- H2O Driverless AI builds propensity and recommendation models from your transaction history, product holdings, and customer profiles, with a reason code on every score.
- A propensity score for each customer and product tells relationship managers and digital channels who to contact, with what, and why.
- Scores feed the campaign, CRM, and channel tools you already run; H2O Driverless AI is the scoring layer underneath them.
- An agent can turn a customer's transaction patterns into a tailored, plain-language suggestion for a relationship manager to review.
The same offer goes to everyone
Your customers already tell you what they need through how they bank. Salary deposits, card spend, a growing savings balance, or a new mortgage enquiry all signal the next product, but most offers are still built from broad segments.
Broad offers waste contact budget and wear out customers. The customer who was ready for a credit card gets a savings promotion, and the one who just refinanced gets a loan offer.
How AI product recommendations work
Your analytics team builds explainable models in H2O Driverless AI on the customer data you already hold. Every customer gets a ranked list of products with a score and a reason.
Step 1: join the customer view
Start with transaction history, product holdings, balances, channel activity, and past campaign responses. The model learns from what customers actually did after earlier offers, not from assumptions about segments.
Step 2: learn which products go together
A recommendation model learns the relationships between banking products from historic transactions: which products are held together, which one usually comes next, and which customers look like those who already took it up.
Step 3: score propensity to buy
H2O Driverless AI trains classification models that estimate how likely each customer is to take up each product. Automatic feature engineering builds behavior features such as spend trends, balance changes, and life-event signals.
Step 4: explain and deliver
Each score carries reason codes, so a relationship manager can see why a product was suggested before calling. Scores flow to your CRM, campaign tool, or mobile app.
Propensity to buy
Propensity to buy is the likelihood that a given customer takes up a given product within a given time. It is the same model family as the recommendations above, scored per customer and per product. Relationship managers use the scores to prioritize their book, and marketing uses them to decide who receives an offer at all. Scores can refresh as new transactions arrive, so the list follows the customer rather than last quarter's segment.
Turning patterns into a suggestion a banker can use
An agent in Enterprise h2oGPTe can take a customer's transaction patterns and turn them into a tailored, plain-language suggestion for a relationship manager to review before it reaches the customer. This keeps a person in the loop on every recommendation that goes out.