Intelligent Expense Management

The Problem

The company was facing two converging problems at once.

First, their CFO was retiring. He had spent years manually assigning general ledger codes to credit card transactions — a process that lived entirely in his head. When he left, that institutional knowledge walked out with him. The company needed a way to automate what he had been doing, consistently and accurately, without having to hire a replacement whose sole job was categorizing expenses.

Second, the COO had inherited a sales team with almost no financial oversight. Historically, the team could spend company money with little accountability — there was no systematic way to see what was being purchased, by whom, or whether it made sense. The COO described it as flying blind. He knew money was going out but couldn't see where it was landing.

Neither problem was dramatic on its own. Together, they represented a real operational and financial risk.

---

The Build

The system has several layers working together.

GL Code Assignment — transactions are analyzed against the company's chart of accounts and assigned codes automatically. The AI doesn't just pattern-match on merchant names; it reasons about the nature of the purchase, the context, and the account structure to make the right call. Edge cases get flagged for human review rather than forced into a wrong category.

Outlier Detection — the system monitors spending patterns and surfaces anomalies. An unusually large transaction, a merchant that doesn't fit historical patterns, a spike in a particular category — these get flagged so the COO can investigate rather than discover problems months later.

Trend Reporting — instead of a static monthly report, the system maintains a running view of how money is moving across categories, teams, and time periods. Patterns that would be invisible in a spreadsheet become visible when everything is being analyzed continuously.

Receipt Capture and AI Parsing — sales team members can upload receipt images. The system reads them using AI image analysis, extracts line-item details, and — critically — parses out state sales tax paid on each transaction. This turned out to have significant tax implications.

---

The Result

The COO now has something he didn't have before: visibility. He can see exactly how his sales team is spending company money, down to the line item, without asking anyone or waiting for a monthly reconciliation.

The GL assignment that previously required a dedicated person runs automatically. The retiring CFO's institutional knowledge has been encoded into a system that applies it consistently, every transaction, without fatigue or inconsistency.

The receipt parsing and sales tax capture created an unexpected but substantial benefit. Previously, when the state conducted a tax audit, the company had limited documentation to prove what sales taxes had already been paid. Without proof, they'd be assessed — and they'd pay. Now the system maintains a complete, searchable record of every transaction where state sales tax was paid and how much. Audit exposure that previously represented real financial risk has been systematically eliminated.

---

The Pattern

This project is a good example of a build that starts as one thing and becomes something bigger. The original ask was "automate GL coding so we don't lose this when our CFO retires." The result was a full financial intelligence layer — expense oversight, anomaly detection, trend visibility, and tax documentation — that the company didn't know it needed until it had it.

That's common. When you remove a manual bottleneck, you often discover that the bottleneck was hiding other problems you'd learned to live with.