AR Collections & Cash Application - Projects

The business problem

With thousands of overdue invoices, collectors need to know whom to call first and what to say. The target metric is days sales outstanding.

Proposed solution

The crew builds each account’s picture, for example: “Customer ABC owes $487K, $212K over 60 days; it usually pays in 45 days; two invoices are disputed; contact about the undisputed $315K now.” It then drafts outreach and records the activity.

A Cash Application Agent matches bank receipts to open invoices.

Workflow

Collections

Analyze Overdue Receivables
Prioritize Accounts by Risk
Gather Customer and Dispute Context
Categorize Disputes
Draft Collection Message
Obtain Collector Approval
Record Collection Activity

Cash application

Ingest Receipts and Remittances
Match Invoices with Confidence Score
Route Exceptions for Review
Apply Cash Within Approved Limits
Record Audit Trail

Agents and responsibilities

Prioritization: aging, payment history, credit status and promises to pay combined into a risk score.
Customer Context: RAG over contracts, collection notes and correspondence.
Dispute Triage: pricing, quantity, delivery and duplicate disputes, routed to an owner.
Outreach: reminders, follow-ups and escalations.
Cash Application: identifies the customer, matches invoices with a confidence score, and sends exceptions for review.

Oracle ERP integration

Oracle objects: Customer accounts, receivables transactions, receipts, adjustments, collection strategies and promises to pay.

Shared Oracle integration: a Python MCP server exposing typed read tools (Fusion Cloud REST resources, BI Publisher/OTBI reports, read-only SQL views for E-Business Suite) and a small set of write tools that execute only after human approval.

Technology stack

Project stack: OpenAI Agents SDK/GPT and Claude Agent SDK/Claude (benchmarked) · CrewAI + MCP · Google ADK/Gemini

Orchestration: CrewAI crews define role-based agents, tasks and hand-offs. Oracle tools attach through CrewAI MCP server adapters, so every agent calls the same governed tool layer.
Agent runtimes: Claude Agent SDK with Claude is the primary build and reasoning runtime. It uses subagents for deep investigation, MCP for tool access, and pre-tool-use hooks that block any ERP write lacking an approval token. The same agents run on Google ADK with Gemini and OpenAI Agents SDK with GPT through a common model-adapter interface, switched by configuration.
Engineering: Python, Pydantic-validated agent outputs, a pytest scenario-evaluation harness comparing Claude, Gemini and GPT on accuracy, latency and cost, OpenTelemetry tracing and Docker, built in VS Code with Claude Code.

Controls and safeguards

Credit holds, write-offs and adjustments require approval.
Emails are sent only after a collector approves them.
High-confidence matches auto-apply only within configured amount limits.
Grounding and control (shared platform): RAG over policies, contracts and approval matrices in Oracle Database 23ai AI Vector Search; a deterministic rule engine for tolerances, thresholds and SoD; an immutable audit log of evidence, tool calls, model, approver and outcome.

Validation plan

Scope: 150 remittances against 300 open invoices.

Targets

Reference build in progress. The figures below are validation targets for the build, not measured results.

90%+ Auto-match on clean remittances
85%+ Correct dispute category