- Banco do Brasil reports 4x contribution margin per client with H2O (FEBRABAN TECH 2025).
- H2O Driverless AI builds client margin, propensity and attrition models from the data your bank already holds, with reason codes on every prediction.
- The scores feed the CRM and relationship tools your bankers already use, so the way they work does not have to change.
- Every model ships with documentation for your model risk process and runs in the deployment option you choose.
Your bankers cover more clients than they can grow
Your relationship managers carry large books, and most of the margin comes from a small share of clients. Portfolios are often assigned by balance, segment or branch, so a banker's week goes to the clients who were assigned, not to the ones with the most room to grow.
You are measured on contribution margin and revenue per client, but a report of last year's profitability tells you where margin was earned, not where it will be earned next. Any model that moves clients between portfolios also has to make sense to the banker who owns the relationship.
How client profitability modeling works
Your analytics team builds explainable margin and propensity models, scores every client on a schedule, and sends the results to the tools your bankers already use.
Step 1: build the client margin view
Start from balances, services held, transaction and fee history, pricing, cost to serve and risk costs. Your finance team's contribution margin calculation becomes the target the model learns to predict.
Step 2: predict margin, propensity and attrition
H2O Driverless AI trains regression models for future contribution margin, and classification models for the chance a client takes up a new service or moves money elsewhere.
Step 3: assign portfolios and set priorities
Rank clients by expected margin and by the lift that coverage is likely to add. Your team decides which clients move into a managed portfolio, which stay in digital coverage, and which a banker should call this week.
Step 4: show the banker why
Reason codes built on Shapley values show which factors drove each client's score, for example a rising payroll flow or a service the client does not hold yet.
Step 5: measure, document and monitor
Hold out a comparison group so you can measure the margin the new assignment adds. H2O MLOps deploys the scoring pipeline, tracks model drift and keeps an audit trail.
Results: Banco do Brasil reports 4x contribution margin per client
Banco do Brasil reports 4x contribution margin per client with H2O (FEBRABAN TECH 2025). This is one bank's reported result. Your margin depends on your client base and your coverage model, which is why step 5 measures the change against a comparison group first.