Case Study: Building Predictive Credit Risk Scores and Business Health Index for Xero

Challenge

Xero, a global cloud-based accounting software platform, engaged Factfin to explore how its extensive business data could be leveraged to understand invoice payment risk at the customer level, support bank lending decisions through credit scoring, and aggregate insights to monitor business health across the broader economy.

While Xero possessed vast volumes of high-frequency, structured financial data, it had not previously explored advanced predictive modelling to support credit risk use cases. This project sought to establish the feasibility and business value of predictive credit scoring, using Xero’s internal data and infrastructure, and build prototype scoring algorithms.

Objectives

The engagement aimed to:

• Assess if invoice payment and business health outcomes could be predicted using Xero’s customer data.

• Build first-generation Trade and Health Scores for individual risk prediction.

• Explore how network-level data (i.e., customer-to-customer interactions) could be incorporated to improve score performance.

• Deliver early versions of aggregated indices (e.g., Health Index) to track macroeconomic trends.

• Identify data gaps, system constraints, and the production readiness of predictive models.

Approach

1. Credit Score Feasibility Study

• Validated the suitability of Xero’s infrastructure and data for building predictive algorithms.

• Developed working definitions of score targets

• Created structured data marts from transaction-level data for scoring.

• Built and tested over 200 predictive variables to evaluate model performance

• Segmented entities to account for data richness.

2. Trade Score Development

• Predicted likelihood of late invoice repayment (30+ days past due) using invoice behaviour, receivables patterns, and entity relationships.

• Achieved strong predictive power.

3. Health Score Development

• Predicted business deterioration using revenue, liabilities, and expense patterns.

• Built a rolling monthly data mart and conducted vintage and cycle stability analysis.

4. Incorporating Network Data

• Conducted a deep dive into how business relationships (via invoice connections) could be incorporated into risk models.

• Identified methods for constructing first- and second-degree connections

• Highlighted that network data is Xero’s unique competitive advantage in the credit analytics space.

Outcomes

• Delivered working prototype credit scoring algorithms using only internal Xero data.

• Demonstrated high predictive accuracy with clear business applications for:

• Invoice payment risk triage

• Lending product eligibility and pricing

• Portfolio health monitoring

• Mapped out clear phased roadmap from feasibility to live product, including future use of network data and external data sources.

• Identified data governance and identity matching as key enablers for unlocking full network potential.

Deliverables

• Trade Score and Health Score prototypes (including stability and validation tests)

• Information Value analysis of 200+ variables

• Health Index prototype and monthly index report

• Network augmentation strategy and feasibility analysis

• Production datamart specification and implementation guide

• Full documentation and insight presentation

Strategic Impact

• Positioned Xero to offer differentiated credit scoring insights to banks, fintechs, and government partners.

• Created pathways for monetisation of aggregated business insights while preserving user privacy.

• Enabled use of Xero’s platform data for early warning signals and economic pulse-checks at national or sectoral levels.

• Delivered a playbook for turning accounting data into decision-grade risk scores—a first in Xero’s history.