Financial services analytics

Financial services data analytics shows insurers and lenders which channels and underwriting years make money, and which figures will stand up in a board or compliance review. In insurance analytics that means loss ratio and combined ratio by underwriting year and channel, on earned premium.

Last reviewed

Insurance dashboard, example

Combined ratio 97.1%, under 100, so the book made £140K on underwriting

Example data

Underwriting year 2025

Combined ratio

97.1%

Earned premium
£4,820K
Earned premium: the share of written premium for cover already given
Claims incurred
£3,040K
Expenses
£1,640K
Loss ratio
63.1%
Expense ratio
34.0%

2.9 points under 100%£140K underwriting result

  1. 63.1% loss + 34.0% expense = 97.1% combined
  2. 100% − 97.1% = 2.9 points
  3. 2.9% of £4,820K earned premium = £140K
Example data: an invented home insurance book. Underwriting year 2025 means policies starting in 2025, with claims as at 30 September 2026. Claims on this year are still being reported, so its loss ratio will move.

Insurance and banking KPIs, with formulas

Seven insurance and lending figures, each with its formula.

The finance dashboard covers the company-level pack that sits above them.

KPIFormulaA good readingThe common mistake
Loss ratioClaims incurred ÷ earned premium × 100, where claims incurred = claims paid + change in claims reserves, including incurred but not reportedSteady across underwriting years compared at the same ageDividing by written premium, which arrives at the start of cover and flatters a growing book
Expense ratioUnderwriting expenses ÷ earned premium × 100, on the same base as the loss ratio. Some insurers quote it on written premium.Falling as the book grows, because fixed costs spread furtherPutting expenses over written premium while the loss ratio uses earned, so the two ratios no longer add up
Combined ratioLoss ratio + expense ratioUnder 100%, which means an underwriting profit before investment incomeComparing last year's underwriting year with one from five years ago. The recent one has had less time for claims to come in.
Claims frequencyNumber of claims ÷ earned exposure in policy yearsSteady by product and channel, with any jump traced to a causeDividing by policies in force at the period end, which ignores policies that were on cover for only part of it
Quote-to-bind ratePolicies bound ÷ unique quotes, deduplicated by customer and riskRead by channel, with price comparison sites on their ownCounting every requote. One customer on a comparison site can generate many quotes for the same home.
Renewal retentionPolicies renewed ÷ policies invited to renewRead by tenure and by the price change offered at renewalIncluding policies the insurer chose not to invite, which counts its own decision as customers leaving
Arrears rateBalances more than 30 days past due ÷ total outstanding balancesRead by the month the loans were made, each month compared at the same ageReading it across the whole book while lending grows. New loans can't be in arrears yet, so growth pulls the rate down.

Where financial services data usually lives

Quote, policy and claims systems, and where joins break.

The rating engine and quote logsEvery quote, including each requote from a price comparison site. One customer collects many quote IDs, and nothing links them until someone builds the match.
The policy administration systemPolicies, mid-term adjustments and renewals. If a renewal issues a new policy number, the policy's history breaks unless the old and new numbers are linked.
The claims systemClaims, payments and case reserves. Reserves change every day, so a report run on Monday and again on Friday gives two loss ratios. Loss ratios are read from dated snapshots.
The finance ledger and actuarial reservingEarned premium and the reserve for claims incurred but not reported, set at quarter end. If that reserve is held for the whole book, splitting it by channel needs an allocation rule agreed with the actuaries.
Loan servicing, for lendersBalances, payments and days past due. Arrears are a month-end snapshot, and the snapshot's timing has to match the ledger's or the two won't reconcile.

A loss ratio on written premium makes a fast-growing book look profitable until the claims arrive.

What an insurance dashboard shows by underwriting year

Reading loss ratio by underwriting year as it settles.

An underwriting year collects every policy that started in that year and follows its claims for as long as they keep coming in. Read that way, each year's loss ratio is a figure that settles over time, and the dashboard shows each year at the same age, such as 12 or 24 months after it started. Split by channel, it shows whether business from comparison sites prices as well as business written directly. Both ratios sit on earned premium, the share of premium for cover already given.

Insurance analytics: the quote-to-bind funnel

How to deduplicate quotes before measuring conversion.

Quote-to-bind only means something once quotes are deduplicated. The match key is the customer (name, date of birth and postcode) and the risk (the property and the cover chosen). Requotes for the same customer and risk inside a window you set, for example 30 days, collapse into one quote, and the last one before binding is the one that counts. Without that step a customer who gets ten quotes on a comparison site and buys once counts as nine lost sales and one win, and that customer's conversion reads 10% when it was 100%.

Analytics services for insurers and lenders

The services insurers and lenders need most, and why.

When a reviewer asks why last quarter's loss ratio moved, the answer is two dated reserve snapshots and one definitions entry. A figure that can't give that answer isn't ready for the board.

Lineage in financial services data analytics

How each figure traces back for a compliance reviewer.

Each figure on a report carries an ID that points to its entry in the definitions log: the formula, the source tables, the snapshot date and the named owner. The SQL that builds it lives in version control, so a reviewer can see who changed a definition and when. The path from a combined ratio in the board pack runs through its ID and definitions entry to the claims and premium tables, and the same path works for the regulator's questions.

Regulation

The FCA's Consumer Duty requires firms to act to deliver good outcomes for retail customers, under four outcomes: products and services, price and value, consumer understanding and consumer support. Firms must assess, test, understand and evidence those outcomes. A firm's board, or equivalent governing body, must review and approve a report on those outcomes at least once a year. That evidence is MI, and a reviewer will want to follow it back to source like any other figure.

Source: FCA, About the Consumer Duty and Consumer Duty: information for firms. Reviewed 20 November 2026.

AI in financial services analytics

What AI can do in regulated MI, with lineage.

An AI assistant can draft the monthly MI commentary or answer "which channel's loss ratio rose this quarter?" in plain English. In a regulated firm, a reviewer will want the same lineage for an AI answer as for any other figure: the query it ran, the table it read and the snapshot date. AI-led analytics shows all three with every answer, so a reviewer can check an AI figure the way they'd check one from the dashboard.

Questions people ask

Answers on Power BI, quote tracking, lenders and AI.

Do you build insurance dashboards in Power BI?

Yes. The model holds premium and claims by policy, with underwriting year, channel and a dated reserve snapshot on every row. Loss ratio, expense ratio and combined ratio are measures on earned premium, so the three always add up on any slice of the book.

Can you set up tracking for insurance quote journeys?

Yes. Each step of the quote form is tracked with the quote reference attached, so the web funnel joins to the rating engine's quotes and the policies bound. Fields a customer types in, such as name or date of birth, are kept out of analytics tags.

Do you work with lenders as well as insurers?

Yes. In banking data analytics for lenders, the core figures are applications, approvals, funded loans and arrears, with arrears read by the month each loan was made. The same rules apply: one definition per figure, a named owner, and a path from the report back to the loan servicing tables.

Can AI be used for MI reporting in financial services?

For drafting commentary and answering questions about the MI, yes, provided every figure it gives shows the query and source table behind it. A figure that can't be traced shouldn't reach a board pack, whoever or whatever produced it.

Why does a growing insurance book need earned premium?

Written premium is booked when the policy starts, while claims arrive over the year of cover and after it. A book that's growing fast writes premium well ahead of its claims, so a loss ratio on written premium looks healthy until the claims catch up.

Want your loss ratio by underwriting year?

Bring a premium and claims extract. The first call works out what earned premium and a dated reserve snapshot would take on your systems.

Book a call