How to choose analytics software for financial services: questions, data and traps

Most buying decisions for analytics start from a feature list. For financial services firms the better start is the questions that come back every month, the files already on hand, and the traps that make a tool look right in a demonstration and wrong in the first board meeting.

Start from the questions

Any tool you consider should answer these from your data, not from a sample. Ask to see it.

The data you already hold

• Product systems
• Customer master
• Holdings
• Segment file
• Transaction file
• Balances
• Revenue
• Segment rules

If a vendor needs a warehouse built before it can read these, count that in the cost and the time.

Ten questions to ask any vendor

1. What does it need in place before the first answer? A warehouse, a data model, a modelling language, a partner? Ask for the list and the typical weeks.
2. Who does the setup, and who maintains it? Your team, the vendor, or a partner, and what that costs after year one.
3. Does the AI calculate figures, or choose from computed ones? A language model that writes queries or code can produce a plausible wrong number. Ask what it is allowed to do.
4. Does every total reconcile to a control figure? Ask to see a bridge that does not sum and what the product does about it.
5. Can every figure be opened to its rows? An answer nobody can check becomes a debate in the meeting.
6. What does it cost in the first year, all in? Licences, consumption, implementation, modelling and training, not only the seat price.
7. How does data arrive, and who holds credentials? A file your systems already export, a scheduled drop, or a live connection with the vendor holding keys.
8. What happens to the data, and where is it stored? Residency, retention, deletion, and whether names can be replaced with identifiers.
9. Can we see it on our own data before we sign? A demonstration on a sample dataset tells you little about your own.
10. What does the tool do when it cannot answer? It should say so. A confident guess does more harm than no answer.

Checks specific to financial services

Ask whether the tool enforces these, and what it does when they fail:

• Customer view: Accounts = linked to a resolved customer + unlinked; revenue likewise
• Holdings: Product revenue by customer sums to the revenue ledger
• Opened and unused: Opened = active + unused + closed within window
• Migration: Customers at start + new − lost = customers at end, per band, with moves netting to zero

The traps

The single customer view as a measured thing. It is treated as a project with an end date, not as a coverage figure reported every month.

Opened and unused. Counted as sales on the day; never revisited.

Tenure against depth. The customers least likely to leave and most likely to say yes, in nobody's campaign.

Measures to leave out

Products per customer as one firm-wide average. Moves with record duplication and unused openings more than with anything real.

Accounts opened. Without activation it rewards the wrong behaviour.

Customer numbers. Counts records, not customers, until the view is resolved.

A scorecard

Where Covirage fits

Covirage reads the exports above, answers the questions with figures our tools compute and check, and is set up for you within a week. See analytics software for financial services compared, AI analytics for financial services and Covirage for Financial services.

For the measures in full, with formulas and exports, read Customer KPIs for financial services firms.