2019-2020 · MARKETING
Propensity & cross-sell.
Propensity models for taking out insurance or a loan, plus a cross-selling recommendation engine, to select which customers each campaign should reach.
CASE STUDY · BANCO GALICIA · 2016-2020
Models that choose the right customer and automate credit decisions.
Four years and nine months at Banco Galicia, first in Credit Risk and then in Marketing & BI. Propensity, segmentation and real-time recommendation models, plus credit automation, that sold more with fewer calls and cut qualification times.
// THE CHALLENGE
Every contact has a cost. The value is in choosing who to contact.
In retail banking, outbound calls and email campaigns sell insurance, loans and cross-sell products. Their return depends on reaching the customers most likely to respond, and on getting each campaign out quickly.
In credit risk, the same idea applies to time: the faster a loan or a credit-card portfolio can be qualified, the faster the business can act on it.
// THE WORK
One question behind every model: is this contact worth making?
Much of the job was acting as a business translator: taking a business pain, proposing an MVP, and putting supervised and unsupervised models into production for the marketing and credit teams.
2020 · MARKETING
Customer Segmentation 2.0.
RFM and unsupervised models to find new groups of customers and new universes to target, feeding communication strategies and email campaigns.
2018-2019 · MARKETING
Real-time recommendation.
An integrated real-time recommendation system on Oracle, built to shorten the time it took for marketing campaigns to become available.
2019 · MARKETING
ReMarketing chatbot.
Text analytics and NLP (trigrams and graphs) to recognize patterns and detect, in real time, when a customer wants to act: take out a loan or unsubscribe a product.
2016-2018 · CREDIT RISK
Lending engine & card portfolio.
A loan qualification and granting system, and the automation of the credit-card portfolio qualification for individuals, replacing manual processes.
// WHAT CHANGED
Fewer calls, more sales, weeks turned into days.
−30%call costsPROPENSITY + CROSS-SELL · 2019-2020
+15%sales, in less timePROPENSITY + CROSS-SELL · 2019-2020
−70%time to get campaigns availableREAL-TIME RECOMMENDER · 2018-2019
+30%CTRREAL-TIME RECOMMENDER · 2018-2019
−60%loan qualification timeLENDING ENGINE · CREDIT RISK
15 → 2 dayscredit-card portfolio qualificationAUTOMATION · CREDIT RISK
Customer Segmentation 2.0 improved customer selection by 30%, and email campaigns gained 3% in CTR and 2% in conversion rate.
STACK
PythonSQLscikit-learnSPSSOracleRFMSupervised & unsupervised modelsNLP · trigrams + graphs
// CONTACT
Working on a similar problem?
Predictive modeling, customer analytics, segmentation and NLP, for banking, insurance and adjacent industries.