Difference in Difference

Welcome to my project!

---

✅ 1. Reframe Your Outcome Metric as “Share of Production”

Instead of showing raw volume (which is down everywhere), show the share of Personalized Cards relative to all cards issued. For example:

“Branches that received promo boxes saw a +X% increase in the share of cards that were personalized, while non-promo branches declined or remained flat.”

Metric:

Personalized Share = Personalized Cards / (Personalized + Standard Cards)

Do this for Time Period 1 and 2, and compute the delta.

Run a t-test or Wilcoxon rank-sum on delta(Personalized Share) between promo vs non-promo branches.

---

✅ 2. Construct a Difference-in-Differences (DiD) Analysis

You already have a natural experiment with a treatment group (branches that received promos) and control group (those that didn’t).

Model:

Outcome: Personalized card production (or share) Predictors: Promo Branch (T/F), Time Period (Pre/Post), and their interaction

The interaction term (Promo Branch * Post Period) gives you the treatment effect, i.e., what we attribute to the promo boxes.

Even if the effect is modest, if the coefficient is positive and significant, you now have causal-ish language like:

“Branches receiving promotional items outperformed the regional trend by X% in personalized card growth, adjusting for the regional decline.”

---

✅ 3. Use Indexing to Normalize for Downtrend

Create a baseline index (set each branch’s Time Period 1 = 100), then express Time Period 2 values as indexed change. This visually makes it clear who “beat the market” even if raw numbers dropped.

Plot: Boxplot or slope graph of % change for promo vs. non-promo branches.

Narrative: “Promo branches retained X% more of their original personalized volume than non-promo branches during a regional slowdown.”

---

✅ 4. Normalize by Foot Traffic or Account Openings (if available)

If you can get branch-level deposit account openings or foot traffic:

Compute personalized cards per DDA opened.

If personalized card density per customer is higher in promo branches, you now have a metric that suggests increased engagement.

---

✅ 5. Highlight High-Performing Teams or Branches

Even if aggregate numbers are soft, highlighting winners helps change the tone:

“Branches with [Team X] boxes saw +12% increase in personalized card share, suggesting the right sports affinity boosts impact.”

Consider an interactive visual that shows team-by-team lift.

Maybe one team outperformed heavily—lean into that.

---

✅ 6. Create a Composite “Engagement Uplift Score”

Aggregate all the elements into a simple uplift score:

Engagement Score = weighted % change in Personalized Cards (new + reissue) + change in personalized share

Then rank branches and show that promo branches dominated the top deciles.

A Sankey diagram or decile bar chart can tell a clean visual story.

---

✅ 7. Language Framing for Executive Presentation

Rather than focusing on decline, focus on resilience and share of engagement:

“In a contracting market, promo branches preserved personalized engagement significantly better.”

“Promo boxes contributed to a +X% lift in personalized card preference share, despite declining total volume.”

---

If You’re Open to Additional Data Pulls:

Add Standard Card Volumes to explicitly show share shifts.

Bring in team-specific timelines if possible—see if timing overlaps with any known sports calendar events (e.g., playoff runs, merchandise campaigns).

Incorporate branch size (e.g., average monthly openings) to contextualize volume changes.

---

Would you like help structuring the DiD analysis in PySpark, R, or Excel? Or should I help you write the storyline for your PowerPoint deck?

That’s excellent progress—and a much stronger story is starting to emerge. The fact that:

✅ Branches with promotional boxes saw an increase in personalized card production,

✅ While non-promo branches in the same region declined across the board, and

✅ This happened in the face of overall declines in standard cards,

gives you a clearer positive treatment signal to lean into.

Let’s now tighten the analysis and positioning to get the most compelling narrative from this.

---

🔍 RECOMMENDED NEXT STEPS FOR STUDY DESIGN & STORY

1. Anchor to the Difference-in-Differences (DiD) Framework

Now that you’ve included all branches, you have:

Treatment Group: Branches with promo boxes

Control Group: All other branches in the same region(s)

Pre vs Post: Your two time periods (T1 vs T2)

Use this to quantify the impact more cleanly:

Treatment Effect = (Personalized_T2_Treatment - Personalized_T1_Treatment) - (Personalized_T2_Control - Personalized_T1_Control)

You can do this:

At the aggregate level (summed card volume)

At the branch level (mean cards per branch)

Narrative Framing:

“Branches receiving promotional boxes grew personalized card production by +X%, while peer branches declined by –Y%, yielding a net engagement lift of Z% attributable to the promotion.”

---

2. Introduce a Normalized Metric: “Personalized Card Penetration”

Use:

Personalized Penetration = Personalized Cards / (Standard + Personalized Cards)

You can now:

Show this increased more or decreased less in promo branches

Frame it as a shift in consumer preference toward personalized cards

Narrative:

“Among card selections, the share of clients choosing personalization increased by X% in promo branches, suggesting a meaningful shift in customer preference.”

---

3. Use “Branch-Level Lift” Visualizations

Use a scatter or slope graph to show the change in personalized cards per branch (T1 to T2), colored by team and promo status. This lets you:

Highlight branches that grew despite the downtrend

Spotlight top-performing teams (e.g., maybe Team A’s branches all grew)

Also consider a decile ranking:

Group all branches by % change in personalized card production

Count how many promo branches land in top deciles vs control

Narrative:

“Promo branches were 3× more likely to land in the top 20% of performers by personalized card growth.”

---

4. Highlight “Promo as an Offset to Market Decline”

Quantify the decline in standard cards vs the rise in personalized cards in promo branches. A good visual: a waterfall chart or a bar chart comparing:

Branch Type Standard Δ Personalized Δ Net Δ

Promo –X% +Y% … Control –Z% –W% …

Narrative:

“While standard card demand declined sharply across all branches, promo branches not only offset the loss—they reversed the trend in personalization.”

---

5. Team-Level Success Framing

Even if not every team performs equally, cherry-pick the winners and show potential for future optimization:

“Branches with [Team X] saw the strongest increase in personalized card volume: +Y%”

“This suggests deeper affinity strategies could further improve results”

A team leaderboard chart works well here.

---

6. Frame Results as “Proof of Engagement Potential”

You don’t need to overclaim. Instead, position the results as:

A signal that personalization resonates

An engagement nudge that moved the needle despite adverse market conditions

Evidence that future promotions could be targeted and scaled

Example framing:

“These early results suggest that localized promotional tie-ins can enhance engagement with optional card features like personalization—even in a declining acquisition environment.”

---

✳️ Summary of Actionable Metrics to Calculate:

Metric Use

Personalized Δ T2 – T1 for personalized cards Personalized Penetration Δ Share change of personalized vs all cards Difference-in-Differences Estimate Causal estimate of promo box impact Branch-Level Lift Score Δ in personalized cards per branch Team-Level Average Δ To rank teams’ effectiveness Promo vs Control Proportion in Top Deciles Evidence of top performance clustering

---

🟦 Slide 1: Executive Summary (Title Slide)

Title: “Impact of Branch-Level Sports Promo Boxes on Personalized Debit Card Engagement”

Subtitle: An analysis of promotional lift across [Region Name] branches | Q4 2024–Q1 2025

---

🟦 Slide 2: Business Context

Slide Title: Why We Promoted Card Personalization with Local Sports Tie-Ins

Bullet Points:

Goal: Increase engagement with debit card personalization feature

Tactic: Distribute themed promotional boxes tied to 5 sports teams across select branches

Period: Q4 2024 – Q1 2025

Hypothesis: Themed promotions would increase client demand for card personalization

Success metric: Increase in new personalized card production

---

🟦 Slide 3: Study Design

Slide Title: How We Measured Promotional Effectiveness

Visual: A simplified treatment vs control schematic

Time Period 1 | Time Period 2

Promo Branches | Personalized Production ↑ Non-Promo Branches | Personalized Production ↓

Key Points:

Compared 2 time periods: Pre (Q4 2023/Q1 2024) vs Post (Q4 2024/Q1 2025)

Measured outcomes:

New personalized card production

Personalized card share of total cards

Change in standard card production

Segmented by:

Branch

Sports team promoted

Region

---

🟦 Slide 4: Overall Trends in the Market

Slide Title: Standard Card Production Declined Region-Wide

Chart Suggestion: Clustered bar chart: Standard vs Personalized cards by branch type (Promo vs Non-Promo), over Time 1 and Time 2

Insight Callout Box:

“Promo and non-promo branches both saw declines in standard card issuance — but personalized cards grew at promo branches.”

---

🟦 Slide 5: Key Metric #1 — Personalized Card Growth

Slide Title: Promo Branches Outperformed Peers on Personalized Card Growth

Visual Option A: Bar chart of % change in personalized card volume: Promo vs Non-Promo branches Visual Option B: Boxplot or violin plot showing branch-level % change (median highlighted)

Key Callout:

“Promo branches saw an average +X% increase in personalized card volume vs a –Y% decrease at non-promo branches.”

---

🟦 Slide 6: Key Metric #2 — Shift in Personalization Preference

Slide Title: Promo Branches Saw a Shift Toward Personalization

Visual: Slope chart or side-by-side bar chart showing Personalized Share of Total Cards (T1 → T2) for Promo vs Non-Promo

Key Callout:

“Personalized cards grew from X% to Y% of all cards in promo branches — a meaningful shift in preference.”

---

🟦 Slide 7: Team-Level Performance

Slide Title: Some Teams Drove Stronger Results Than Others

Visual: Bar chart of average % change in personalized card production per branch by team

Key Takeaway:

“Branches promoting [Team A] saw the highest lift: +Z% growth in personalized cards per branch.”

---

🟦 Slide 8: Engagement Resilience Index (Optional)

Slide Title: Promo Branches Were More Resilient in a Declining Market

Visual: Waterfall chart showing:

Drop in standard cards

Gain in personalized cards

Net branch-level engagement change

Key Callout:

“Despite overall declines, personalized engagement increased at promo branches.”

---

🟦 Slide 9: Strategic Implications

Slide Title: What These Results Tell Us

Bullet Points:

Promotional boxes drove a measurable increase in personalized debit card selection

Personalization preferences grew even as total card demand fell

Some teams resonate more with clients — opportunity for targeting

Personalization can serve as a lever to strengthen client affinity in branch channel

---

🟦 Slide 10: Next Steps & Recommendations

Slide Title: Scale What Worked — And Test What’s Next

Bullet Points:

Expand promotional tie-ins with top-performing teams

Introduce similar kits in declining branches to test turnaround effect

Explore digital variants of the promotion for online personalization

A/B test messaging around personalization preference drivers (e.g. team loyalty vs individuality)

Here is a comprehensive data story framework for an executive-level slide deck on assessing deposit attrition risk tied to the sunset of a specific debit card product. This version includes analysis at both the customer and household level, integrates your hypotheses, and provides data wrangling instructions for each slide.

---

📘 Executive Slide Deck Framework:

“Assessing Deposit Attrition Risk from [Product X] Debit Card Sunset”

---

🟦 Slide 1: Executive Summary

Title: 📌 “[Product X] Sunset: Who’s at Risk & How Much Is at Stake?”

Purpose: Summarize key findings and highlight urgency. Meant for senior leaders with limited time.

Contents:

Total impacted: [# of customers / households]

% product-dependent (highest risk group)

Estimated deposit dollars at high risk

Primary recommendation (e.g., targeted outreach, preemptive migration)

Data Wrangling Tips:

Filter all customers and households with active [Product X] debit cards

Aggregate total deposits, flag product-dependent customers

Group by risk segments (defined in later slides)

---

🟦 Slide 2: Context & Goals

Title: 🎯 Why This Matters: Business Context

Purpose: Frame the situation, link to business strategy (e.g., product rationalization, cost optimization), and introduce your objectives.

Contents:

We’re retiring [Product X] as part of modernization

Risk of attrition for product-dependent customers

Goals:

1. Profile impacted base

2. Segment by relationship depth

3. Quantify deposit balances at risk

4. Recommend retention strategy

Data Wrangling Tip: No data needed — narrative only.

---

🟦 Slide 3: Population Definition & Methodology

Title: 🔍 Who’s in Scope: Defining the Population

Purpose: Clarify who was included in the analysis.

Contents:

All active debit card holders with [Product X]

Both Consumer and Business

Household roll-up applied to measure full relationship

Two-level analysis:

Customer-level risk

Household-level risk and entrenchment

Data Wrangling Tips:

Extract all active [Product X] cards with valid links to checking accounts

Join to:

Customer master

Household roll-up table

Product ownership (deposit, loan, card, CD)

Digital and transaction data

Deduplicate at both customer and household level

---

🟦 Slide 4: Customer & Household Profile

Title: 👥 Who Holds This Product?

Purpose: Give executives a persona-style overview of impacted users.

Visuals:

Charts by:

Customer segment (mass market, affluent)

Generation (Gen Z, Millennial, etc.)

Consumer vs Business

Digital enrollment rate

Primacy rate (e.g., active DD or bill pay)

Data Wrangling Tips:

Use age + generation mapping

Use product flags to determine primacy (ACH > $500, bill pay, recurring debit)

Use household roll-ups to capture “entire relationship” engagement

---

🟦 Slide 5: Relationship Depth Segmentation

Title: 🏦 How Deep Are These Relationships?

Purpose: Segment customers by product dependence and relationship complexity.

Segmentation Framework:

Segment Name Criteria

Product-Dependent 1 Checking + [Product X] only, no other products Light Relationship + Savings or Online Banking, no credit or loan products Moderate Relationship + Credit Card or CD, but no lending or investment Deep Relationship 3+ products across deposit, card, loan, or wealth

Data Wrangling Tips:

Join product ownership data

Create flags for product types

Sum flags to assign depth tier

Create same flags at household level to determine Household Depth Tier

---

🟦 Slide 6: Engagement & Behavior

Title: 💳 How Are These Customers Engaging?

Purpose: Show transactional behavior and digital engagement for each segment.

Visuals:

Avg. monthly debit transactions (by segment)

% with online/mobile banking

Avg. login frequency (if available)

Data Wrangling Tips:

Join card transaction data (aggregate 3-month average)

Join digital banking logins table (if available)

Group by relationship depth and/or customer vs household level

---

🟦 Slide 7: Deposit Balances by Risk Segment

Title: 💰 What’s at Risk? Deposit Value by Relationship Tier

Purpose: Quantify how much money is exposed in high-risk segments.

Visuals:

Bar or waterfall chart: deposit $ by risk level

Include both customer and household roll-ups

Segment # Cust Est. Risk Avg Bal Total Deposits Household Risk %

Product-Dependent X,XXX High $Y,YYY $Z MM X%

Data Wrangling Tips:

Calculate avg and total checking/savings balances

Group by segment at both individual and household level

Estimate attrition risk multiplier (e.g., 60% of Product-Dependent assumed to leave)

---

🟦 Slide 8: Digital & Primacy Gaps

Title: 📉 Who’s Missing Key Engagement Anchors?

Purpose: Surface gaps in stickiness: these are retention opportunities.

Visuals:

Heatmap by segment:

% with Direct Deposit

% with Mobile Banking

% with active debit transactions

Data Wrangling Tips:

Use ACH/Payroll/Bill Pay flags for primacy

Use digital channel usage

Group by depth and household

---

🟦 Slide 9: Recommendations by Segment

Title: 🎯 Retention Strategy by Risk Tier

Purpose: Translate findings into action.

Table:

Segment Risk Strategy Example Offer

Product-Dependent High Proactive outreach & upgrade $X checking bonus Light Relationship Med-High Cross-sell CD/CC + incentives Digital bundle Deep Relationship Low Passive migration None needed

Data Wrangling Tips: None. Based on synthesis from analysis above.

---

🟦 Slide 10: Final Takeaways

Title: ✅ What You Need to Know

Bullet Points:

[Product X] customers represent $X billion in deposits

[Y]% are product-dependent with limited engagement

Attrition risk concentrated in [Z] customer/household profiles

Action now = $[amount] in deposit savings retained

---

📘 Executive Slide Deck Framework

Title: Assessing Deposit Attrition Risk from [Product X] Debit Card Sunset

---

🟦 Slide 1: Executive Summary

📌 Title: “[Product X] Sunset: Who’s at Risk & How Much Is at Stake?”

Purpose: High-level summary for senior leaders
Contents:

• Total impacted customers and households
• % in highest risk group (“product-dependent”)
• Estimated deposit dollars at risk
• Primary recommendation (e.g., targeted outreach)

🔧 Data Wrangling:

• Filter active [Product X] debit card holders
• Aggregate:

  • of customers and households

  • Total deposits
• Apply risk segmentation to compute totals at risk

---

🟦 Slide 2: Context & Goals

🎯 Title: Why This Matters: Business Context

Purpose: Set the stage for business urgency and analytical goals
Contents:

• Strategic reason for sunsetting the product
• Risks associated with disengagement
• Study objectives:
  • Customer & household profiling
  • Relationship depth segmentation
  • Deposits at risk quantification
  • Actionable recommendations

🔧 Data Wrangling: None (narrative context)

---

🟦 Slide 3: Population Definition & Methodology

🔍 Title: Who’s in Scope: Defining the Population

Purpose: Clearly define scope and data sources
Contents:

• Population = active [Product X] cardholders (Consumer + Business)
• Household roll-up applied
• Two levels of analysis: Customer & Household
• Source systems and analytical timeframe

🔧 Data Wrangling:

• Extract all active [Product X] debit cardholders
• Join to:
  • Customer master
  • Household roll-up
  • Product ownership
  • Digital banking & transactions

---

🟦 Slide 4: Customer & Household Profile

👥 Title: Who Holds This Product?

Purpose: Describe demographics and segment makeup
Contents:

• Generation (e.g., Gen Z, Millennial)
• Customer segment (e.g., Mass Market, Affluent)
• Business vs. Consumer mix
• Digital enrollment
• Primacy rates (DD, bill pay, etc.)

🔧 Data Wrangling:

• Join cardholders to:
  • Customer demographics
  • Product flags
  • Digital enrollment & activity tables
• Derive generation from DOB
• Calculate primacy using ACH, DD flags

---

🟦 Slide 5: Relationship Depth Segmentation

🏦 Title: How Deep Are These Relationships?

Purpose: Categorize customer/household product depth
Segments:

Segment	Criteria Example
Product-Dependent	Checking + [Product X] only
Light Relationship	+ Savings or Online Banking
Moderate Relationship	+ Credit Card or CD
Deep Relationship	3+ products across Deposit/Card/Loan/Wealth

🔧 Data Wrangling:

• Join to product ownership data
• Flag product types
• Count product categories owned
• Derive depth tier for both customer & household levels

---

🟦 Slide 6: Engagement & Behavior

💳 Title: How Are These Customers Engaging?

Purpose: Show usage patterns to infer risk/loyalty
Contents:

• Avg. monthly debit transactions
• % using digital (online/mobile)
• Avg. logins per month (if available)

🔧 Data Wrangling:

• Join to:
  • Card transaction tables
  • Digital login activity
• Calculate:
  • Monthly transaction averages
  • Enrollment flags & frequency by user

---

🟦 Slide 7: Deposit Balances by Risk Segment

💰 Title: What’s at Risk? Deposit Value by Relationship Tier

Purpose: Quantify deposits at risk across segments
Table Format Example:

Segment	# Cust	Risk Level	Avg Balance	Total Deposits	HH Risk %
Product-Dependent	X,XXX	High	$Y,YYY	$Z MM	XX%

🔧 Data Wrangling:

• Join deposit balance snapshot (checking/savings)
• Summarize per segment at both customer & household levels
• Estimate attrition multiplier by risk tier (e.g., 60% of Product-Dependent)

---

🟦 Slide 8: Digital & Primacy Gaps

📉 Title: Who’s Missing Key Engagement Anchors?

Purpose: Identify where stickiness is missing
Contents:

• % enrolled in mobile/online
• % with Direct Deposit or bill pay
• % with >5 debit txns/mo

🔧 Data Wrangling:

• Join to ACH, DD, bill pay flags
• Aggregate and bucket by relationship tier
• Produce % tables or heatmaps

---

🟦 Slide 9: Recommendations by Segment

🎯 Title: Retention Strategy by Risk Tier

Purpose: Translate insights into targeted actions
Example Table:

Segment	Risk	Action	Offer Type
Product-Dependent	High	Targeted upgrade campaign	$X incentive
Light Relationship	Med	Bundle + cross-sell savings	Rate promo offer
Deep Relationship	Low	Passive migration	No action needed

🔧 Data Wrangling: None — strategy based on above findings

---

🟦 Slide 10: Final Takeaways

✅ Title: What You Need to Know

Purpose: Wrap-up with key business-impact stats
Contents:

• [Product X] holders = $X billion in deposits
• [Y]% are Product-Dependent
• $Z MM in deposits estimated at risk
• Key next steps: targeting, A/B testing, messaging

🔧 Data Wrangling: Summary of previous analysis

---

🟦 Optional Slides (Appendix)

📎 Title: Field Definitions, Logic & Assumptions

Contents:

• Segmentation logic
• Product code map
• Digital/primacy flag definitions
• Attrition risk multipliers
• Known data caveats (e.g., business users without login info)

🔧 Data Wrangling: Collate and summarize technical reference logic