Treasury Management • Colab • 50-75 min

Invoice-Level Collections Prediction

Launch a Colab notebook that builds invoice-level late-payment risk features, trains models to predict both late-payment risk and expected payment date, ranks collection priorities, and exports a calendar-ready expected cash inflow feed — shown both without AI (naive due-date forecast) and with AI (regressor-adjusted forecast) side by side. Includes a downloadable data template so you can run it on your own ERP invoice export instead of synthetic data.

Invoice-level feature engineering Collections prioritization Calendar-ready cash inflow forecast Without AI vs. with AI comparison Bring-your-own-data template Colab-ready Python workflow
Run in Colab Open About Demo Read README

What this demo is about

Students move beyond a browser simulation and inspect how payment history, invoice terms, dispute signals, channel mix, and seasonality can be turned into late-payment risk features — and now, into a predicted payment date per invoice, rolled up into a day-by-day expected inflow calendar.

Why Colab helps here

The notebook keeps the data generation, feature engineering, model scoring, and ranking logic visible, so the class can critique the modeling assumptions instead of only consuming a final recommendation.

Run Modes and Expectations

Colab

Best for a classroom walkthrough with zero local setup and visible Python cells for each modeling step.

What students do

Generate synthetic invoices (or upload their own ERP export using the provided template), engineer risk features, train a late-payment classifier and a days-late regressor, score invoices, export a calendar-ready expected inflow feed, and estimate cash release from better DSO discipline.

Wait expectation

The notebook is lightweight by design. Most delay comes from package startup and first-run Colab environment loading.

How to use it

  1. Open the notebook in Colab and run the setup cell.
  2. Optional: download the data template, fill it with your own ERP invoice export, and upload it back in instead of using synthetic data.
  3. Review the synthetic (or uploaded) invoice data and ask which features should matter most.
  4. Run the model section and compare predicted late-payment risk with invoice amount and customer context.
  5. Use the priority table to discuss which invoices should receive immediate collections attention.
  6. Compare the "without AI" and "with AI" calendar charts and the near-term dollar difference between them — decide which one you'd actually trust for short-term cash planning.
  7. Export the calendar feed as CSV (or a Google Calendar-style import file).
This is a teaching notebook, not a production collections model. The point is to make feature choice, model behavior, and trade-offs visible enough for classroom critique.