Mastercard Advisors · May 2026 – Jul 2026
Finding ₹500 Cr in a 5.2M-cardholder portfolio
Many cardholders barely use their cards, and every unused card is lost revenue for the bank. For India's #1 card issuer, I split 5.2M credit cardholders into 7 groups, worked out which groups are worth a campaign, and ranked 2.6M debit cardholders by how likely they are to start using their cards. The plan is worth about ₹500 Cr in incremental value.
- Company
- Mastercard AdvisorsMastercard's consulting and analytics arm
- My role
- Associate Consultant Intern, Performance Analytics
- When
- May 2026 – Jul 2026
- Tools
- SQL · PySpark · Hadoop · Impala · Customer Segmentation · ROI Modelling
- 5.2M
- 7
- SourceGo/No-Go ROI framework
- Sourcevs −0.72% baseline forecast
The client stays anonymous, and no client data appears on this page.
The problem
The bank had 5.2M credit cardholders, and 2.6M debit cardholders who only used their cards at ATMs, never to pay. An unused card earns the bank nothing.
Activation campaigns cost money, though. Offers, rewards and communication all add up, so targeting everyone wastes budget on people who won't respond, or who don't earn enough to pay back the offer.
What I did
- Segmented the credit portfolio. I grouped cardholders by activity state (how they use the card today) and card tier into 7 segments.
- Chose where to focus. I proposed an activation set of 5 of those segments, the ones worth a campaign.
- Rebuilt the ROI model. I modelled what each segment earns the bank through revolve (interest on balances) and interchange (fees on each payment), and turned it into a Go/No-Go framework the client can apply segment by segment.
- Ranked the debit cardholders. For the 2.6M ATM-only debit cardholders, I built propensity tiers: a High tier of ~280K and a Medium tier of ~2.25M people, by how likely they are to start paying with their card.
- Checked the offers. I modelled merchant-level offer economics, so each offer is judged on what it returns, not just how attractive it looks.
Along the way I analysed customer lifecycles, transactions and past campaign performance. Everything ran on SQL and PySpark over Hadoop/Impala.
What I'd do next
Run the activation set as a controlled test, with some cardholders held back as a comparison group, so the projected uplift becomes a measured one.
Key terms (7): new to the jargon?
- Activation
- Getting a cardholder to actually start (or resume) using their card.
- Segmentation
- Splitting customers into groups that behave similarly, so each group gets the right offer.
- Revolve
- Interest the bank earns when a customer carries a balance on their credit card from month to month.
- Interchange
- A small fee the bank earns every time a card is used to pay.
- Present value (PV)
- Future earnings expressed in today's money, so options can be compared fairly.
- Propensity
- How likely someone is to do something, here to start using their debit card for payments.
- Go/No-Go framework
- A simple rule for each segment, telling the client whether a campaign pays back (Go) or not (No-Go).
What's next