Invoice Chaser

Inspiration

Small businesses lose countless hours — and real money — chasing late-paying clients. Manually tracking who's overdue, figuring out the right tone for a reminder, and negotiating payment plans is tedious and often gets ignored. We wanted an AI agent that could handle this entire workflow autonomously, so business owners can focus on their work instead of playing debt collector.

What it does

Invoice Chaser is an AI agent built on the Strands Agents SDK that:

Analyzes client payment risk using historical payment data to predict who's likely to pay late
Writes personalized reminder emails with a tone that adapts to the client relationship and how overdue the invoice is
Negotiates payment plans for clients who need flexibility
Forecasts cash flow so businesses can plan ahead
Generates a weekly brief summarizing outstanding invoices and recommended actions

How we built it

We used the Strands Agents SDK to build the core agent logic, with AWS Bedrock powering the underlying language model calls. The agent's decision-making modules (risk scoring, reminder writing, payment plan negotiation, forecasting) are built in Python and trained/tested against IBM's public "Finance Factoring - Late Payment Histories" dataset from Kaggle — real accounts receivable data rather than synthetic samples.

The web interface is built with Flask, with a polished dark/light-theme frontend that lets users see their invoice dashboard, risk breakdowns, and generated reminders in one place.

Challenges we ran into

Getting Bedrock model access approved on a new AWS account took longer than expected
Balancing reminder tone so it feels firm but not aggressive — especially across different client relationships
Structuring the agent so multiple decision modules (risk, negotiation, forecasting) could work together coherently rather than as isolated scripts

Accomplishments we're proud of

A fully working end-to-end agent pipeline: data in, risk-scored, reminder drafted, plan negotiated
Using real-world late payment data instead of synthetic data for meaningful predictions
A clean, usable web dashboard built by our team in a short timeframe

What we learned

Building with the Strands Agents SDK taught us how to structure multi-step agentic workflows, and working with real financial data gave us a much better sense of what actually predicts late payments versus what looks intuitive but isn't.

What's next for Invoice Chaser

Connect directly to accounting tools like QuickBooks or Zoho Books for real-time invoice syncing
Add multi-language reminder support
Expand forecasting to give longer-term cash flow projections
Let the agent auto-adjust reminder cadence based on client response patterns