Payment Delay Prediction¶

What It Predicts¶

Customer Payment Delay Prediction enables organizations to enhance short-term and mid-term cash flow planning by forecasting whether customer payments will occur earlier or later than their due dates.

The model predicts the expected payment delay for each customer invoice installment, expressed as a number of days. The prediction indicates how many days before or after the due date the payment is likely to be made.

Where the Prediction Appears¶

The prediction results are displayed on the following IFS Cloud page(s):

• ​ Instant Invoice – Installment Plan and Discount

• ​ Customer Order Invoice - Installment Plan and Discount

• ​ Project Invoice - Installment Plan and Discount

• ​ Accounts Receivable Lobby

• ​ Cash Plan Details

The following field is shown:

Predicted Payment Delay Days - The estimated number of days by which a payment is expected to occur before or after its due date.

Target Definition¶

Formal Definition¶

Payment Delay (days) = Actual Payment Date – (Installment Due Date or Installment Original Due Date)

Interpretation¶

Positive values indicate late payment (payment expected after the due date).

Negative values indicate early payment (payment expected before the due date).

Sign Convention¶

> 0 → Paid after due date

< 0 → Paid before due date

Examples¶

Due 10 March, paid 15 March → +5 days

Due 10 March, paid 6 March → –4 days

Availability and Prerequisites¶

AI Services and Model Availability¶

• IFS.ai services are enabled.
• The Payment Delay Prediction model is trained and activated.
• Sufficient historical invoice and payment data is available to support reliable predictions.
• The feature is enabled for the specific customers for whom predictions are to be generated.

Although not mandatory, it is highly recommended to define the parameter values prior to training to ensure that the relevant data is considered during the training process.

Data Access and Security¶

• Access follows standard IFS Cloud role-based security.

• Customer data is isolated per tenant and not shared across customers.

• Users must have access to PAYLED and INVOIC components. Further, if you wish to see the predictions in cash plan details CSHPLN component should also be available.

Data Availability and Quality¶

Prediction quality depends on the volume of historical payment data, as well as the completeness, accuracy, and consistency of that data. The presence of significant outliers may also impact model performance. Defining appropriate parameters at the time of training generally results in improved prediction accuracy.

If the ‘INVOICES_FROM_DATE’ parameter is not specified by the user, the system automatically filters the training dataset to include only invoices created within the last 48 periods.

Only include invoices/installments that meet all of the following conditions:

• The invoice gross amount is greater than zero

• The installment amount is greater than zero

• The payment date is on or after the payment term base date

• The installment due date is on or after the payment term base date

• The original installment due date is on or after the payment term base date

• The invoice date is not in the future

• The payment date is not in the future

• The invoice date is within the last 48 months → If a “from date” is provided, use that → Otherwise, include invoices from the past 4 years only

• If “Exclude Internal Customers” is NOT enabled (the option is set to FALSE or not provided): → Include all customers, both internal and external.

Prediction Execution¶

Trigger¶

The prediction process is triggered once the relevant invoice data is available, and the model has been trained and activated, and this feature is enabled for the particular customer that the invoice is made. Predictions are generated automatically when an invoice transitions to the PostedAuth status.

If an error occurs during inference, a null value is recorded for the predicted delay.

Prediction Drivers¶

The model is trained using the customer’s own historical invoice and payment data through the IFS Machine Learning Service. This ensures that predictions are tailored to the organization’s specific payment patterns and customer behavior.

Operational Considerations¶

Model Behavior¶

Predictions are generated only when the model is trained and Model Status is Active.

Retraining is supported to maintain and improve prediction accuracy as new invoice and payment data becomes available. This enables the model to continuously adapt to changes in customer payment trends and business conditions.

Known Limitations¶

• Predictions are advisory

• Prediction quality depends on data volume, completeness, correctness, and outliers in the data may adversely impact the accuracy of prediction results.

• Process changes may require retraining. Predictions are advisory

• Prediction quality depends on data volume, completeness, correctness, and outliers in the data may adversely impact the accuracy of prediction results.

• Process changes may require retraining.

Troubleshooting¶

If predictions are missing or incorrect, verify:

• AI services and model status

• Data availability and quality conditions

• Feature enabled on the customer

• Data access and permissions

Use the Machine Learning Logs page in IFS Cloud for diagnostics.