Why Amazon Fraud Detector?

Amazon Fraud Detector, a fully managed service built on over 20 years of insights from Amazon, helps customers identify potentially fraudulent activities and catch more online fraud faster. With Amazon Fraud Detector, you pay only for what you use, and there are no minimum fees or upfront commitments. You are charged based on the compute hours used to train and host your models, the amount of storage you use, and the quantity of fraud predictions you make.

Free trial

Sign up now to try Amazon Fraud Detector free for two months. The offer includes 50 compute hours for model training, up to 500 compute hours for model hosting, 20GB of stored event data per month, 30,000 Online Fraud Insight predictions, 30,000 Transaction Fraud Insights predictions, 30,000 rules-based fraud predictions per month and 1 Million Account Takeover Insight predictions for the first two months.

Pricing at a glance

You are charged by the Gigabyte (GB) for storing event data in Amazon Fraud Detector. Data storage is optional. Event data can be stored both through uploads of historic events and when generating predictions.

Amazon Fraud Detector charges for the compute hours consumed to train a custom model with your data. A compute hour represents one hour of compute capacity using 8v CPUs and 32 GB memory. Fraud Detector automatically chooses the fastest, most efficient instance type to train your data, and the instance may exceed baseline specifications. Therefore, the number of compute hours billed may be greater than the number of elapsed training hours.

Fraud Detector charges for compute capacity by the hour for on-demand hosting of deployed models, so that they are available for real-time predictions.

Fraud Detector chargers per fraud prediction. The price you are charged varies based on whether you are using an Amazon Fraud Detector machine learning model or only rules. The per prediction price is the same for real-time and batch predictions. Your fraud predictions are aggregated for each month's usage and billed according to the pricing tiers. Predictions that use a model imported from Amazon SageMaker are priced as rules-based predictions.

Pricing details

Data Processing and Storage    Pricing
Data Processing and Storage $0.10 per GB
Model Training and Hosting             Pricing
Model Training $0.39 per hour
Model Hosting $0.06 per hour
Fraud Predictions Pricing
Online Fraud Insights  
First 100,000 predictions per month                                 $0.0300 per prediction
Over 100,000 predictions per month $0.0075 per prediction
Transaction Fraud Insights  
First 100,000 predictions per month $0.0300 per prediction
Over 100,000 predictions per month $0.0075 per prediction
Rule-based Fraud Predictions  
First 400,000 predictions per month $0.00500 per prediction
Next 800,000 predictions per month $0.00250 per prediction
Over 1,200,000 predictions per month $0.00125 per prediction
Account Takeover Insights  
First 10,000,000 predictions per month $0.0010 per prediction
Next 90,000,000 predictions per month $0.0005 per prediction
Over 100,000,000 predictions per month $0.0003 per prediction

Example 1: Real-time online fraud detection for an ecommerce merchant

You are an ecommerce merchant who’s looking to protect yourself from high risk guest checkout orders and chargebacks. Let’s say you upload 5 GB of data and train a single model twice per month, with each training consuming 10 compute hours to complete. Further, you deploy one of the models for the entire month and use it to generate 1,000 real-time fraud predictions per day. The bill for the month for using Amazon Fraud Detector will be:

Data processing and storage charge = 5 GB x $0.10 per GB = $0.50

Training charge = 10 compute hours x 2 trainings x $0.39 per compute hour = $7.80

Hosting charge = 30 days x 24 hours x 1 model x $0.06 per compute hour = $43.20

Fraud prediction charge (real-time and batch) = 1,000 predictions / day x 30 days x $0.03 per Online Fraud Insights prediction = $900

Total cost = $0.50 + $7.80 + $43.20 + $900 = $951.50

Example 2: Transaction fraud detection for a payment service provider

You are a payment service provider that offers card-not-present acceptance solutions looking to reduce transaction fraud by flagging suspicious payments. You plan to implement a machine learning model and a rules-based decisioning system. Let’s say you upload 20GB and train two models once a month, with each model training consuming 10 compute hours to complete. You then pick the better performing model and deploy it for the entire month. You generate 20,000 real-time fraud predictions per day (resulting in 600,000 transactions per month) and 1000 rules-based fraud prediction decisions per day (resulting in 30,000 transactions per month). The bill for the month for using Amazon Fraud Detector will be:

Data processing and storage charge = 20 GB x $0.10 per GB = $2

Training charge = 2 models x 10 compute hours x 2 trainings x $0.39 per compute hour = $15.60

Hosting charge = 30 days x 24 hours x 1 model x $0.06 per compute hour = $43.20

ML-based fraud prediction charge (real-time) for first 100,000 transactions = 100,000 predictions x $0.03 per Transaction Fraud Insights prediction = $3,000

ML-based fraud prediction charge (real-time) for next 500,000 transactions = 500,000 predictions x $0.075 per Transaction Fraud Insights prediction = $3,750

Rules-based fraud prediction charge = 1,000 predictions / day x 30 days x $0.005 per Rules-based fraud prediction = $150

Total cost = $2 + $15.60 + $43.20 + $3,000 + $3,750 + $150 = $6,960.80