Toolkit Store Courses Take Assessment About Blog
Take Assessment

Predict. Prioritize. Collect.

AR Late Payment Predictor

From AR data to risk-ranked actions

Get the AR Late Payment Predictor

Features List

8

Structured Outputs

2

Ready-to-Use Model Playbooks

1500

Invoice Sample Dataset

Why AR and Collections Teams Use This Prompt Book

Risk Score Before Outreach

Every open invoice gets a probability score for paying 30+ days late. The Actions tab ranks invoices by risk score (probability x amount), so your collections team contacts the right accounts first, not the noisiest ones.

Two Model Playbooks Included

Logistic Regression and Gradient Boosting β€” two production-ready Excel workbooks built from the prompts. Use Logistic Regression for interpretability and simpler datasets. Switch to Gradient Boosting when you need higher accuracy on complex patterns.

Daily Collections Queue

The Actions tab outputs a prioritised list of who to contact today, based on your team's daily capacity. Contact_Today = YES if the invoice is flagged and within your set outreach limit. No more manual prioritisation on Monday morning.

Built-In ROI Calculator

The ROI tab calculates expected benefit, cost, and net value per invoice using your recovery lift rate, recovery percentage, and contact cost. Adjust the inputs and the entire playbook recalculates. You can quantify the return on every collections action before you make it.

Prompts That Teach

The 8-prompt sequence doesn't just produce outputs. Prompt 4 explains in plain business terms which model type fits your AR problem. Prompt 2 suggests the additional features you can build. You leave with a working model and an understanding of why it works.

Everything Inside AICFO-029

  • Prompt Book PDF (8 Prompts)

    The core product. A structured ChatGPT prompt sequence that takes your AR dataset from raw invoices to a scored, ranked collections queue. Each prompt includes the exact text, what files to upload, expected output, and the decision you need to make before moving to the next step.

  • AR Late Payment Playbook (Excel)

    The Logistic Regression model output. 9 tabs: Target Definitions, Features Used, Model Training Summary, Evaluation Dashboards, Customers Expected to be Late, False Positives, False Negatives, ROI & Ops Inputs, and Actions (Daily Collections Queue). Fully formula-driven ROI tab.

  • AR Gradient Boosting Playbook (Excel)

    The Gradient Boosting model output. 10 tabs β€” same structure as the Logistic Regression playbook plus a Guide tab with detailed instructions. Actions tab includes prob_late_30d, risk_score, Contact_Flag, Rank_by_Risk, Contact_Today, and Exp_Net per invoice.

  • AR Feature Specification (Excel)

    A structured feature spec sheet listing every predictive feature: Historical Late Ratio, Rolling Average DPD (3/6/12 months), Prior Overdues, Dispute Frequency, Trend and Volatility metrics, Aging Bucket, and Balance Exposure. Includes calculation logic and business rationale for each feature.

  • Synthetic AR Dataset (Excel, 1,500 invoices)

    A ready-to-use sample dataset with 1,500 invoice records across Enterprise, Mid-Market, and SMB segments. Includes stress-tested overdue scenarios. Use it to run the full prompt sequence before loading your own data.

  • Data Dictionary (PDF)

    Full field-by-field documentation for the AR dataset: data types, null rules, example values, and derivation logic for key fields including DPD, Aging Bucket, Balance, and Status.

  • Logistic Regression Playbook Guide (PDF)

    Step-by-step instructions for reading and using the Logistic Regression playbook: target definition, feature selection, model training, threshold calibration, ROI inputs, and best practices for monthly retraining.

  • Gradient Boosting Playbook Guide (PDF)

    Same structure as the Logistic Regression guide but for the Gradient Boosting model. Covers the ROC-AUC evaluation dashboard, threshold selection at 0.60 default, daily capacity input, and how to interpret the Exp_Net calculation per invoice.

Β What Each Prompt Does β€” and What You Get Back

Prompt: Customer Profiling

What You Ask ChatGPT:Β Upload AR dataset and data dictionary. Ask ChatGPT to calculate average days late, % invoices paid late, and whether each customer currently has open invoices.

Output You Receive: Per-customer payment behaviour summary. Your baseline before building any model.

Prompt: Feature Engineering

What You Ask ChatGPT: Ask ChatGPT to suggest additional predictive features you can derive from the AR dataset to improve late payment and default forecasting.

Output You Receive: A list of engineered features with construction logic: rolling DPD averages, historical late ratio, dispute frequency, trend metrics.

Prompt: Feature Spec Sheet

What You Ask ChatGPT: Ask ChatGPT to draft the feature specification as a downloadable Excel file.

Output You Receive: The AR Feature Specification workbook β€” one row per feature, with calculation logic and business rationale.

Prompt: Model Selection

What You Ask ChatGPT: Ask ChatGPT which model type fits AR payment prediction better: binary classification (late/not late) or regression (days late). Request a plain-language business explanation.

Output You Receive: A clear recommendation with the trade-offs of each approach explained for an AR or collections audience.

Prompt: Logistic Regression

What You Ask ChatGPT: Ask ChatGPT to write Python code training a logistic regression model on 30+ day late payments, then generate the full 9-tab playbook as a downloadable Excel file.

Output You Receive: AR Late Payment Playbook: scored invoices, collections queue, ROI calculator, false positive/negative analysis.

Prompt: Gradient Boosting

What You Ask ChatGPT: Extend Prompt 5 to generate a Gradient Boosting model and playbook using the same feature set. Ask ChatGPT to add a Guide tab with usage instructions.

Output You Receive: AR Gradient Boosting Playbook: same 9 tabs plus a Guide, with higher accuracy predictions and risk_score ranking.

Prompt: Threshold Review

What You Ask ChatGPT: Optional. Ask ChatGPT to explain how to adjust the probability threshold in ROI & Ops Inputs, and what happens to false positives and false negatives as you move it.

Output You Receive: A plain-language explanation of threshold trade-offs with a worked example for your specific collection capacity.

Prompt: Retraining Checklist

What You Ask ChatGPT: Optional. Ask ChatGPT to produce a best practices checklist for model retraining, drift monitoring, and expanding features for future model runs.

Output You Receive: A retraining and model maintenance checklist. Covers monthly cadence, drift signals, and feature expansion ideas.

What Each Tab in the Model Playbook Does

Tab Name What It Does
1 Target Definitions Defines the prediction target: Target_Late30 = 1 if the invoice was paid 30+ days late, or is unpaid with DPD > 30. Sets the rules for training vs. scoring split.
2 Features Used Lists every feature in the model with its calculation method. Historical metrics use only data prior to each invoice date to prevent data leakage.
3 Model Training Summary Records model type, training data summary, and key performance metrics: ROC-AUC, accuracy, precision, recall, F1, and confusion matrix.
4 Evaluation Dashboards Interactive threshold selector. Adjust Risk_Threshold to shift the balance between precision and recall. Displays the full performance curve at your chosen threshold.
5 Customers Expected to be Late Aggregates risk by customer: invoices at risk, average probability, and total open amount. Use for account-level prioritisation alongside invoice-level actions.
6 False Positives Invoices flagged as high risk that paid on time. Review to identify patterns causing over-flagging. Helps calibrate the threshold over time.
7 False Negatives Invoices not flagged that paid late. Review to identify missed patterns β€” new customers, unusual terms, or segment-specific risks the model didn’t catch.
8 ROI & Ops Inputs The control panel. Set your probability threshold, daily contact capacity, recovery lift, recovery percentage, discount rate, and contact cost. All other tabs reference these dynamically.
9 Actions / Daily Queue Today’s prioritised collections list. Contact_Flag = 1 if probability β‰₯ threshold. Rank_by_Risk orders flagged invoices. Contact Today = YES for the top N within daily capacity. Exp_Net calculated per invoice.
10 Guide (Gradient Boosting only) Full usage guide explaining every tab, how to read evaluation metrics, how to adjust inputs, and how to use the model for credit policy and treasury cash flow forecasting.

Common Questions

Know Which Invoices Will Pay Late Before They Do.

Updated for 2026. 8 prompts, 2 model playbooks, and a daily collections queue β€” built in ChatGPT.

Get the AR Late Payment Predictor