Overview & Steps for Sense Platform
At its core, our fraud prevention platform operates in three simple steps.
- You send user / transaction / device data
- We enrich the data and deliver a risk score based on rules
- You give feedback on the results
You can find a detailed timeline here to see how long it takes to get results with SEON Sense.
Step 1 - Providing the Data
All the user, transaction, and device data is sent via the Fraud API. Your first step is to define payloads for the API, populating it with as many relevant data points as possible. All the fields are optional, but the more you fill, the more precise our results will be.
- For custom business-specific data points, use the
- The config object helps you to fine-tune settings such as versions, response, and aggregating data enrichment APIs when required.
- You must define the authentication points aka.
purchase, etc.) where risk assessment data can be collected or fraud should be prevented.
This allows for a more comprehensive overview of each transaction, allows us to establish meaningful connections across different users, and it makes for better data quality which is important for the machine learning functions to work properly.
Please get in touch with your dedicated customer success team member to tailor and validate your specific payloads. Our team is ready to support you via email, phone, the shared Skype, or Slack channel for any queries you might have.
You can find examples for an airline company and an e-wallet here, including custom fields specific to each industry.
Step 2 - Enrichment and Scoring
SEON Sense is designed to give you full transparency behind every score and decision (a.k.a. state). This is why every data point will be available in the response.
By default, the fraud scores are based on preset rules, which can be reviewed in the Scoring Engine. A score of 10+ is considered risky. Standard thresholds for each state are:
|APPROVE||0 - 10|
|REVIEW||10 - 20|
Step 3 - Feedback
Providing feedback is the key to refining the rules and getting more precise fraud scores. This is particularly important when discovering false positives and false negatives.
Every transaction state should therefore be set to the appropriate category:
|REVIEW||Suspicious transaction, not confirmed fraud yet.|
|DECLINE||Confirmed fraudulent transaction.|
You can also create categories of fraud reasons in the Machine Learning section of your Settings page, which support the Label API (e.g. chargeback, bonus abuser, or postback data from payment: authorized, lost or stolen, etc.).
Please jump to the Machine Learning part to learn more about how SEON's ML module can help to fine-tune its algorithm.