Predicted risk bands
How customers are distributed across the fixed reference bands.
How to read these bands
These bands describe relative predicted risk. They are not action thresholds and do not change queue selection.
PORTFOLIO INSIGHTS
Portfolio overview
A clear view of predicted risk and review capacity for your customer portfolio.
NEW ANALYSIS
Start with a customer portfolio
Upload a prediction-time CSV, set your review capacity, and score the full portfolio.
AT A GLANCE
Portfolio snapshot
Summary of model estimates for the full uploaded portfolio.
Review capacity
How the requested capacity translated into the queue.
0%
of target selected
Interpretation: Predicted probabilities summarize model estimates, not observed outcomes or expected losses. Portfolio patterns are descriptive and do not show that an intervention will change outcomes.
COHORT EXPLORATION
Explore portfolio
Filter scored customers and compare descriptive groups.
Find a cohort
Every filter applies to the full scored portfolio. Counts below show the filtered denominator.
Predicted-risk distribution
Fixed 10 percentage-point bins; shares use the filtered cohort.
Compare groups
Descriptive summaries for the filtered cohort.
| Group | Customers | Share | Mean risk | Selected |
|---|
For numeric fields, groups use quartile boundaries from the full portfolio.
Matching customers
Customers are ordered from highest to lowest predicted risk. The table shows a page at a time.
| Rank | Customer ID | Predicted risk | Risk band | Review status | Details |
|---|
Page 1 of 1
Interpretation: Filtered groups describe this uploaded portfolio. They do not establish causes, treatment effects, or fairness conclusions. Encoded input values are shown as supplied; category codes are not given inferred labels.
CAPACITY DECISION
Review queue
How your review capacity shaped the ranked customer list.
How selection works
A consistent selection rule determines who enters the review list.
- Rank by predicted risk.Customers are ordered from highest to lowest predicted default probability. Equal scores receive the same rank, so later ranks account for all customers ahead.
- Keep equal scores together.Customers with the same risk score enter the review list together when the group fits within your capacity.
- Stop at a boundary tie.If the next tied group would exceed the remaining target, that whole group and all lower-scoring groups are left out.
To change capacity: edit Review capacity above and click Run analysis. This queue shows the result from the last completed run.
Why customers were selected or left out
These totals explain why customers entered the review list or remained outside it.
| Queue outcome | Customers |
|---|
Unused capacity can occur when a full tied score group cannot fit. The queue never splits that group or exceeds its target.
Selected and not selected
Compare customer profiles under the current review capacity.
Summaries use the full uploaded portfolio, regardless of filters in Explore portfolio. Each measure shows the median, middle half of values, full range, and available and missing record counts.
Average bill and payment amounts are calculated per customer across the six supplied monthly fields. Small groups are shown with their exact count; individual records can have more influence in a small group. These descriptive differences do not show causes, outcomes, or intervention effects.
Selected customers
Customers are listed from highest to lowest predicted risk.
| Rank | Customer ID | Predicted risk | Selection reason | Details |
|---|
Page 1 of 1
Interpretation: This view explains a model-ranked review policy under the requested capacity. It does not recommend an intervention or estimate the effect of reviewing a customer.
PORTFOLIO PATTERNS
Feature insights
What appears in this uploaded portfolio, alongside patterns the model used on validation data.
OBSERVED IN THIS PORTFOLIO
Historical repayment and amount patterns
Summaries describe the uploaded rows. They do not represent future behavior or outcomes.
Repayment status codes by month
Share of customers in each input code, across the full portfolio.
Codes are shown as supplied. Their integer values are not relabeled as payment outcomes here.
Bill statement amounts
Monthly distribution across all uploaded customers.
| Month | Median | Middle half | Range | Records |
|---|
Previous payment amounts
Monthly distribution across all uploaded customers.
| Month | Median | Middle half | Range | Records |
|---|
MODEL VALIDATION REFERENCE
Features associated with model performance
Global permutation importance from the validation cohort.
Top validation features
Mean Average Precision decrease after shuffling each feature.
How to interpret: These validation results describe how the model performed when a feature was shuffled. They do not show that a feature causes default or explain an individual customer's score. Engineered features can share information with their source fields, so their importance values should not be added together.
Your portfolio overview will appear here
Upload a customer file above to see the portfolio summary and review-capacity breakdown.