they spend, how much they want to be approached, and how much of their financial life they run themselves. Treat them as one audience and every offer is wrong for most of them.
02 The model
Behavior and consumption, not personas agreed in a workshop.
I directed the data science team to build a clustering model from behavioral and consumption data. Then I asked them to go one level further and single out the variables that actually changed behavior inside each cluster, down to the individual client.
I asked for the model to be built from data and history, because a model built that way can be checked against what clients actually did. That is what makes a commercial team willing to change the way it has always served its clients.
03 What we did with it
One client at a time, not eleven ways to be a client.
The clusters told us how a group behaves. The level below them told us who we were actually talking to. That is where the personalization happened: which offer belongs to this client, which service model fits them, which message is worth sending at all, and when.
So the answer was never eleven ways to be a client. It was one individual way of being a client, read inside the group that gives it context.
04 What we could suddenly see
Reactivating them without overloading the commercial team.
The question was never only how to find portfolios that had stopped moving. It was how to reactivate them without adding load a commercial team could not absorb. So we used the model to write personalized communications that gave each client a reason to reinvest, instead of handing relationship managers a longer call list.
We read the same client from four angles: their investment profile, the moment they were in, how digital they were, and how often the commercial team was already in contact. Then I worked with communication, marketing and service to turn that into decisions they could make. Which client to approach, with which argument, through which channel, and at which moment.
05 The reframe
What actually had to align.
- Client expectations
- Data
- Relationship models
- Business priorities
- Technology
- Commercial service
The app was the visible layer. The real design challenge was aligning client expectations, data, relationship models, business priorities, technology and the commercial service itself.
05 How I ran it
What I asked the data team for, and why.
I ran the scope of this project and directed the data science team. My job was not to build the model. It was to keep it pointed at what the business actually needed, so that what came out of it could be used by the people who talk to clients every day.
-
I started from the business need, not from the data
Before anyone opened a dataset I wrote down what the bank needed the model to answer. How do we meet the expectations of clients who are used to being known by name. How do we grow investment volume. Where is there opportunity sitting untouched. Every request I made to the data team traced back to one of those three.
-
I asked for the clusters, then asked what was inside them
Eleven clusters told us how groups behave, which was useful but not enough. So I asked the team to go further and single out the variables that actually changed behavior inside each cluster, down to the individual profile. Two clients can sit in the same cluster and be at completely different points in their lives.
-
I asked how to reactivate quiet portfolios without burying the team
Finding the portfolios that had stopped moving was the easy half. The real question I handed the team was how to bring them back without adding load a commercial team could not absorb. We answered it with personalized communications that gave each client a reason to reinvest, rather than a longer call list for their relationship manager.
-
I asked for the client's moment, not only their profile
Alongside behavior and consumption we looked at how digital each client was, how long the relationship had run, how often we were in contact, which products they already held and how satisfied they were right now. A recommendation can be right for the person and wrong for the moment, and when that happens it still lands as noise.
-
I took the model to the teams who had to act on it
I worked with communication, marketing and service to turn the output into decisions they could actually make. Which client to approach, with what, and at what moment. A model that nobody downstream can use is a slide, not a capability.