Polaris Microfinance Bank: three revenue figures and a board that wants one
Polaris Microfinance Bank serves 900,000 customers through 42 branches. The board's monthly pack has shown 'revenue' as ₦4.82bn, ₦5.07bn and ₦4.65bn in three consecutive months, produced by finance, operations and the BI team respectively. Investigation finds 14 documented definitions of 'active customer'; 22% of customer records duplicated across the core banking system, the mobile wallet and the loan book; no named data owners for customer or product data; the BI team rebuilds 60 Excel-based extracts manually each month; one analyst is the only person who understands the loan arrears query; and regulatory returns are prepared from a separate spreadsheet that reconciles to none of the three figures. The board wants 'one version of the truth' before a capital raise. The CIO proposes a ₦600m enterprise data warehouse replacement; the CFO wants the definitions fixed first.
1. Diagnose the root causes of the conflicting revenue figures and explain why a new warehouse alone would not resolve them.
Read the worked answer
The three figures are a symptom, not the disease
Trace each figure to its origin:
- Definitional conflict: 14 definitions of 'active customer' mean the population being measured differs before any arithmetic happens. Revenue per active customer, product mix and accruals will all diverge accordingly.
- Entity resolution failure: 22% duplication across core banking, wallet and loan book means the same customer is counted more than once and revenue may be attributed to different records in different systems.
- No ownership: nobody is accountable for the definition, the quality or the reconciliation. Each team has built a locally rational answer.
- Fragile process: 60 manual Excel extracts create opportunity for error and make results unreproducible month to month.
- Key-person dependency: the arrears query depends on one analyst; when they leave, the number changes or disappears.
- Broken control: regulatory returns reconcile to none of the figures, which means the external reporting process has its own undocumented logic.
Why a warehouse will not fix it
A warehouse moves data; it does not decide what 'revenue' means, who owns the customer record, or which duplicate survives. Re-platforming now would industrialise the ambiguity - producing the same three answers faster and at greater cost. Architecture is necessary but it comes after definition and ownership, otherwise the new platform simply encodes the old disputes and the business will not trust it either.
2. Recommend a data governance operating model, specifying roles, decision rights and the first three artefacts to produce.
Read the worked answer
Operating model
- Board/executive data council: chaired by the CFO or COO (not the CIO, to keep it a business accountability), meeting monthly. Decision rights: approves enterprise metric definitions, arbitrates cross-domain disputes, approves data quality thresholds and prioritisation.
- Data owners: accountable executives for each domain - customer (retail director), product (product head), finance (financial controller), risk (CRO). They sign off definitions and accept quality targets for their domain.
- Data stewards: named operational staff in each domain who maintain definitions, resolve issues and manage reference data day to day. This is a formal role in objectives, not a volunteer activity.
- Data management office: a small central team (3–5 people) providing method, tooling, quality monitoring and secretariat. It enables; it does not own the data.
- BI/analytics: builds and operates reporting against approved definitions; has no authority to redefine metrics.
Decision rights to make explicit
Who approves a new metric; who changes a definition and with what notice; who adjudicates when two domains disagree; who authorises access to sensitive data; who accepts residual data quality risk.
First three artefacts
- A metric dictionary for the top 20 board metrics, starting with revenue and active customer - formula, inclusions, exclusions, source system, owner, approver, effective date.
- A data domain ownership matrix (RACI) naming owners and stewards for customer, product, finance, risk, employee and reference data.
- A data quality issue log with severity, business impact, owner and target date, reported monthly to the council.
These are inexpensive and produce visible progress within 60 days.
3. Evaluate the CIO and CFO positions and recommend a sequenced 12-month plan with measurable success criteria.
Read the worked answer
Both are partly right, and the sequencing matters
The CFO is correct on priority: without approved definitions and ownership, a warehouse cannot produce a trusted number. The CIO is correct that the current extract-and-spreadsheet architecture is unsustainable and will re-create the problem at scale. The error is treating these as alternatives. Definitions without a platform remain manually enforced; a platform without definitions simply automates disagreement.
Sequenced plan
Months 0–3 - define and reconcile
- Approve the revenue and active-customer definitions at the data council; publish with an effective date.
- Produce one reconciled revenue number for the board, with a documented bridge explaining each prior variance.
- Reconcile regulatory returns to the same source. Stop issuing three figures immediately, even if the first reconciled figure is imperfect.
- Appoint owners and stewards.
Months 3–6 - fix the customer record
- Customer master data initiative: matching rules, survivorship rules, deduplication of the 22%.
- Replace the highest-risk manual extracts with controlled, versioned queries. Document the arrears query and cross-train at least two analysts.
Months 6–12 - platform
- Business case for the warehouse, scoped to approved definitions and prioritised reporting, delivered incrementally rather than as a big bang. Fund from decommissioned manual effort where possible.
Success criteria
One board revenue figure with zero unexplained variance; duplicate customer rate below 2%; at least 80% of board metrics automated from governed sources; regulatory returns reconciled monthly with sign-off; median data quality issue closure within agreed SLA; and no single-person dependency on any board metric.