Sign in

Blog · Finance metrics and formulas · Finance and FP&A teams

Outlier analysis in financial data

Outlier analysis helps finance teams investigate unusual transactions and account movements. Work through an invoice screen, an account variance review and the reconciliation that keeps exceptions connected to the ledger.

The short answerOutlier analysis in financial data identifies transactions or account movements that differ substantially from a relevant comparison group. Start with reconciled exports, group comparable records, and apply a stated amount or movement rule. Investigate flagged rows against invoices, approvals and period timing. An unusual value is a review lead; it does not establish an accounting error or fraud.

Outlier analysis in financial data flags unusual transactions and account movements for investigation. Begin with a complete export, compare like records, and apply an explicit review rule. Keep the ledger total intact while the finance team determines whether each exception reflects a valid business event, a timing issue or an error.

Establish the comparison before setting a threshold

A large amount is unusual only relative to a useful reference group. A freight invoice for an expedited shipment may differ from routine deliveries for a valid reason. A new revenue contract may be large without being recorded incorrectly.

NIST defines outliers relative to the other observations and emphasizes understanding unusual points before considering their removal. For a finance review, that means defining the account, entity, currency, service period and type of transaction first.

Export transaction ID, posting date, account, department, vendor or customer, document number, currency and signed amount. Add service dates or quantities where they determine comparability. Keep credits separate from ordinary invoices unless the review deliberately covers net activity. A blank amount needs resolution; it is not a zero-dollar transaction.

First reconcile the export to the ledger control total for the same scope. Financial data reconciliation covers that underlying matching job. A screen applied to an incomplete extract can miss the very rows finance needs to investigate.

Screen comparable invoice amounts

Consider eight freight invoices at a fictional US distributor. They initially share the same delivery account and review period. Put the document IDs in column A and positive invoice amounts in B2:B9, with one invoice per row.

Document Invoice amount (USD) Initial screen
F-101 950 Within fences
F-102 1,000 Within fences
F-103 1,000 Within fences
F-104 1,050 Within fences
F-105 1,100 Within fences
F-106 1,150 Within fences
F-107 1,200 Within fences
F-108 4,800 Review
Total 12,250

Use the interquartile range, or IQR, to describe the spread of the middle part of the group. This screen uses fences at the lower quartile minus 1.5 times IQR and the upper quartile plus 1.5 times IQR.

In Excel, Microsoft documents the inclusive quartile function used below. Put the summary calculations in E2:E6:

E2: =QUARTILE.INC($B$2:$B$9,1)
E3: =QUARTILE.INC($B$2:$B$9,3)
E4: =E3-E2
E5: =E2-1.5*E4
E6: =E3+1.5*E4
C2: =IF(OR(B2<$E$5,B2>$E$6),"Review","Within fences")

Fill C2 down through C9. Keep the total outside B2:B9 so it does not enter the quartiles.

Screen step Calculation Amount (USD)
Lower quartile, Q1 Inclusive quartile of eight invoices 1,000.00
Upper quartile, Q3 Inclusive quartile of eight invoices 1,162.50
IQR 1,162.50 - 1,000.00 162.50
Lower fence 1,000.00 - 1.5 × 162.50 756.25
Upper fence 1,162.50 + 1.5 × 162.50 1,406.25

The inclusive calculation interpolates between sorted observations. Q1 falls between the two $1,000 invoices. Q3 falls one-quarter of the way from $1,150 to $1,200, producing $1,162.50. This states the quartile convention so another reviewer can reproduce it.

Only F-108 lies outside the fences. The screen does not assign a probability that it is wrong. With a small group, use this as a review aid and inspect the underlying distribution before extending the rule to other accounts.

Investigate the row and prove the total

The AP reviewer retrieves F-108 and its shipment approval. In this example, the records show a valid expedited delivery; the other invoices cover ordinary deliveries. Record that explanation and identify whether future comparisons should separate delivery types.

The review population still contains $12,250. The unflagged invoices total $7,450, and the flagged invoice is $4,800:

Full total: =SUM(B2:B9)
Review total: =SUMIF(C2:C9,"Review",B2:B9)
Other total: =SUMIF(C2:C9,"Within fences",B2:B9)
Control difference: =SUM(B2:B9)-SUMIF(C2:C9,"Review",B2:B9)-SUMIF(C2:C9,"Within fences",B2:B9)

The control difference is 12,250 - 4,800 - 7,450 = $0. Every amount remains accounted for. If management wants a view of routine delivery costs, show the expedited amount as a separate reconciling line rather than silently deleting it.

This follows NIST's distinction between labeling and investigating outliers: flagging a point does not establish that it is erroneous. For the commercial interpretation of a valid exceptional order, see the one-off order example.

Review account movements with a separate rule

Invoice-size screens do not explain changes in account totals. Run a second review on matching periods. Here, use a stated internal rule: review a movement when its absolute dollar change is at least $10,000 and its absolute percentage change is at least 20%. Route a newly nonzero account to a separate review regardless of that percentage test.

These are illustrative management thresholds, not accounting or assurance standards. The following table is a separate monthly account extract; it is not a rollup of the invoice example.

Account May (USD) June (USD) Change (USD) Change % Result
Revenue 1,200,000 1,320,000 120,000 10.0% Within screen
Freight 40,000 66,000 26,000 65.0% Review
Software expense 10,000 12,000 2,000 20.0% Within screen
Travel expense 0 3,500 3,500 Not defined Review new balance

Put account, May and June in A2:C5. Calculate the change in D2, relative movement in E2 and review label in F2, then fill down:

D2: =C2-B2
E2: =IF(B2=0,"",D2/ABS(B2))
F2: =IF(B2=0,IF(C2=0,"No change","Review new balance"),IF(AND(ABS(D2)>=10000,ABS(E2)>=20%),"Review","Within screen"))

Percentages are shown to one decimal place. Revenue misses the percentage threshold, and software misses the dollar threshold. Travel has no meaningful percentage growth from zero. This prevents a division error from hiding a newly active account.

FP&A assigns each selected movement to its business owner. Procurement can explain freight rates or urgent shipments; department managers can explain travel; the controller resolves postings and cutoff. The wider financial analysis then considers whether those explanations affect spending decisions or forecasts.

Where it goes wrong

  • Mixing unlike populations. Different currencies, contract lengths or delivery types can create artificial exceptions. Separate them before interpreting distance from the group.
  • Deleting valid exceptions. A cleaner-looking average can conceal real expenditure. Preserve reported totals and show every exclusion in a reconciliation.
  • Using percentage movement alone. Small or zero starting balances make percentages unstable or undefined. Retain the dollar change and a new-balance review.
  • Assuming a flag proves fraud. Unusual timing or size identifies a question. Documents and the finance team's investigation determine the finding.
  • Confusing detection with explanation. A rule can flag freight growth without distinguishing higher shipment volume from higher rates. Check quantities and supporting terms.

Covirage is not affiliated with Microsoft.

Compare the underlying rows in Covirage

Bring your reconciled transaction export or prepared account-movement workbook, including the identifiers needed for review. Covirage maps the columns, its tools compute and check the comparisons, and the external AI model explains the results with every figure traced to its rows; your finance team chooses the review criteria and evaluates exceptions. Create a free Covirage account and bring a sample of your own data.

See analysis of your spreadsheet files for the supporting workflow.

Questions people ask

Should finance remove outliers before reporting results?

Keep valid transactions in the reported totals. If a transaction is erroneous, follow the normal correction process and retain the explanation. A separate recurring-cost view can be useful, but show the excluded amount and reconcile that view back to the complete financial result.

What is a useful threshold for a financial outlier?

Choose a threshold that fits the account, transaction size and review capacity. A combination of dollar movement and relative movement helps prioritize account changes. A quartile screen can help with comparable invoice amounts. Neither method provides a universal accounting materiality threshold.

Can an outlier screen detect fraud?

A screen can identify records worth examining, including unexpected amounts or unusual timing. It cannot determine intent or establish fraud. Finance staff need supporting documents, business context and the appropriate investigation process before reaching a conclusion about a transaction.