Financial services analytics: risk scoring, compliance reporting and customer intelligence

Large banks have risk, compliance and analytics departments. A small lender, a fintech with forty people, a credit union's finance team or an independent advisory firm has the same questions and one or two people to answer them. This guide covers the metrics that matter in three areas (credit risk, compliance operations and customer value) with the formulas behind them and a sensible order to build them in.

It is about analytics, not regulation. What your regulator requires is a question for your compliance officer and counsel.

Credit risk: the numbers that move first

Delinquency rates

Delinquency rate = balances (or accounts) past due by N days ÷ total balances (or accounts)

Track 30, 60 and 90 days past due separately, by product and by origination month. A rising 30-day rate is the earliest visible signal of future losses.

Roll rates

Roll rates measure how many accounts move from one bucket to the next in a month: current to 30 days, 30 to 60, 60 to 90. They predict losses better than a single delinquency snapshot because they show momentum.

Example: 2,000 accounts were current last month and 60 of them are 30 days late this month. The current-to-30 roll rate is 60 ÷ 2,000 = 3%. If that was 2% a quarter ago, something changed in underwriting, in the customer base or in the economy.

Expected loss

Expected loss = probability of default (PD) × loss given default (LGD) × exposure at default (EAD)

A worked example with round numbers: a portfolio segment with $5,000,000 outstanding, an estimated 4% probability of default over the next year and a 60% loss after recoveries:

$5,000,000 × 4% × 60% = $120,000 expected loss

Break it down by segment and the table tells you where provisions and pricing need attention.

Concentration

A simple measure is the share of total exposure held by the largest 10 borrowers, or by the largest industry or region. If one industry holds 35% of the book, a downturn there is your downturn.

Model monitoring

If you use a scoring model, track how well it still separates good and bad borrowers over time. The Gini coefficient and the Kolmogorov-Smirnov (KS) statistic are the usual measures. A steady decline means the population or the economy has moved away from the data the model was built on.

Compliance operations: measuring the process

Analytics does not make a firm compliant. It shows whether the compliance process is working and where it is stuck. These are the operational numbers most teams track.

Transaction monitoring alerts

• Alert volume per week, by rule.
• False positive rate = alerts closed with no finding ÷ alerts closed. Transaction monitoring systems commonly run very high false positive rates. A rule that closes almost everything with no finding is a candidate for tuning; a sudden drop can mean a rule changed without anyone noticing.
• Alert ageing: how many alerts are older than your internal target, and which analyst queues they sit in.
• Escalation rate: share of alerts that become cases, and of cases that lead to a report.

Customer onboarding (KYC)

• Onboarding cycle time from application to approval, split by customer type. Median is more useful than average here, because a few stuck files distort the mean.
• First-time-right rate: applications that needed no follow-up request for documents.
• Enhanced due diligence volume and average processing time.
• Periodic reviews due against completed, by month.

Findings and exceptions

• Open audit or examination findings by age. An old open finding is worse than a new one.
• Active policy exceptions, with owner and expiry date.
• Training completion by team.

Customer intelligence

Customer value

In financial services, value has to include risk and cost, not just revenue:

Customer value per year = interest income + fee income − cost to serve − expected credit loss

Example: a small-business customer brings $1,800 in interest and $300 in fees, costs $400 a year to serve and has an expected loss of $500. Annual value is $1,800 + $300 − $400 − $500 = $1,200. A customer with twice the balance and three times the expected loss may be worth less.

Products per customer

Average products held per customer is one of the most watched retention indicators in retail banking. Customers with a single product tend to leave more easily than those with several. Track it by segment and by tenure.

Attrition early warnings

Account closures are often preceded by quieter signals: falling balances, fewer transactions, fewer logins, a direct deposit that moves elsewhere. Build a list of customers showing two or more of these signals and review it monthly. The customer segmentation guide covers how to score behavior like this.

Channel mix

Share of transactions and service contacts by channel (app, web, phone, branch), by segment. It tells you where to invest and who will be affected if you close a channel.

Market and economic context

Many firms also watch outside data next to their own: interest rates, inflation, unemployment, exchange rates, equity and commodity prices. Public sources such as FRED (Federal Reserve Economic Data) and market data providers such as Twelve Data make this easy to chart next to delinquency or deposit trends. The clariBI page for investment research and market data lists the connectors in this area.

Where to start

Most analytics projects in financial firms fail from trying to do everything at once. Pick the report that takes your team the most manual hours each month, usually the board risk pack or the monthly delinquency report, and automate that first. Then:

1. Agree on one customer identifier across systems. Without it, customer-level analysis is guesswork.
2. Write down the definition of each metric: which balance, which date, which exclusions.
3. Decide who can see customer-level data and who only needs aggregates.
4. Keep a log of definition changes so a number that moves for technical reasons is not mistaken for a real change.

Our guide to data governance for growing companies goes into the last three steps.

Where clariBI fits, and where it does not

clariBI is a general BI tool, not a risk engine, a transaction monitoring system or a regulatory reporting product. It fits the reporting layer for a smaller firm:

• Connect a reporting copy of your data (a PostgreSQL, MySQL, SQL Server or Oracle database, or scheduled CSV exports) and market data sources, then build dashboards for delinquency, roll rates, onboarding times and alert ageing.
• Ask questions such as "Show the 30-day delinquency rate by origination month for the last 12 months".
• Access is read-only: clariBI reads your sources and never writes back.
• Organization roles (Owner, Administrator, Analyst, Member, Viewer) control what each person can do; see organization roles. Custom roles and the organization audit trail are on Professional and Enterprise.

Two things to weigh before connecting customer-level data. First, the AI features send column names, a few sample values and calculated results to a third-party language model provider listed in our privacy policy; each question session uses only the sources you pick for it, and the Free and Lite plans make no AI calls at all. Second, clariBI does not hold a SOC 2 or ISO 27001 certification. If your vendor-risk process needs either, aggregated or de-identified data is the safer starting point.

A reasonable first step: connect FRED and a de-identified monthly loan-level export, and chart your 30-day delinquency rate next to the unemployment rate for the last five years.