A vendor increased their pricing by 8% three months ago. Nobody noticed. The charge cleared the AP approval because it was within the budget line's tolerance. It wasn't large enough to flag in the monthly review. But compounded over a year, it's $48K of margin quietly leaving the business.
This is the expense monitoring problem. The leaks that kill mid-market profitability aren't dramatic overspends: they're small, persistent charges that slip below every threshold. Nobody's looking at them because nobody has time to look at every line item every month.
What Expense Monitoring Actually Involves
Expense monitoring means more than comparing actuals to budget. Budget-to-actual catches the big misses: the department that blew its quarterly allocation by 30%. But the expensive problems are subtler.
Vendor pricing creep: a supplier gradually raises rates over successive invoices, staying just below the threshold that triggers a review. Orphan subscriptions: a SaaS tool that someone signed up for during a trial, the trial ended, and the $200/month charge has been running for a year with no active users. Structural cost drift: a department's spending grows 15% while its revenue contribution grows 5%, slowly eroding its margin contribution.
Catching these requires pattern analysis across time, not a single month's snapshot, but a trailing comparison that shows when something changed and whether the change is accelerating.
How It's Done Today
It mostly isn't. The monthly budget review catches line items that are significantly over plan. Everything else passes through. Nobody has the bandwidth to review every vendor invoice against its historical pricing, check every subscription against user activity logs, or compute each department's cost-to-revenue ratio on a monthly basis.
Some teams do a quarterly deep-dive into expenses. This takes 1-2 days of analyst time and typically uncovers 3-5 issues that have been compounding for months. The savings are real but the effort is manual and inconsistent: it happens when someone remembers to do it, not when it needs to be done.
The result is a predictable pattern: spend inefficiencies accumulate quietly between reviews, get caught eventually, and the conversation is always "how long has this been happening?"
Why General AI Can't Monitor Expenses
Expense monitoring is an ongoing process, not a one-time question. You can't paste six months of GL entries into ChatGPT every day and ask it to find anomalies. It can't remember what was normal last month. It can't track vendor pricing trends over time. It can't cross-reference a subscription charge against user activity in another system.
Even for a point-in-time analysis, the accuracy problem is critical. A false positive, flagging a legitimate expense as an anomaly, wastes the CFO's time. A false negative, missing an actual issue, lets the leak continue. General AI's tendency to hallucinate makes both scenarios common enough that the output can't be trusted without manual verification of each finding.
What Proper Monitoring Looks Like
Effective expense monitoring is always-on, not periodic. It needs to:
- Compare every vendor charge against its 6-month trailing average and flag deviations
- Identify recurring charges that don't match an active contract or PO
- Track department-level spend growth relative to revenue contribution
- Know which charges are expected (the quarterly insurance premium) vs. unexpected
- Distinguish between one-time spikes (a conference sponsorship) and structural increases (a vendor rate change)
This requires both pattern recognition and business context. The system needs to know that the $35K charge in August was a planned offsite (don't flag it) while the $2K monthly increase in the marketing agency bill was not in the contract (flag it immediately).
How DataWyse Handles It
DataWyse's Eagle Eye Agent runs a daily financial health check against your GL data. It doesn't wait for the monthly review. Every day, it compares current charges against trailing averages, contracts, and budget allocations.
When you ask "What expense anomalies should I know about this month?", you're not asking it to start looking. It's already been looking. The output is a prioritized list of findings, each with the dollar amount, the duration, whether it's a new issue or recurring, and the recommended action.
Each finding is traceable to specific GL entries, vendor invoices, and the comparison methodology used. "AWS spend up 22% MoM" links directly to the invoice line items showing the increase, the 6-month trend chart, and the note that no product launch or infrastructure change was logged for the period.
The Difference
Without continuous monitoring, a $4K/month vendor overcharge runs for six months before someone catches it: that's $24K gone. With DataWyse, it's flagged on day one.
The $1,800/month in unused SaaS subscriptions doesn't silently compound for a year. The marketing agency's billing discrepancy doesn't persist because nobody has time to audit invoices against the contract.
The shift is from periodic audits to continuous vigilance. Small leaks add up. Catching them early is the easiest margin improvement a mid-market company can make, no revenue growth required, no cost-cutting, just stopping the money that shouldn't be leaving.