Card Activation Prediction
“Which new cardholders will activate within 30 days?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
Which new cardholders will activate within 30 days?
25-35% of newly issued credit cards are never activated (Mercator Advisory Group). Each unactivated card represents $500-$800 in lost annual revenue from interchange, interest, and fees. For a large issuer with 5M new cards per year, that is $625M-$1.4B in unrealized revenue. The activation window is narrow: if a cardholder does not activate within 30-45 days, the probability drops to under 10%. Most issuers send generic reminder emails to all new cardholders, but conversion rates on these campaigns are just 3-5%.
Quick answer
The most effective card activation models connect application data, existing banking relationships, digital engagement signals, and card delivery status to predict which new cardholders will activate within 30 days. Relational ML captures the difference between a cardholder with 3 existing products and high app engagement vs a cold applicant from an online ad. This lets marketing teams target the 30% at highest dormancy risk with incentivized offers instead of sending generic reminders to everyone.
Approaches compared
4 ways to solve this problem
1. Uniform reminder campaigns
Send the same activation reminder email and push notification to all new cardholders on a fixed schedule (day 3, day 7, day 14).
Best for
Simple to execute. No model needed. Captures the low-hanging fruit: cardholders who intended to activate but forgot.
Watch out for
3-5% conversion on these campaigns. The 70% who would have activated anyway get unnecessary reminders, and the 25-35% who will never activate get the same generic nudge that does not address their actual barrier.
2. Application-data scoring
Score new cardholders based on application features: channel (branch vs online), existing relationship (products held), credit score, and card type.
Best for
Quick to build. Captures the basic relationship: branch applicants with existing products activate at higher rates than cold online applicants.
Watch out for
Static at approval time. Does not update as behavioral signals arrive (app logins, email opens, card delivery status). A cardholder who applied online but is actively logging into the app may be more likely to activate than the model predicts.
3. XGBoost with behavioral features
Add post-approval behavioral signals (app logins, email opens, push notification opt-in, card delivery status) to application data and train a classification model.
Best for
Meaningful lift over application-only scoring. Captures the engagement trajectory in the first 7-14 days after approval.
Watch out for
Features are manually engineered from separate systems (app analytics, email platform, fulfillment tracker). Keeping these features fresh and joined correctly is an ongoing data-engineering burden.
4. KumoRFM (relational graph ML)
Connect new cardholders to application data, existing account activity, digital engagement events, and card delivery status. The GNN learns activation patterns from the full relational context, updating daily as new engagement data arrives.
Best for
Captures the full story: zero app logins + no email opens + online-ad channel + no prior relationship = 18% activation probability (critical risk). Branch referral + 5 app logins + 2 emails opened + 3 existing products = 89% activation (no intervention needed). Marketing spends budget only where it changes the outcome.
Watch out for
Requires real-time digital engagement data feeds (app events, email tracking). Cardholders with no prior relationship have less relational signal to work with.
Key metric: 25-35% of new credit cards are never activated (Mercator Advisory Group), representing $625M-$1.4B in unrealized annual revenue for a top-10 issuer.
Why relational data changes the answer
Card activation depends on signals scattered across application data, existing banking relationships, digital engagement, and fulfillment status. A flat model using only application features cannot see that the cardholder has been logging into the app 5 times since approval, opened both welcome emails, and has 42 transactions on their existing card. These post-approval behavioral signals are the strongest predictors of activation, and they live in different tables from the application data.
Relational models connect all of these signals and update daily. The result: clear separation between cardholders who will activate on their own (89% probability, high engagement, existing relationship) and those heading for dormancy (18% probability, zero engagement, no prior relationship). This separation lets marketing teams allocate activation incentives ($100 bonus offers, personal calls) only to the high-risk cohort where intervention changes the outcome. For a top-10 issuer with 5M new cards per year, that targeted approach recovers $200-400M in annual unrealized revenue.
Sending the same activation reminder to every new cardholder is like a gym sending the same 'come back!' email to every January signup. The person who has been coming 4 times a week does not need it. The person who signed up, never visited, and has not opened a single email needs a different conversation entirely. Relational ML knows who is showing up and who is ghosting.
How KumoRFM solves this
Relational intelligence built for banking and financial data
Kumo connects new cardholders to their application data, existing product relationships, transaction history on other accounts, digital engagement signals, and demographic patterns. The model identifies that Cardholder CH-5501 applied through a branch referral, has 3 existing products, high mobile-app engagement, but has not received the physical card yet (shipping delay). Meanwhile, CH-5520 applied through a generic online ad, has no prior relationship, and has not logged into the app. Kumo scores activation probability so marketing teams can target high-risk-of-dormancy cardholders with incentivized activation offers.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
NEW_CARDHOLDERS
| cardholder_id | card_type | application_channel | approval_date | existing_products |
|---|---|---|---|---|
| CH-5501 | Rewards Platinum | Branch Referral | 2025-09-01 | 3 |
| CH-5520 | Cash Back | Online Ad | 2025-09-03 | 0 |
| CH-5534 | Travel Elite | Pre-approved Mail | 2025-09-05 | 2 |
CARD_STATUS
| cardholder_id | card_shipped | card_delivered | activated | days_since_approval |
|---|---|---|---|---|
| CH-5501 | True | True | False | 14 |
| CH-5520 | True | True | False | 12 |
| CH-5534 | True | False | False | 10 |
DIGITAL_ENGAGEMENT
| cardholder_id | app_logins_7d | email_opens | push_enabled |
|---|---|---|---|
| CH-5501 | 5 | 2 of 2 | True |
| CH-5520 | 0 | 0 of 2 | False |
| CH-5534 | 3 | 1 of 2 | True |
EXISTING_ACCOUNT_ACTIVITY
| cardholder_id | other_card_txns_30d | checking_balance | direct_deposit |
|---|---|---|---|
| CH-5501 | 42 | $8,200 | Active |
| CH-5520 | N/A | N/A | N/A |
| CH-5534 | 28 | $15,400 | Active |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(CARD_STATUS.ACTIVATED = 'True', 0, 30, days) FOR EACH NEW_CARDHOLDERS.CARDHOLDER_ID WHERE CARD_STATUS.ACTIVATED = 'False'
Prediction output
Every entity gets a score, updated continuously
| CARDHOLDER_ID | CARD_TYPE | ACTIVATION_PROB | RISK_OF_DORMANCY | RECOMMENDED_ACTION |
|---|---|---|---|---|
| CH-5501 | Rewards Platinum | 0.89 | Low | Standard Welcome |
| CH-5534 | Travel Elite | 0.64 | Medium | Bonus Offer |
| CH-5520 | Cash Back | 0.18 | Critical | Call + $100 Bonus |
Understand why
Every prediction includes feature attributions — no black boxes
Cardholder CH-5520 (Cash Back card)
Predicted: 18% activation probability (critical dormancy risk)
Top contributing features
No existing banking relationship
0 products
30% attribution
Zero app logins since approval
0 in 12d
26% attribution
Email engagement absent
0 of 2 opened
20% attribution
Application channel (low-intent)
Online Ad
14% attribution
Push notifications disabled
False
10% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about card activation prediction
What percentage of credit cards are never activated?
25-35% of newly issued credit cards are never activated (Mercator Advisory Group). Each unactivated card represents $500-$800 in lost annual revenue from interchange, interest, and fees. The activation window is narrow: if a cardholder does not activate within 30-45 days, the probability drops below 10%.
How do you increase credit card activation rates?
Targeted intervention on high-dormancy-risk cardholders is more effective than blanket reminders. Use a predictive model to identify the 30% at highest risk of never activating, then deploy incentivized offers (bonus points, cashback) and personal outreach to that specific cohort. Generic reminders convert at 3-5%. Targeted incentive offers convert at 15-20% among high-risk cardholders.
What data predicts card activation?
The strongest predictors are post-approval behavioral signals: app logins (zero vs multiple), email engagement (opens, clicks), push notification opt-in, and card delivery confirmation. Application-level features (channel, existing products, credit score) set the baseline, but behavioral signals in the first 7-14 days after approval are the most predictive of 30-day activation.
How long do you have to get a new cardholder to activate?
The critical window is 30-45 days. After that, activation probability drops below 10% and continues declining. The most effective activation campaigns start immediately after card delivery confirmation and escalate intensity through day 21. By day 30, unactivated cards should receive the highest-value incentive or personal outreach.
What is the revenue impact of card dormancy?
For a large issuer with 5M new cards per year, 25-35% dormancy means 1.25M-1.75M unactivated cards. At $500-$800 per card in annual revenue (interchange, interest, fees), that is $625M-$1.4B in unrealized revenue. Even recovering 20% of dormant cards through targeted intervention generates $125-280M annually.
Bottom line: Target the 30% of new cardholders at highest dormancy risk with personalized activation offers, recovering $200-400M in annual unrealized revenue for a top-10 issuer.
Related use cases
Explore more financial services use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
