The challenge
Ledgerly’s customers, small finance teams, were reconciling hundreds of bank lines by hand each month. The company wanted a premium AI tier, but its batch-based platform could not process transactions in real time, and its AWS bill was already growing faster than revenue.
What we built
We re-platformed transaction ingestion onto a streaming pipeline and built two AI features on top: a matching model that pairs bank lines with invoices and bills, with an LLM explaining each match in plain language, and an anomaly detector that learns each customer’s normal patterns and flags outliers within seconds.
- Event-driven ingestion replacing nightly batch jobs
- Explainable auto-matching with an accountant-friendly review screen
- Per-tenant anomaly baselines with tunable sensitivity
- Infrastructure as code, autoscaling and a cost dashboard per feature
The results
Customers on the AI tier close their month 3.4× faster. The new tier lifted average revenue per account by 18%, and the re-platforming cut overall cloud spend by 27%.