Know

Anomaly Detector agent

Checks nine business figures for unusual movements whenever an administrator or your scheduler starts a scan, and lists them in the Anomaly Feed, most serious first.

Validated with each client

Short answer

What does Anomaly Detector replace? The work of a finance manager or CA: reviewing the books for unusual movements. Checks nine business figures for unusual movements whenever an administrator or your scheduler starts a scan, and lists them in the Anomaly Feed, most serious first.

Shows why

Every draft comes with its reasons and how sure the agent is. If it is unsure, it only offers a suggestion.

Waits for you

It does not file, pay or post anything until a person says so.

Gets reviewed

Your approvals and corrections are recorded, so you can see how well it does on your books.

A day with Anomaly Detector

A typical week for an owner or finance head who wants to hear about surprises before the month-end review.

Kinnow keeps nine figures under watch. Gross profit percentage, collection efficiency, stockout rate, debtor days, creditor days, stock turnover, dead stock percentage, on-time delivery and the share of agent tasks that resolve without manual work.
A scan compares now with normal. An administrator starts it, or a scheduler you arrange with us does. For each figure it takes the last seven days and compares them with that figure's usual range over the previous 30 to 90 days, depending on the figure.
Only real departures are flagged. Most figures are flagged when they sit about two standard deviations from their usual level (a little more for the noisier ones, such as gross profit). A figure with no data for the week is flagged as missing data.
Each flag gets a severity and a sentence. Severity runs from low to critical, depending on how far the figure has moved. A short explanation says what moved, by how much, and what to look at. If the AI cannot be used for that sentence, a plain template states the same numbers.
The Anomaly Feed lists them. Flags appear most serious first. Analytics (Know) landing screen shows the top three under "needs your attention", and the feed has the rest.
A person reviews and closes each one. Your finance head looks at the cause, acts, and marks the flag resolved with a written reason. Kinnow does not repeat the same flag for seven days unless the figure gets worse.

See what a scan finds on your own numbers. Send us a month of sales, collections and stock figures.

What it reads and what it hands back

What it reads	What it hands back	Where it lands
Nine business figures, built from your sales, collections, purchase and stock records	Whether each is out of its usual range, and by how much	Anomaly Feed in Know
The last 90 days of the same figures, as the normal level	Direction (up or down), severity from low to critical	Each flagged item
Figures with no data for the week	A "no data" flag, so a stopped feed is not mistaken for a quiet week	Anomaly Feed

What a person decides

Nothing is filed, paid or posted without a person. For flagged movements that means:

You decide whether a flag matters. A seasonal sale or a one-off large order can move a figure honestly. Resolve it and write down why.
You act on the cause. The agent suggests where to look. It does not change a record, hold a payment or send a message.
You decide how often to scan. A scan runs when an administrator starts it, or on a schedule you arrange when it is set up.
It never does these things. It does not correct a figure, reverse an entry or block a user.

Ask Kinnow about this

Ask Kinnow shows its source and cost before it runs, and says so when a question is outside the records it can read. We confirm which of these work on your data in the demo.

You could ask	The answer shows
What changed unusually in the last week?	The flagged figures, most serious first, with how far each moved from its usual level.
Why did debtor days go up?	The flag for debtor days with the current value, the usual value and the short explanation.
Which figures have stopped updating?	Figures flagged for missing data, so you can find the feed that has gone quiet.

Ask us to run a scan on your last quarter and show you what it would have caught.

Works with

Anomaly Detector belongs to Know. These pages sit closest to it:

NL-query: ask the follow-up question in plain English.
Policy Agent: checks activity against the rules you have written down.
Every number you track, watched for surprises: the capability behind this agent.
Journal entries screened for fraud and errors: the separate check on hand-typed entries.

Questions

What does the Anomaly Detector agent do?

Checks nine business figures for unusual movements whenever an administrator or your scheduler starts a scan, and lists them in the Anomaly Feed, most serious first.

Does Anomaly Detector act on its own?

No. Anomaly Detector hands back each unusual movement it found, ranked by how serious it looks, with its reasons, and a person approves before anything is posted, filed, sent or paid.

Does Anomaly Detector learn?

Yes. Each time you approve or reject a flagged movement, Kinnow records it against how sure the agent said it was, so you can see how it is doing on your books. Your corrections are kept as feedback.

How is it paid for?

It is part of your agent pack, and there is no charge per user.

Will it flag the same problem every time it scans?

No. A figure already flagged in the last seven days is not flagged again unless it has got worse. After seven days an unresolved one comes back.

Can it catch an unusual journal entry, such as a duplicate or one posted and approved by the same person?

Those checks belong to the journal screening that sits in Account. This agent looks at business figures such as debtor days, not at single entries.

Works alongside

Know

Tell Kinnow what slows you down.

Answer three questions. Get the applications, the agents and a two-week plan for your business.

Book a walkthrough