All agents
    ExpenseAvailable now

    Expense anomaly watcher

    Looks across all expense claims in a month and lists, highest first, what is worth a look, with the evidence for each.

    How it works

    For Finance. Where the Expense policy checker looks at one claim, this looks across every claim submitted in a period: the same receipt claimed twice (same merchant, amount and date, or the same receipt file or invoice number), expenses split just under a limit, work expenses on leave days, weekends or holidays, local travel on days attendance shows a full office day, round-number patterns, spend far above colleagues in the same grade or department, and merchants or categories new to the company. Every finding is worked out from the data and shows its evidence; the AI only explains a finding in plain words when you ask. Findings are pointers to check, never accusations. When you confirm, it emails the employee a neutral question; it never rejects, approves or changes a claim. A monthly digest is available if you switch it on.

    The check itself uses no credits; about 1 credit per finding you ask it to explain.

    What it reads

    • All expense claims of the period, their lines and receipt fingerprints, and the expense policies
    • Claimants' approved leave, the company holidays and attendance check-ins
    • The company's expense lines of the 12 months before (to know what's new)

    When it starts

    • When you open it or ask the assistant about expense anomalies
    • At the start of each month, if you switched the digest on

    What it can do

    • Email an employee a neutral question about an expense (after you confirm)
    • Send you a monthly digest of last month's findings

    What it never does

    • Reject, approve, hold or change a claim

    Safe by design

    • One assistant, many agents. Ask the Employee assistant in the app or on WhatsApp; it calls the agents your company has hired that you may use.
    • Hired, not switched on. Your admin hires each agent from the catalogue and chooses who can use it.
    • Agents propose, people confirm. Anything that changes data, like a leave request, waits for a person to confirm it.
    • Same access as the user. An agent sees only what the person asking is allowed to see.
    • AI processing in India. Agents run on models hosted in India (AWS Mumbai).
    • Budgets and a hard stop. Set a monthly credit limit per agent; agents stop when credits run out.
    • Every run recorded. Admins see each run, the credits it used and what it did.

    Frequently asked questions

    What does the Expense anomaly watcher do?

    For Finance. Where the Expense policy checker looks at one claim, this looks across every claim submitted in a period: the same receipt claimed twice (same merchant, amount and date, or the same receipt file or invoice number), expenses split just under a limit, work expenses on leave days, weekends or holidays, local travel on days attendance shows a full office day, round-number patterns, spend far above colleagues in the same grade or department, and merchants or categories new to the company. Every finding is worked out from the data and shows its evidence; the AI only explains a finding in plain words when you ask. Findings are pointers to check, never accusations. When you confirm, it emails the employee a neutral question; it never rejects, approves or changes a claim. A monthly digest is available if you switch it on.

    Can the Expense anomaly watcher change data on its own?

    No. It asks a person to confirm before it will: email an employee a neutral question about an expense. It never does the following: reject, approve, hold or change a claim.

    How much does the Expense anomaly watcher cost?

    The check itself uses no credits; about 1 credit per finding you ask it to explain.. Credits are a paid add-on to any flowwork.ai plan at ₹100 (US$ 1.19) per 1,000 credits, and new companies get 500 free trial credits for 30 days.

    Where is the data processed?

    Agents run on AI models hosted in India and see only what the person using them is allowed to see. Admins see every run and the credits it used.

    Other agents