Spot the Silo: Auditing Utkarsh Small Finance Bank's Broken Data Flow
Objective: Given a description of Utkarsh Small Finance Bank's current disconnected marketing stack, identify every point where the composable-CDP principle (one warehouse, one source of truth) is being violated and rank the fixes by urgency.
You're the digital marketing analyst at Utkarsh Small Finance Bank, the Varanasi-founded small finance bank that listed on the NSE and BSE in July 2023. Your CMO just asked why the loan cross-sell campaign keeps targeting customers who closed their accounts three weeks ago.
Read the current-state stack description, sort every step into 'silo' (violates single-source-of-truth) or 'fine', then rank the fixes by urgency.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, and a shared audit doc every stakeholder can review without new tool access
The process
1 step
Step 01 of 01
The lesson's composable CDP splits data work into three layers: a collection tool (Segment) gathers events, a warehouse (BigQuery) becomes the single source of truth, and reverse ETL syncs it back out. A packaged, siloed tool that keeps its own private customer database breaks this because two teams end up looking at two different definitions of the same customer.
Utkarsh's current setup: the collections team exports a CSV of closed accounts from the core banking system every Monday morning and emails it to the marketing intern, who manually removes those customers from the email tool's suppression list. The mobile app team runs its own analytics tool with its own device-based user ID that doesn't match the core banking account number. Which of these is the highest-priority silo to fix first, and why?
Procedure
- List every system in the current stack (core banking, email tool, mobile analytics, ad accounts) as rows
- For each system, note what customer ID it uses and how often it's updated
- Flag any system whose customer ID doesn't match another system's ID as a 'silo, critical'
- Flag any manual CSV/email step as a 'silo, moderate' (works today, breaks under scale or staff turnover)
- Rank the flagged items by how much bad targeting they cause per week
SYSTEM AUDIT, Utkarsh Small Finance Bank System | Customer ID | Update cadence | Flag Core banking | account_no | real-time | source of truth Email tool | email address only | weekly, manual | SILO - critical (no shared ID, 3-week lag) Mobile app analytics | device-based UUID | real-time | SILO - critical (can't join to account_no at all) Google Ads | hashed email | manual upload | SILO - moderate (works, but stale) Priority fix: the mobile app UUID has no path back to account_no, worse than the email lag because it can never be joined, not even manually.
Healthy
Every system either shares one customer ID or has a documented, automatable path to it.
Unhealthy
A system whose ID can't be joined to any other system, even manually, as with the mobile app UUID here.
What this means
A weekly manual lag is a process problem you can patch; an unjoinable ID is an architecture problem that blocks every future fix until it's solved.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| Suppression list is always ~3 weeks stale | Replace the manual CSV export with an automated daily sync once a warehouse exists | dev ticket |
| Mobile app events can't be tied to a bank customer | Add a login-linked customer ID to the app's analytics SDK before any personalization work | dev ticket |
Final deliverable
A one-page current-state data flow map with every silo flagged and ranked by fix urgency.
See a reference example
FIVE-STAR BUSINESS FINANCE, data flow audit (excerpt) System | Customer ID | Flag LOS (loan origination)| loan_account_id | source of truth Collections call tool | phone number only | SILO - critical, can't join to loan_account_id when a customer holds 2 loans SMS reminder tool | phone number only | SILO - critical, same join problem as above Priority fix: the collections tool first, a phone-only ID means a repayment reminder can land against the wrong loan.
Success criteria
You're done when you can:
- Correctly flags all silo systems (2+ critical, 1+ moderate)
- Ranks the mobile app UUID issue above the manual CSV lag with a stated architectural reason
- Names at least one concrete step to fix each flagged system