Overview

Begini's scoring models improve over time as more data becomes available. The Data Upload feature allows you to share loan performance outcomes — repayment behaviour, defaults, delinquency — back with Begini so that models can be recalibrated against real-world results from your portfolio.

The more performance data you share, the more accurately the model reflects the behavioural patterns of your specific users and market. This is one of the most significant levers available for improving predictive performance over time.

Why performance data matters

At deployment, Begini's scoring models are built on broad behavioural data. They are predictive from day one — but they are not yet calibrated to the specific characteristics of your applicant population, your market conditions, or your lending product.

The data collected in the first months of operation is pivotal. It is during this period that the relationship between Begini's behavioural signals and actual repayment outcomes in your portfolio begins to emerge. This is the data that makes recalibration possible.

Without performance data sharing, the model continues to operate on its initial calibration. With it, the model can be tuned specifically to your deployment — improving risk separation, reducing false positives, and better reflecting how your applicants actually behave.

How recalibration works

Recalibration follows a structured process:

1. Data upload

You share loan performance outcomes with Begini via the Data Upload feature. This typically includes repayment status, delinquency flags (DPD 30/60/90), default outcomes, and the Unique IDs used during the original assessment sessions (A template is available via the Beacon Dashboard). Begini matches these outcomes to the corresponding assessment sessions and behavioural data.

2. Recalibration

Begini's AI engine analyses the performance data and recalibrates MLA s in the scoring model against your real-world outcomes. This is done without requiring input from your side at this stage — the team works on the model adjustments independently, ensuring the updated model reflects the behavioural patterns specific to your portfolio and market.

3. Review

Once recalibration is complete, Begini shares the findings and updated model outputs with your organisation via the Beacon Dashboard. This is the point at which you review the recommendations, understand what has changed, and agree on the approach before anything goes live. Changes are not applied to production without your sign-off.

4. Deploy

The updated model is deployed to your environment. Depending on your preference and risk appetite, this can be done in two ways:

Direct deployment — the recalibrated model replaces the existing model immediately and is applied to all subsequent assessments.

A/B testing — the recalibrated model runs in parallel with the existing model across a defined split of traffic. This allows you to validate performance uplift against real applicants before committing to a full rollout.

Regional vs deployment-level recalibration

The scope of recalibration depends on your data volume and subscription plan:

Deployment-level recalibration produces the strongest performance gains because the model is optimised exclusively for your applicant population. Regional recalibration still delivers meaningful improvement and is a natural starting point for new deployments building up data volume.

What good performance data looks like

To enable effective recalibration, the performance data you share should:

• Cover a sufficient time window — typically at least 3 months of repayment history per cohort
• Include clear outcome flags — repaid, delinquent (with DPD bucket), defaulted, or written off
• Be linked to Begini Unique IDs so outcomes can be matched to assessment sessions
• Represent a meaningful volume of completed assessments — the more outcomes available, the more statistically robust the recalibration

Begini's data science team will guide you on the specific format and fields required.

The ongoing benefit

Recalibration is not a one-time event. As your portfolio grows and market conditions evolve, periodic uploads and recalibrations ensure the model continues to reflect current reality. Each cycle of data sharing and recalibration compounds the improvement — the model gets progressively better aligned to your market over time.

Teams that invest in regular performance data sharing typically see the strongest long-term gains in risk separation and portfolio performance.

Next steps

• Understanding Assessment Results — how to interpret scores and outputs
• Data, Exports & Monitoring — access and export your assessment data
• How Scoring Works — how behavioural data becomes a risk score