> ## Documentation Index
> Fetch the complete documentation index at: https://docs.stockful.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Anomalies

> Stockful scans sales data daily and flags SKUs whose velocity has shifted out of pattern, each with a short AI-written explanation and likely cause.

Stockful scans your sales data every day and flags SKUs whose velocity has changed in a way that doesn't fit their usual pattern. Each anomaly comes with a short AI-written explanation of what happened and a likely cause.

<Frame>
  <img src="https://mintcdn.com/ljsstudios/TUCamDUIjkRf5Okj/images/user-guide/ai--anomalies.png?fit=max&auto=format&n=TUCamDUIjkRf5Okj&q=85&s=9ac129d24d1c649df830155508303422" alt="The anomalies list" width="2490" height="722" data-path="images/user-guide/ai--anomalies.png" />
</Frame>

<Note>
  Anomaly detection runs on all plans. Turn it off any time under **Settings** → **AI**.
</Note>

## What gets flagged

The daily scan compares each SKU's recent sales against its own baseline (the rolling average + variability) per location. When today's velocity falls far enough outside the normal band, Stockful records an anomaly.

Anomalies have a **direction**:

* **Drop** - sales fell sharply versus the baseline
* **Spike** - sales jumped sharply versus the baseline

And a **severity**, based on how far outside the normal band the change sits:

| Severity   | Meaning                                                           |
| ---------- | ----------------------------------------------------------------- |
| **High**   | Big move that's almost certainly real - worth investigating today |
| **Medium** | Notable change worth a look this week                             |

Anything smaller than the medium threshold is treated as noise and isn't recorded.

## Where you see them

* **`/anomalies` page** - the dedicated list, filterable by location, severity, and direction. Each row shows the SKU, the observed vs. baseline numbers, the AI narrative, and a link to the inventory detail page.
* **Inventory detail page** - if the variant you're looking at has an open anomaly, a banner appears at the top of the page with the narrative and a **Dismiss** button.
* **Home dashboard** - a compact card on the home page surfaces the most recent unresolved anomalies so they don't get lost.
* **Slack or email** (optional) - set up a **Sales spike or drop** alert and anomalies are pushed to you like any other alert.

## Getting anomalies pushed to you

Anomalies are an [alert](/user-guide/alerts/index) trigger, so they use the same alerts everything else does - with the cooldown, quiet periods, per-SKU snooze, scope and history that come with them.

1. Go to **Alerts** and click **Create alert**
2. Choose the **Sales spike or drop** trigger
3. Set the **Severity** - **High only** (default) or **Medium and above**
4. Set the **Direction** - **Spikes and drops** (default), **Drops only**, or **Spikes only**
5. Choose where it goes and how often under **Delivery**, and which SKUs and locations to watch

Each message includes the SKU, the size of the change, and the AI narrative. Anomaly alerts are checked once a day, in the morning, after the overnight scan.

<Note>
  Anomaly detection has to be on for these alerts to have anything to send. If **Enable Stockful AI** is off in **Settings → AI**, the alert form says so.
</Note>

## Dismissing an anomaly

Anomalies stay open until you dismiss them. Click **Dismiss** on the inventory-detail banner or use the dismiss action on a row in the anomalies list. Dismissed anomalies are excluded from cards and alerts but stay in your history (filter the list by **Dismissed** to see them).

## What the AI narrative does

Anomaly detection itself is statistical - it doesn't need AI to fire. The AI step writes a short explanation: looks at the SKU's product type, vendor, recent events (price changes, restocks), and suggests a likely cause ("matches a Black Friday lift across the catalog", "stockout in the prior week may explain the drop", etc.). When the model isn't confident, the narrative says so rather than guessing.

## Plan availability

Available on all plans.

## Troubleshooting

* **No anomalies for a SKU** - the baseline needs enough recent sales history to compute. Brand-new SKUs won't generate anomalies until they have enough data.
* **Too many alerts in Slack** - raise the severity threshold to **High severity only** and/or switch the direction to **Drops only**.
* **Anomalies stopped appearing** - check **Settings** → **AI** that **Enable Stockful AI** is on. If it's off, no scan runs and no anomalies are recorded.