> ## Documentation Index > Fetch the complete documentation index at: https://docs.verisoul.ai/llms.txt > Use this file to discover all available pages before exploring further. # Best Practices > Best practices for integrating Verisoul For optimal fraud detection and user experience, follow these two critical best practices when integrating Verisoul ## 1. Initialize the Client SDK as Early as Possible **Why this matters:** Early initialization allows Verisoul to collect the maximum amount of behavioral and device signals, resulting in more accurate risk assessments. ### Implementation by Platform #### Web SDK Add the Verisoul script to the `` section of your HTML to ensure it loads before the rest of your page content. For production web integrations, use a [custom hostname](/integration/custom-hostname) so the SDK loads from your own subdomain: ```html theme={null} ``` If your custom hostname is not active yet, you can use `https://js.verisoul.ai/{env}/bundle.js` for sandbox, pilots, or initial testing. #### Mobile SDK For our mobile SDKs, initialization happens automatically when the SDK is integrated. Ensure you're following the platform-specific integration guides: * [iOS](/integration/frontend/ios) * [Android](/integration/frontend/android) * [React Native](/integration/frontend/react-native) * [Flutter](/integration/frontend/flutter) ### Benefits of Early Initialization * Allows more time for signal collection * Captures the full user journey * Improves accuracy of behavioral analysis * Enables detection of suspicious patterns from the start ## 2. Call the Verisoul API at the Time of Critical Actions **Why this matters:** While you should initialize the SDK early, you should wait to call the authenticate endpoint until the moment of a critical action. ### Critical Action Examples * User registration * User login * Payment processing * Rewards redemption * Account withdrawal * Profile or settings changes * High-value transactions ### Implementation Guidance 1. Initialize the SDK early in the user journey 2. Collect the session ID when needed 3. Call the [Authenticate](/integration/backend/authenticated) endpoint at the exact moment of the critical action 4. Use the decision to allow, block, or verify the action ### Why This Approach Works * **Maximum signal collection**: Client SDKs continuously collect data, so waiting allows more time for signals to be gathered * **Defeats "human in the loop" fraud**: Fraudsters often create accounts manually but later automate actions * Example: A fraudster creates an account as a human, verifies their email, and then logs in and starts a bot * **Captures intent signals**: The most revealing behavioral signals often occur right before a critical action * **Reduces false positives**: More context leads to more accurate decisions