← Back to the fleet FTE-04 · AI Cost Controller

OperaNova Auto-FTE

The back-office employee who never opens a report — because it already read them all. Five live anomaly detectors screen every delivery and flag the costly mistakes with the rupee exposure attached, pushed to WhatsApp and email without anyone visiting a dashboard.

Strong MVP F&B back-office ERP 5 detectors AI-native 3/5 FTE-fit 4/5

Portal scope

113routes · 37 migrations

Detectors

5Reactive + proactive

Alerts

₨ exposureAttached to each flag

Push

2 channelsEmail + WhatsApp

What runs today vs. the roadmap

Running today

Integration-test-verified against real data.

✦Auto-FTE anomaly agent — 5 detectors: wrong-unit/over-receipt, receiving price-spike vs weighted-average cost, theoretical-vs-actual usage variance, sales anomaly, and loss-making margins.
✦Reactive + proactive — screens every stock-in receipt before approval and sweeps usage/sales/margins; findings carry severity + estimated exposure, deduped and persisted.
✦Out-of-band alerts — branded HTML email + WhatsApp, surfaced on the dashboard feed. "Detected automatically… no one had to open a report."
✦Recipe Advisor with agentic cross-consultation — a chef-bot and a finance-bot consult each other, grounded in real BoM, costing workbook, SOPs and 12 months of sales.
✦Full back-office underneath — menu, inventory with an append-only stock ledger, double-entry accounting, loyalty, HR, reports, RBAC and audit on every mutation.

On the roadmap

From advisory to a true 24/7 employee.

◆Wire a real scheduler — the single change that makes the proactive scan literally 24/7 (today it runs from a manual script).
◆Close the loop — let the agent auto-hold a flagged receipt, draft a re-price, or raise a Recipe Change Request behind human approval (today it alerts but doesn't act).
◆Multi-tenant + hosted demo — onboarding ETL so a new chain's menu/costing/sales feed the agents; per-tenant notification config.
◆Trust layer — cost metering, an answer-quality eval set, and an audit link from each AI suggestion to the change it drove.

Honesty note: the AI is real but narrow — ~110 of 113 routes are classic CRUD, and the agent is advisory (it flags, a human acts). Payroll is intentionally parked.

The role it replaces

A junior cost-controller.

The manual report-pulling and reconciliation a food-cost / inventory-audit seat does — catching the wrong-unit receipt at the door, the price creep, the margin leak — running continuously instead of monthly.

120–160hrs/month of audit work

The 60-second showcase

veloce://fte-04 — receiving gate

Receive 10,000 cases of a sachet — a fat-finger for 10,000 pieces.

The agent instantly flags a critical over-receipt — "12+ months of stock in one delivery" — with the implied loss in rupees.

A branded WhatsApp + email fire and it drops on the dashboard feed — live, in under a second. No one had to open a report.

OperaNova is the most literal embodiment of the thesis in the fleet — an agent the team already calls the "Auto-FTE," named for the role it replaces, alerting out-of-band with quantified impact. The path to flagship is short and concrete: a real scheduler so it's truly always-on, and a guarded action loop so it can hold the suspicious receipt, not just point at it.