50 Financial Analytics: Risk, Fraud, and Performance

50.1 Why Financial Analytics Matters

A marketing dashboard that misreports campaign reach by 5 percent might survive a quarter; a finance dashboard that misreports cash position by 5 percent will not survive an hour.

Of all the analytical disciplines, finance has the longest history of working with numbers and the lowest tolerance for sloppy ones. The audience — auditors, the CFO, the board, regulators — is trained to read tables and to spot the number that does not add up. The bar is high, and the visualisation grammar must be tuned for it.

Financial analytics on a BI desk clusters into three jobs. Risk answers how much could we lose, and how likely is it? — credit risk, market risk, operational risk, liquidity risk. Fraud answers did somebody do something they should not have? — transaction fraud, expense fraud, vendor fraud, insider trading. Performance answers did the business deliver against its own and the market’s expectations? — margins, returns, working capital, growth. Philippe Jorion (2007) frames risk analytics around the single discipline of putting a defensible number on potential loss, with Value at Risk as its most recognisable expression. Richard J. Bolton & David J. Hand (2002), surveying the fraud-detection literature, show that the visualisation lens — outlier dashboards, network graphs, time-series anomaly views — has always done at least as much work as the algorithms behind it.

The finance-dashboard contract

Three rules separate finance dashboards from every other kind:

The numbers must tie. If the dashboard total does not equal the GL total to the rupee, finance will not use the dashboard.
The audit trail must exist. Every figure must trace to its source row. Tooltips with drill-through are not optional.
The view must show what is missing. Late filings, unposted journals, unreconciled accounts — finance audiences want to see gaps, not just data.

50.2 Risk Analytics: Quantifying What Could Go Wrong

Risk is the discipline of putting a number on a future loss. The number is always wrong — risk is by definition probabilistic — but the shape of the wrongness, expressed in distributions, percentiles, and tail measures, is what risk dashboards visualise.

Four Risk Families and Their Canonical Visualisations

Family	Question	Core metric	Visualisation lens
Credit risk	Will the borrower or counterparty pay?	Probability of Default, Loss Given Default, Exposure at Default.	Vintage curves, transition matrices, score-band heatmaps.
Market risk	Will prices, rates, or FX move against us?	Value at Risk, Expected Shortfall, beta, duration.	Distribution histograms with VaR cut-line, drawdown charts.
Operational risk	Will a process, person, or system fail?	Loss event frequency, severity, capital charge.	Heatmap of likelihood vs impact, loss scatter.
Liquidity risk	Can we meet obligations as they fall due?	Liquidity Coverage Ratio, cash runway, days payable outstanding.	Maturity ladder, cash-runway chart.

Value at Risk and Expected Shortfall

VaR answers: what is the maximum loss we expect, with X percent confidence, over Y days? A one-day 99 percent VaR of 12 crore means: on 99 of every 100 trading days we expect to lose less than 12 crore; on 1 of 100 we may lose more. The natural visualisation is a histogram of simulated portfolio P&L with the 1st-percentile cut-line drawn through it.

flowchart LR
  A[Historical returns<br/>or Monte Carlo paths] --> B[Simulated P-and-L<br/>distribution]
  B --> C[Sort outcomes<br/>worst to best]
  C --> D[VaR at 99 percent<br/>= 1st percentile]
  C --> E[Expected Shortfall<br/>= avg of worst 1 percent]
  style D fill:#FCE8E6,stroke:#D93025
  style E fill:#FCE8E6,stroke:#D93025

Philippe Jorion (2007) warns that VaR is not a worst-case loss number; it is a percentile. Expected Shortfall — the average of losses beyond the VaR cut — answers the harder question if it goes wrong, how wrong? A defensible risk dashboard shows both, and shows the distribution that produced them, not just the headline number.

Credit-risk vintages and transition matrices

In credit portfolios, the vintage curve is the workhorse view. Each cohort — for example, all auto loans booked in Q1 FY24 — is plotted as cumulative default rate against months on book. Vintages stack on the same chart so the audience sees whether newer cohorts are deteriorating or improving. The eye picks up structural risk shifts faster than any table.

A transition matrix is the second canonical view: rows are starting credit ratings (AAA … D), columns are ending ratings, and cells are the probability of moving from row to column over a fixed period. Visualised as a heatmap, the diagonal dominates for healthy portfolios; off-diagonal mass downward and to the right is migration risk.

Risk dashboards must show the tail

A line chart of average default rate hides everything that matters. Always pair the average with a histogram of individual loan losses and a tail metric (worst 5 percent, worst 1 percent). The CRO is paid to look at the tail.

50.3 Fraud Analytics: Finding the Anomaly

Fraud detection is a search problem with a brutal cost asymmetry — false positives annoy honest customers, false negatives let real fraud through. Richard J. Bolton & David J. Hand (2002) survey three families of approach: supervised (train on known fraud labels), unsupervised (flag outliers without labels), and behaviour profiling (compare each entity to its own past). Visualisation is essential to all three because investigators ultimately have to look at the suspect transactions.

Four Fraud-Visualisation Archetypes That Cover Most Use Cases

Archetype	What it shows	Tool path
Outlier scatter	Each transaction as a point in a feature space; outliers stand out.	Power BI scatter with feature axes; Tableau scatter; Excel chart.
Time-series anomaly	Volume or amount over time with prediction bands; spikes break out.	Power BI line chart with shaded confidence band; Tableau dual-axis with reference band.
Network graph	Customers, accounts, devices as nodes; suspicious links light up.	Power BI Force-Directed Graph; Tableau Sankey or external Gephi; NetworkX.
Benford-law deviation	First-digit distribution of amounts; deviations flag manipulation.	Excel COUNTIF on first-digit; Power BI custom column with LEFT().

Benford’s Law and the digit-distribution test

In genuinely natural data — populations, river lengths, accounting entries — the leading digit follows Benford’s distribution: 1 appears about 30 percent of the time, 9 about 4.6 percent. Fraudsters making up numbers tend to spread leading digits more evenly, which produces a measurable deviation. The simplest fraud chart is a bar chart with two series: expected Benford share and observed share. Audiences without a statistics background read it immediately.

flowchart LR
  A[Population of<br/>journal entries] --> B[Extract leading digit<br/>of amount]
  B --> C[Tally distribution<br/>1-9]
  C --> D[Compare to Benford<br/>expectation]
  D --> E{Deviation<br/>significant?}
  E -->|Yes| F[Flag for review<br/>chi-square or MAD]
  E -->|No| G[Pass]
  style F fill:#FCE8E6,stroke:#D93025
  style G fill:#E6F4EA,stroke:#137333

Behavioural profiling and the customer baseline

For transaction fraud, the most effective dashboards do not look at thresholds in absolute terms — they look at deviation from the customer’s own baseline. The card that has averaged 4 swipes a week, 80 percent in one city, 90 percent under 5,000 rupees, suddenly logs ten swipes in two hours across three cities including one international one. No single transaction breaks a fixed rule; the pattern breaks the customer’s own rule, and the dashboard surfaces that.

The visualisation is two parallel small-multiples — baseline week and current week — for each at-risk customer, with deviation flags coloured red. Power BI does this with a small-multiples line visual; Tableau handles it natively with the Trellis chart pattern.

Investigators need triage, not the entire haystack

A fraud dashboard that lists 11,000 alerts a day is unusable. The dashboard must rank by an investigator-friendly score and cap the daily list at what the team can actually work — typically 50 to 200 cases. Richard J. Bolton & David J. Hand (2002) note that ranked alert lists with rich context are the only deployment shape that actually reduces fraud loss in practice.

50.4 Performance Analytics: The Financial Operating Picture

Performance dashboards report whether the business is making money, using capital well, and meeting its own plan. They are read by people who already know the business, so the design discipline is signal density — every square centimetre must carry a number that matters.

Five Layers of Financial Performance Metrics, Each With a Canonical Visual

Layer	Examples	Visualisation lens
Profitability	Gross margin, EBITDA margin, operating margin, net margin.	Waterfall (revenue to net income), bullet against plan and prior year.
Returns	ROCE, ROIC, ROE, ROA.	Bullet, sparkline trends, peer-comparison bars.
Working capital	DSO, DIO, DPO, Cash Conversion Cycle.	Line trend with target band; component stacked area.
Growth and quality	Revenue growth YoY, organic vs acquired, recurring share, churn.	Combo column-line with growth rate as line.
Cash	Free cash flow, FCF conversion, cash runway, net debt over EBITDA.	Rolling-balance line, runway chart, debt-maturity ladder.

The waterfall: revenue to net income

The waterfall (or bridge) chart reconstructs the income statement visually: starts at revenue, subtracts COGS to gross profit, subtracts opex to operating profit, adjusts for tax and interest, and lands at net income. Each subtraction is a downward bar; each addition (rare in this context) is an upward bar. The chart answers where is the money going? in five seconds, where the table form takes a minute.

Power BI has a native Waterfall visual; Tableau builds it with running-sum tricks and Gantt marks; Excel ships a Waterfall chart type since 2016. All three handle the variance bridge — actual versus plan, expressed as a series of contributing increments — equally well.

Variance bridges and the budget conversation

The most consequential finance chart in most companies is the variance bridge between plan and actual. It opens at Plan on the left, then walks through volume variance, price variance, mix variance, cost variance, FX variance, and other variances, landing at Actual on the right. Each variance is a contributing bar, signed in the colour convention (green favourable, red unfavourable). The CFO uses this chart to brief the board; the BI analyst owns its accuracy.

Bullet charts beat dials and gauges

Many finance dashboards reach for dials and speedometers because they look executive. Stephen Few (2013) showed that the bullet chart — a horizontal bar with a target tick and qualitative bands — encodes the same information in less space, more readably, and prints better. Replace every gauge in a finance dashboard with a bullet chart and the reader will thank you.

50.5 Common Pitfalls

What goes wrong

Numbers that do not tie to the GL. Even by a few rupees. Finance will reject the dashboard.
VaR reported without distribution. A single percentile number with no shape behind it is uninformative and easy to misread.
Mean-only credit reporting. Average default rate hides cohort and tail risk. Always pair with vintages and a tail histogram.
Fraud alerts without ranking. Investigators get the haystack, not the needles.
Gauges and speedometers. Take three times the space of a bullet chart and encode less.
Variance bridges with the wrong sign convention. Half the audience reads green-up as good, the other half as growth — establish and label the convention every time.
Headline KPI without a comparator. A number alone is meaningless. Always show prior period, plan, or peer benchmark.
Dashboards over close-of-month data only. Daily flash views matter more for cash and risk than month-end views; both must exist.
Currency and unit ambiguity. Tooltips and titles must be explicit — crore INR, USD millions. Mixing crore and lakh on a single chart is a classic foot-gun.
Privacy and access creep. Salary, vendor pricing, customer-level credit data — finance dashboards leak quickly. Row-level security and audit logs are mandatory.

50.6 Illustrative Cases

Three case sketches

Yuvijen Forge Components Ltd. credit-risk dashboard. The treasury team builds a Power BI dashboard for trade-receivables risk. Vintage curves, exposure heatmap by customer tier, and a top-20 watchlist with rolling 90-day exposure trends. When two large auto-OEM customers begin slipping, the dashboard flags rising exposure four weeks before the credit committee would have spotted it from monthly reports. Treasury tightens credit limits in time and avoids a 14 crore write-off.

Yuvijen Stores expense-fraud audit. Internal audit team uses a Tableau dashboard built on three years of vendor invoices. Benford’s-law digit distribution flags two vendors whose invoice amounts cluster heavily on round-number digits. Drill-through reveals that both vendors are routing invoices just below the manager’s approval threshold. Audit recovers 86 lakh in inflated billings and the procurement system gains a new tolerance check.

Yuvijen Telecom CFO performance pack. The CFO’s monthly board pack is rebuilt from a 38-page slide deck into a six-page Power BI report. Page 1 is a P&L waterfall; page 2 is the variance bridge against plan; page 3 is the working-capital trio (DSO, DIO, DPO) with target bands; page 4 is the cash runway and debt-maturity ladder; page 5 is segment-level gross margin; page 6 is a watchlist of risk events. Board reading time drops from 35 minutes to 8 minutes; the CFO uses the saved time for a structured Q and A that consistently produces sharper decisions.

50.7 Hands-On Exercise: Build a CFO Performance Dashboard with a Risk Watchlist

Three-page finance dashboard

Aim. Build a three-page financial-analytics dashboard in Power BI that ties together a P&L waterfall, a variance bridge, and a credit-risk watchlist on the same data spine. Tableau equivalents are noted at each step.

Scenario. You are the BI lead in finance at Yuvijen Forge Components Ltd. The CFO has asked for a single report that lets her tell the story of the quarter, walk the variance against plan, and check whether any customer credit exposure is moving into the danger zone.

Deliverable. A three-page Power BI report — Performance, Variance, Risk — with row-level security so that division CFOs see only their own division’s numbers, and an audit-trail tooltip that traces every visual back to its GL line.

50.7.1 Step 1 — Load and validate the data

Use Get Data in Power BI to load four CSVs:

gl_actual.csv — Date, Account, Division, Amount.
gl_plan.csv — Date, Account, Division, Amount.
accounts_receivable.csv — CustomerID, InvoiceDate, DueDate, Amount, Status.
customer_credit.csv — CustomerID, CreditLimit, RatingTier.

In Power Query, type the columns. Build a DimDate table with CALENDARAUTO() and a DimAccount hierarchy (Statement → Group → Account) so the waterfall can roll up cleanly. Validate by checking that the sum of gl_actual[Amount] for the latest closed month matches the closing trial balance to the rupee. If it does not, stop and reconcile — the rest of the dashboard is built on this.

50.7.2 Step 2 — Page 1: Performance waterfall and KPI strip

Drop a Waterfall visual with Account on the category axis (sequenced Revenue → COGS → Gross Profit → Opex → EBITDA → Depreciation → Operating Profit → Tax → Net Income) and Amount as values, scoped to the latest closed quarter.

Above the waterfall, add a strip of five KPI cards:

Revenue (with YoY arrow)
Gross margin %
EBITDA margin %
Operating cash flow
Net debt / EBITDA

Each card uses conditional formatting against the prior period: green for improved, red for deteriorated, with absolute and percentage deltas in the tooltip.

Tableau alternative: build the waterfall using a calculated Running Sum with Gantt bars; build KPI cards as text-only worksheets in a tile container.

50.7.3 Step 3 — Page 2: Variance bridge against plan

Build a second waterfall, this time using calculated drivers. The bridge starts at Plan EBITDA, walks through Volume Variance, Price Variance, Mix Variance, Cost Variance, FX Variance, and Other, and lands at Actual EBITDA.

The variance components require DAX calculated tables that decompose the actual-versus-plan difference. A simple price-volume decomposition uses:

VolumeVariance =
SUMX(
    VALUES(DimProduct[ProductID]),
    (CALCULATE(SUM(gl_actual[Quantity])) - CALCULATE(SUM(gl_plan[Quantity])))
    * CALCULATE(AVERAGE(gl_plan[Price]))
)

PriceVariance =
SUMX(
    VALUES(DimProduct[ProductID]),
    (CALCULATE(AVERAGE(gl_actual[Price])) - CALCULATE(AVERAGE(gl_plan[Price])))
    * CALCULATE(SUM(gl_actual[Quantity]))
)

Establish the colour convention up front: green for favourable variances, red for unfavourable, with the legend always visible. Add a date slicer that lets the CFO pull the bridge for any closed period.

Tableau alternative: same Gantt-mark waterfall pattern; use parameters for period selection.

50.7.4 Step 4 — Page 3: Credit-risk watchlist

This page combines two visuals.

Vintage chart. Line chart with months-on-book on the x-axis, cumulative default rate on the y-axis, one line per origination quarter. Use the accounts_receivable table joined to customer_credit to compute, for each cohort, the cumulative percentage of receivables that hit Status = Defaulted by each month-on-book. The chart reveals deteriorating cohorts at a glance.

Top-20 exposure table. A table visual ranked by current outstanding with the columns CustomerID, RatingTier, ExposureNow, ExposureLimit, %Utilisation, and a sparkline of 90-day exposure trend. Conditional formatting flags utilisation above 85 percent in amber, above 100 percent in red.

Add a threshold parameter so the credit committee can set the alert utilisation themselves; the highlights respond live.

Tableau alternative: build the vintage as a line chart with the cohort dimension on Colour; build the table with sparklines via reference-line tricks or the embedded sparkline calculation.

50.7.5 Step 5 — Audit-trail drill-through

For each visual, configure a drill-through tooltip page that shows the underlying rows: for a waterfall bar, the GL entries that compose it; for a vintage point, the loan IDs in that cohort and month. The tooltip page must be short — ten rows, plus a View All button that opens a paginated table page. This is the audit trail the dashboard contract demands.

50.7.6 Step 6 — Row-Level Security and audit logging

Implement RLS so that a division CFO sees only their own division’s numbers. In Power BI, build a DivisionSecurity table, define a DAX role with [Division] = USERNAME(), and test by viewing as different identities. Enable workspace audit logs so the data team can see who opened the report and which pages they viewed — for finance dashboards, this auditability is a compliance line, not a nice-to-have.

50.7.7 Step 7 — Sign-off and the close-of-month protocol

Before the dashboard is published for board use, run a three-way sign-off: BI analyst (data integrity), financial controller (accounting accuracy), and CFO (storytelling and emphasis). Record the sign-off in a metadata page on the report. Establish a close-of-month protocol so that the dashboard is locked from edits between books-close and board-pack distribution; emergency edits require a controller approval recorded in the audit log.

Connect to the Visualisation Layer

Financial analytics is the most mature application of the visualisation grammar in this book. Waterfalls and variance bridges from Chapter 12 carry the P&L story. Bullet charts from Chapter 12 stand in for the gauges finance audiences keep asking for and the analyst keeps replacing. Reference bands and tail histograms render risk distributions. Vintage line charts render cohort behaviour. Conditional formatting and RLS from Chapter 36 enforce the audit-trail and access-control rules the function lives by. Mobile design from Chapter 47 means the CFO can answer a board question from a hotel room. The storytelling discipline of Chapter 48 is what turns the dashboard into a five-minute briefing instead of a thirty-minute scroll.

Files and Screen Recordings

Power BI three-page CFO pack with RLS (yuvijen-forge-cfo-pack.pbix), Tableau equivalent (yuvijen-forge-cfo-pack.twbx), workshop dataset (yuvijen-forge-finance-data.xlsx), screen recording of the dashboard tour (yuvijen-forge-cfo-walkthrough.mp4), and the three-way sign-off template (yuvijen-forge-signoff.docx) will be embedded here.

Summary

Concept	Description
Why Finance Is Different
Numbers Tie to the GL	Dashboard total must reconcile to the trial balance to the rupee
Audit Trail Required	Every figure must trace to its source row via drill-through tooltips
Show What Is Missing	Late filings, unposted journals, unreconciled accounts must be visible
Low Sloppy-Number Tolerance	Finance audiences will reject dashboards that contain casual numbers
Three Jobs
Risk Job	How much could we lose, and how likely is the loss?
Fraud Job	Did somebody do something they should not have?
Performance Job	Did the business deliver against its plan and the market?
Risk Families
Credit Risk	Probability of default, loss given default, exposure at default
Market Risk	Value at risk, expected shortfall, beta, duration
Operational Risk	Loss event frequency and severity, capital charge
Liquidity Risk	Liquidity coverage ratio, cash runway, days payable outstanding
Risk Family Visuals	Each risk family carries its own canonical chart
Risk Visualisation
VaR with Distribution	Histogram of simulated P and L with percentile cut-line
Expected Shortfall	Average loss beyond the VaR cut answers if it goes wrong, how wrong?
Vintage Curves	Cumulative default rate per origination cohort against months on book
Transition Matrix Heatmap	Probability of moving from start rating to end rating over a period
Average plus Tail	Always pair averages with tails to expose hidden risk
Fraud Archetypes
Outlier Scatter	Each transaction is a point in feature space; outliers stand out
Time-Series Anomaly	Volume or amount over time with prediction bands; spikes break out
Network Graph	Customers, accounts, devices as nodes; suspicious links light up
Benford-Law Deviation	First-digit distribution flags fabricated or manipulated amounts
Behavioural Profiling	Compare entity behaviour to its own past, not to fixed thresholds
Ranked Alert Lists	Capped daily lists ranked by investigator-friendly score
Performance KPIs
Profitability Layer	Gross margin, EBITDA margin, operating margin, net margin
Returns Layer	ROCE, ROIC, ROE, ROA
Working Capital Layer	DSO, DIO, DPO, cash conversion cycle
Growth and Quality Layer	Revenue growth, organic versus acquired, recurring share, churn
Cash Layer	Free cash flow, FCF conversion, runway, net debt over EBITDA
Layered Visualisation	Each KPI layer has its own visualisation idiom
Canonical Charts
P and L Waterfall	Walks revenue through to net income with downward bars
Variance Bridge	Walks plan to actual through volume, price, mix, cost, FX variances
Bullets Replace Gauges	Bullet charts encode target and qualitative bands more readably than gauges
Sparkline Trends	Sparkline columns inside tables show trends without consuming space
Cash Runway and Maturity Ladder	Cash runway line and debt-maturity ladder show liquidity headroom
Common Pitfalls
Pitfall: Numbers Do Not Tie	Even a few rupees of variance from GL kills credibility
Pitfall: VaR Without Distribution	Single percentile number with no shape behind it is uninformative
Pitfall: Mean-Only Credit	Average default rate hides cohort and tail risk
Pitfall: Unranked Fraud Alerts	Investigators receive the haystack rather than the needles
Pitfall: Gauges Instead of Bullets	Dials and speedometers take more space and encode less information
Pitfall: Unclear Sign Convention	Half the audience reads green-up as good, the other half as growth
Pitfall: KPI Without Comparator	A number alone is meaningless without prior period, plan, or peer comparator
Pitfall: Currency Ambiguity	Mixing crore and lakh on a single chart is a classic error
Pitfall: Access Creep	Salary, vendor, customer-credit data leaks without RLS and audit logs
Hands-On Pack
Page 1 — Performance Waterfall	P and L waterfall with revenue-to-net-income bars and KPI strip
Page 2 — Variance Bridge	Walks plan to actual through DAX volume and price decompositions
Page 3 — Risk Watchlist	Vintage chart plus top-20 customer exposure watchlist
Audit-Trail Drill-Through	Tooltip pages traceable to the GL or loan-level rows
Division RLS	Division CFOs see only their division through Power BI RLS
Sign-Off Protocol	BI analyst, controller, CFO sign-off recorded in metadata page