Anomaly detection on every journal entry. Variance attribution by operational lever. Root-cause walks down to the specific transactions and locations that drove the miss. Every finding comes with an AI explanation grounded in your driver history.
The platform continuously watches your ledger for movements outside expected envelopes. An anomaly is a journal-entry rollup that deviates more than 2σ from the trailing twelve months of the same account, period, and location combination. The detection runs nightly after close; the high-severity findings hit your dashboard before your controller hits coffee.
Each finding lands with the expected value, the observed value, the z-score, and an AI-written one-sentence explanation that references the driver history that contextualizes it.
When a scenario or budget has periods that have now elapsed, the platform compares projected against actual and decomposes the variance — not by account, by operational lever. Each variance walks lever and driver mappings backward to identify the most likely lever that explains the miss.
The variance attribution engine treats Scenarios and Budgets uniformly through a shared interface. Whether you set the target in FP&A or in the planning suite, the attribution logic is identical.
For each variance, the engine walks lever and driver mappings backward to identify the most likely lever. When multiple levers map to one account, the attribution uses observed driver values: enrollments grew 8%, so most of the revenue variance attributes there.
A table of variances ranked by absolute miss, each with its inferred lever and a one-paragraph CFO summary. Drill any row into the EBITDA bridge to see the waterfall, or into the root-cause walk to descend into the specific entries.
A guided drill from a variance down to the journal entries that drove it. The starting point is a metric. The destination is a list of transactions. The walk reuses the report service and account service for the balance queries at each step — the novelty is the navigation and the AI-written summary that contextualizes each level.
Once time has passed, the platform measures how well saved scenarios predicted what actually happened. Per scenario, per metric, per period: projected, actual, signed error, rolling MAPE. The accuracy page surfaces aggregate AI-vs-manual performance so operators see whether AI scenarios are out-performing their hand-built ones — and the answer feeds back into the next AI proposal.
This is the loop. Drivers whose AI projections were persistently wrong get larger uncertainty in their next envelope. The model becomes better at proposing what's plausible because the platform has watched it propose, watched the operator execute, and watched reality respond.
Connect your books in an afternoon. See the diagnostics and the first scenarios on your real data, the same week. No implementation fees, no six-month rollout, no SOW.