# Data-drivenData-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.Voir la définition complète → kpis for product and risk teams
A fraud analyst at a mid-size digital bank pulls up the morning dashboard: match rate on identity verification dropped from 96% to 89% overnight. No one changed the model. What changed was a data feed from a credit bureau that started returning incomplete address fields. Three hours later, onboarding conversion is down 14%, and nobody in the room knows if it's a data problem or a risk problem. It's both, and that's the point of this lesson.
Product and risk leadership in fintech don't read raw data-quality logs. They read KPIs (key performance indicators): compact numbers that translate underlying data health into business decisions. Your job, whether you're a data analyst, product manager, or risk lead, is to build the bridge between the two.
Before any KPI means anything, you need to know what feeds it.
Identity and verification data: government ID scans, selfie biometrics, device fingerprints, phone carrier data. Used in KYC (Know Your Customer, the regulatory requirement to verify customer identity) and KYB (Know Your Business) checks.
Credit and bureau data: from providers like Experian, Equifax, TransUnion in the US, or Schufa in Germany and Credit Kudos-style open bankingopen bankingCadre réglementaire (PSD2 en Europe) obligeant les banques à partager les données clients via des API standardisées, avec consentement, transformant les données bancaires en actif compétitif. data in the UK. Feeds underwriting and risk scoring.
Transaction data: card networks (Visa, Mastercard), ACH (Automated Clearing House, US bank transfer rail), SEPA (Single Euro Payments Area) transfers. This is the highest-volume, highest-velocity data most fintechs handle.
Open banking / account data: via aggregators like Plaid (US) or providers regulated under PSD2 (the EU's Payment Services Directive 2, which mandates bank data-sharing via APIs) in Europe.
Behavioral and device data: session logs, app usage, geolocation. Increasingly used in fraud models.
Each source has a different refresh rate, error profile, and regulatory sensitivity. A KPIKPIKey Performance Indicator, a measurable value that shows how effectively you're achieving a specific objective, tracked over time against a target.Voir la définition complète → is only as trustworthy as the weakest data source feeding it.
Here's the translation layer. Data teams measure quality; leadership wants performance.
What it measures: the percentage of records that successfully link across two datasets, for example, a submitted ID matching a bureau record, or a transaction matching a known customer profile.
Why leadership cares: low match rate means higher manual review costs, slower onboarding, and potential regulatory exposure (unverified customers under KYC rules).
Worked example: A neobank processes 10,000 onboarding applications. 8,700 auto-match against bureau data on the first pass. Match rate = 8,700 / 10,000 = 87%. If the acceptable benchmark internally is 92%+ (a common informal target cited by identity-verification vendors like Jumio and Onfido, as an estimate), a 5-point gap signals either a data feed issue or a genuine rise in thin-file applicants (people with little credit history).
What it measures: time from data receipt to a risk or credit decision. Often broken into sub-metrics: data ingestion time, model scoring time, human review time.
Why leadership cares: latency directly affects conversion. Every extra second in a loan approval flow is correlated with drop-off; fintech UX benchmarks commonly cite noticeable abandonment increases past 3 to 5 seconds of wait, though exact figures vary by product and should be treated as directional estimates.
Worked example: A BNPL (buy-now-pay-later) provider targets sub-2-second decisions at checkout. If bureau APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.Voir la définition complète → response time balloons from 400ms to 1.8 seconds during peak hours, total decision latency might cross 2.5 seconds, breaching SLA (service-level agreement) and triggering cart abandonment.
What it measures: the proportion of customers or transactions for which a required data field is present and usable, versus missing or null.
Why leadership cares: coverage gaps quietly bias risk models. If income data is missing disproportionately for gig-economy applicants, the model either rejects them by default or scores them on incomplete information, a fair-lending concern under laws like the US Equal Credit Opportunity Act (ECOA).
Worked example: Out of 50,000 active loan accounts, 46,500 have complete income-verification data. Coverage ratio = 46,500 / 50,000 = 93%. Segment that by channel: if coverage drops to 78% for a specific partner integration, that's the actionable insight, not the aggregate number.
Product and risk KPIs rest on governance fundamentals that rarely appear on the leadership dashboard directly but explain why it moves:
A simple governance check, run in Python on a transaction feed, might look like this:
import pandas as pd
df = pd.read_csv("transactions.csv")
completeness = df["customer_id"].notna().mean()
timeliness_hours = (pd.Timestamp.now() - df["ingested_at"].max()).total_seconds() / 3600
print(f"Completeness: {completeness:.1%}")
print(f"Data freshness (hours since last record): {timeliness_hours:.1f}")This kind of check feeds the coverage ratio and latency KPIs above. It's not glamorous, but it's the plumbing.
Vérification des acquis
1. In the opening scenario, the identity verification match rate dropped and onboarding conversion fell shortly after. What does this sequence illustrate about KPIs in fintech?
2. Why does the lesson emphasize that 'a KPI is only as trustworthy as the weakest data feed behind it'?
3. A risk analyst wants to explain a sudden change in a fraud KPI to product leadership, who don't review raw data-quality logs. What is the most effective approach based on the lesson's framing?
4. Select ALL correct answers about why different data sources feeding fintech KPIs matter for risk and product teams.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers about the categories of data described as feeding fintech KPIs.
Sélectionnez toutes les réponses correctes.
When you brief leadership, structure the dashboard in three tiers:
1. Headline KPIs (match rate, decision latency, coverage ratio): trend lines, week over week, with a clear threshold or target line.
2. Segment breakdowns: by channel, product, geography. Aggregates hide the real story, as in the coverage ratio example above.
3. Root-cause flags: a short annotation layer noting known data incidents (e.g., "bureau APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.Voir la définition complète → degraded 03:00 to 06:00 UTC") so leadership doesn't mistake a pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.Voir la définition complète → outage for a strategy failure.
Benchmark sources worth knowing: the Consumer Financial Protection Bureau publishes fair-lending and complaint data useful for sanity-checking coverage gaps in the US. In Europe, the European Banking Authority publishes guidelines on ICT and data risk management relevant to DORA compliance.
🎬 [VIDEO: "How Fintechs Use Data to Make Real-Time Decisions" - youtube.com/results?search_query=fintech+real+time+data+decisions - search for recent explainer content from fintech data platforms like Plaid or Marqeta on real-time decisioning architecture]