--- title: "Fraud detection" description: "Screen every order for fraud, decide on risky orders and tune the thresholds and rules the screening uses." url: "https://docs.poltura.com/en/apps/fraud-detection/" lang: en alternates: en: "https://docs.poltura.com/en/apps/fraud-detection/" hu: "https://docs.poltura.com/hu/apps/fraud-detection/" --- # Fraud detection Screen every order for fraud, decide on risky orders and tune the thresholds and rules the screening uses. - Where: Apps → Fraud Detection Fraud detection checks every placed order. An AI score and your own rules give the order a risk score from 0 to 100. Orders with a low score pass on their own. Risky orders wait for your decision. The app suits any store that takes online payments and wants a human look at suspicious orders before goods leave. ## Before you start - The app comes installed on new stores. - The AI check uses your store's [AI credits](https://docs.poltura.com/en/assistant/ai-credits/). When credits run out or the AI is unavailable, the rules keep scoring orders and the review says why. - To get risky orders in a chat with Approve and Refund buttons, set up the [Telegram](https://docs.poltura.com/en/apps/telegram/) app as well. This is optional. ## Set up fraud detection 1. Open **App Store** in the sidebar and choose **Fraud detection**. 2. If the app is not installed yet, select **Install**. See [App Store](https://docs.poltura.com/en/apps/#install). 3. Select **Turn on**. There is nothing to paste: no account or key is needed. 4. Open **Fraud Detection** in the **Apps** group of the sidebar. From now on every new order is screened and shows up in the numbers on this page. 5. Open **Settings** under **Fraud Detection** and review the thresholds, see [Settings](https://docs.poltura.com/en/apps/fraud-detection/#settings). ## Read the dashboard **Fraud Detection** opens on the dashboard. It shows eight numbers: - **Queued**, **Processing**, **Exhausted** and **Completed Today** describe the checks themselves. Exhausted means a check failed on every attempt and needs your attention in the queue. - **Pending**, **Approved**, **Blocked** and **Blocked Amount** describe the reviews: how many orders wait for a decision, how many were let through, how many were stopped and what they were worth. The buttons **Queue**, **History** and **Settings** at the top open the other pages. ![The Fraud detection dashboard with eight numbers in two rows and links to the Queue, History and Settings pages](https://docs.poltura.com/screenshots/en/apps/fraud-detection.webp) _The second row counts the orders waiting for you, approved and blocked._ ## Decide on a risky order A risky order waits as a pending review. You can decide in three places. 1. Open the order. The **AI Fraud Review** panel shows the **Risk score** out of 100, the **Risk level** (**LOW RISK**, **MEDIUM RISK** or **HIGH RISK**) and the **Signals** that led to it. 2. If you trust the order, select **Allow Order**. The panel status changes to **Approved**. 3. If you do not trust it, select **Cancel and refund** and confirm. The order is canceled, the payment is refunded and the panel status changes to **Refunded**. If the order already has a fulfillment, cancel the fulfillment first, see [Payments and refunds](https://docs.poltura.com/en/orders/payments/). A card payment is only reserved on the shopper's card for a limited time, set by the payment provider. While an order waits for your decision, its panel and the reminders show the date until which the payment can be collected. Decide before that date: afterwards, allowing the order may fail. Fraud verdicts arrive in the notification bell with Approve and Refund buttons, and in Telegram too when you switch the fraud notifications on for it in your notification settings. > **Warning** > > A pending review does not wait forever. When the review timeout passes, a low or medium risk order is approved. A high-risk order is never canceled or refunded for you: it stays on hold with the status **Needs your decision**, and you get a reminder every day until you allow it or cancel and refund it. You get one more reminder in the last 24 hours before the card authorisation ends. ## Work through the check queue The **Queue** page lists the fraud check task of each order. 1. Open **Fraud Detection** and choose **Queue**. 2. Search by order number, or filter by status: **Queued**, **Processing**, **Exhausted** or **Completed**. 3. Read the row: **Attempts**, **Next Retry** and **Last Error** tell you whether the check is still trying. 4. Select **Detail** to open the task. The **Attempt Timeline** lists every attempt with its result. 5. Select **Retry** to run the check again. The message **Task queued for retry** confirms it. From an order, **View in Fraud Detection** in the **AI Fraud Review** panel opens the queue filtered to that order. ## Look up past checks The **History** page lists every check attempt, newest first. 1. Open **Fraud Detection** and choose **History**. 2. Filter by order ID, or by status: **Success** or **Failure**. 3. Read the **Score**, **Tier** (**Low**, **Medium**, **High**), **Duration** and, for a failed attempt, the **Error**. ## Settings The app has no fields in the App Store. Its settings are on **Fraud Detection → Settings**. Select **Save Changes** after editing. - **Score Thresholds.** **Auto-approve threshold**: an order at or below this score is approved on its own (30 to begin with). **Auto-block threshold**: an order at or above this score is high risk (70 to begin with). - **Bulk Quantity Rules.** Large quantities add to the score. Under **Global Defaults**, **Max quantity** is the quantity above which the rule applies, **Score addition** the points added per item, and **Max total delta cap** the most points bulk rules can add to one order. With **Add Bulk Rule** you set a different **Max Quantity** and **Score Addition** for one product or one category. The most specific rule wins: product, then category, then the global default. Bulk rules never apply to wholesale tier members or company contacts, who buy in trade quantities by design. - **ASN Blacklist.** Type a network number such as `AS12345` and select **Add**. Orders from a listed network never score below 25. - **Review timeout (minutes).** How long a pending review waits before a low or medium risk order is approved. A high-risk order stays on hold for your decision. - **Auto-capture on fraud approval.** Captures the payment when a review is approved. When it is off, you capture the payment yourself on the order page. ## Good to know - Trust signals, such as a returning customer or an old account, raise the approval threshold for that order. They do not hide risk signals. - The location and network of the buyer's IP address are part of the score when that data is available. Checks continue without it otherwise. ## If something goes wrong - **A task is "Exhausted".** The check failed on every attempt. Open **Detail**, read **Last Error**, then select **Retry**. - **"Numeric fields must be whole numbers."** A threshold or rule field holds a decimal or text. - **"Pick a product or category for the rule first."** Search and select a target in **Add Bulk Rule** before you save the rule. - **No Telegram message arrives.** Check that the fraud notifications are switched on for Telegram in your notification settings and that the [Telegram](https://docs.poltura.com/en/apps/telegram/) app is set up with a staff chat. - **Allow Order does not go through.** The payment could not be collected, for example because the card authorisation has lapsed. The order stays on hold and the panel shows the reason. Cancel the order and refund it, or contact the shopper to pay again. - **A good order is on hold.** The review timeout passed while the order was high risk, so it waits for you. Select **Allow Order** in the order's panel. ## Related - [App Store](https://docs.poltura.com/en/apps/) - [Telegram](https://docs.poltura.com/en/apps/telegram/) - [Utánvét Ellenőr](https://docs.poltura.com/en/apps/utanvet-ellenor/) - [Order details](https://docs.poltura.com/en/orders/order-details/) - [Payments and refunds](https://docs.poltura.com/en/orders/payments/) - [Notifications and alerts](https://docs.poltura.com/en/settings/notifications/) - [AI credits](https://docs.poltura.com/en/assistant/ai-credits/)