Transaction Anomaly Watchdog

Catches double-charges, duplicate vendors, and fraud patterns nightly. Replaces a $60k Junior Accountant.

The Problem

An SMB owner pays $60k/yr to a Junior Accountant whose primary job is spotting the $847 charge that shouldn't be there — the duplicate vendor, the missed Stripe refund, the credit-card-skimmer test transaction. The role is 100% pattern-deviation detection — exactly what unsupervised ML libraries like PyOD do at scale.

The Outcome

Nightly 11pm → agent runs Isolation Forest + KNN + LOF over today's transactions vs your 90-day baseline → surfaces 'Top 5 anomalies tonight: charge to BLUE_VENDOR_LLC ($1,247) is 4.2σ from norm; same vendor charged 6 hrs earlier at $1,247 → likely duplicate.' All drafts; you decide whether to dispute.

Day in the Life

Nightly 11pm: pulls today's bank + card + Stripe txns → ensembles 3 PyOD detectors → ranks top 5 anomalies

On match: drafts dispute email to vendor or chargeback request to Stripe — queues for one-click approve

Weekly: writes 'fraud + duplicate scan summary' — '$0 lost this week vs $847 caught and reversed' for owner peace of mind

Deploy Specs

Runtime

python

Pattern

native-agent

Tier

medium

Setup Time

hours

Required Env Vars

ANTHROPIC_API_KEYSTRIPE_API_KEYPLAID_CLIENT_IDPLAID_SECRETQBO_REFRESH_TOKEN

Integrations

PyODStripePlaidQuickBooks Online

GitHub Source

Repository

yzhao062/pyod

Stars

9,828

License

BSD-2-Clause

Last Commit

2026-04-29

Replace

Junior Accountant

$60,000/yr

AgentDepot

$299/month

Save $56,412/yr · 16.7x cheaper

Deploy this skill →

14-day free trial · No credit card

Also Replaces

×Trustpair $1,200/mo (enterprise)

×Manual reconciliation review hours

Not Technical?

Expert Setup — $499

Our team deploys this exact skill for you — integrations connected, tested, live in your AWS within 48 hours.

Book your setup →