Algo-based Personalization: Troubleshooting & FAQs
Refer to our FAQ section to troubleshoot and resolve any issues related to Algo-based Personalization.
Explore troubleshooting tips and FAQs for Email for detailed information by expanding the sections below.
Algorithm and Data Usage
Q. How does the algorithm work?
A. Refer to this document to learn how Algo-based Personalization works.
Q. How does the algorithm work for new customers?
A. For first-time visitors, the algorithm shows popular products based on contextual factors like location and time. It learns from the customer's interactions and improves recommendations in real time.
Q. How does the algorithm work for non-signed-up customers?
A. The algorithm tracks customers using a cookie ID, allowing us to personalize recommendations even for non-signed-up users.
Q. How does the algorithm handle frequently changing products?
A. The algorithm allocates a percentage of recommendations to new products to capture data. Over time, as new products gain enough data, they are included in the main personalization bucket.
Q. What data does the algorithm use?
A. The algorithm uses three sets of data:
- User interactions (eyeballs, clicks, purchases, add to carts, wishlists, searches, filters).
- Product catalog data.
- User context (location, device details).
Q. How will the algorithm benefit my business?
A. The algorithm delivers personalized product recommendations, enhancing the user experience and increasing engagement. More details are available here.
Data Collection and Privacy
Q. What data will you collect from my website/app?
A. We collect data on user interactions, product catalog, and user context. See the previous question for details.
Q. Do you need our historical data?
A. While historical data is helpful, it is not necessary. We can start fresh with the data collected from the sign-up date.
Q. How safe is my data? What are the privacy policies?
Your data is securely stored on Amazon Web Services. For more details on data security, visit AWS's data privacy FAQ. We can sign an NDA to ensure data safety.
Q. How does customer mapping work in personalization?
A. Default: Different browsers = different boxx token IDs.
Case 1: Logged-in users have one customer ID and one boxx token ID.
Case 2: Non-logged-in users are identified by boxx token ID, later mapped to customer ID upon login.
Q. What is a customer ID? Is it the same as a unique ID?
A. Customer ID is set by the client, while a unique ID is a UUID/GUID stored in a cookie by us for internal identification.
Integration Process
Q. How can the algorithm be incorporated into my E-commerce website?
A. Install our JS on your website, and we'll handle the rest.
Q. How about app integration?
App integration is more complex. We start with the website and move to the app in Phase 2 after demonstrating success.
Q. How long does app integration take?
A. App integration takes about 40 hours of tech effort and is linked to a new app version release.
Q. How do you get my data?
A. Netcore JS automatically reads it from your website.
Q. How do you maintain the look and feel of my website?
A. We replicate your CSS files to ensure the interventions match your website's design.
Q. How do you get my product feed?
A. Our JS reads your PLP and PDP pages every 2 hours. We may also request your product feed if available.
Q. How do you track new products and out-of-stock items?
A. Our JS reads your PLP and PDP pages every 4 hours to update product status. Ideally, we use the product feed you provide.
Q. Will your integration slow down my website?
A. Our installations are asynchronous, ensuring they do not slow down your website.
Q. What if your API fails?
A. If our API does not respond in 200 ms, the intervention collapses, and your website functions as usual.
Q. Where is your server located? What about network latency?
A. Our server is in Mumbai on AWS, using Cloudfront with 1-hour caching to ensure a 200 ms response time globally.
Q. We use caching on our website. Will it be a hindrance?
A. We handle it by creating a hole in your cached HTML for our recommendations.
Implementation and Testing
Q. How will the integration process work?
A. 1. Install our JS and provide product feed access.
- Integrate data ingestion and start collecting interaction data.
- Integrate personalization and make it limited live for testing.
- Go live after testing and approval.
Q. Can we see the personalization before it goes live?
A. It will be made limited live at a specific link for your testing.
Q. Can we make it live for a limited number of customers initially?
A. Yes, we can start with 1% of customers and gradually increase as you gain confidence.
Q. Can we test the recommendations visually before they go live?
A. We recommend relying on data rather than human judgment for the best results.
Q. How long does it take for the algorithms to optimize?
A. The algorithm needs an average of 100 clicks per product to optimize. For example, with 10,000 products and 1 million monthly visits, it would take about 10 days to start converging.
Q. Can we test this on staging first?
A. Testing on staging is not useful due to the lack of data. Instead, use the "limited live" link for UI testing.
Post-Launch Process
Q. What happens after we go live?
A. Week 1-2: The algorithm tests multiple models and delivers initial results.
Week 3-4: Optimal mix of models is learned.
Week 5-6: Move to longer-term planning based on results.
Q. Do you provide a management panel?
A. Yes, a basic panel is provided to manage interventions and view results.
Q. What do you need from us?
Tech: Install JS, provide product feed, and create a blank page.
Biz: Test the limited live version and provide sign-off.
Commitment: Long-term engagement based on initial results.
Q. How about our m-site?
A. It works the same as the website.
Q. How about our app?
A. App integration is done in Phase 2.
Q. What are the different phases of integration?
A. Phase 1: Quick JS-plugin-based integration for website and m-site.
Phase 2: Deeper API-based integration for website, m-site, and app.
Q. Is personalization different on the app versus the web for the same user?
A. Yes, due to different integration methods, go-live processes, UX, and user behavior.
Performance and Expectations
Q. What can we expect in the results?
A. Month 1: 20-50% increase in CTR.
Month 2: 5-15% increase in key metrics like page views, ATC, and purchases.
Month 3: Continued improvements in these metrics.
Month 6: Improved bounce rate, repeat visitors, and organic traffic.
Month 6-12 (with personalized marketing): Better email CTOR, notification open rates, ad CTR, and ad revenue.
Technical Specifications
Q. What is the size of the JS?
A. Approximately 40kb.
Q. What is a flicker effect, and how can we avoid it?
A. Flicker is a visible shift in UI elements after page load due to JS-loaded elements on a slow network. It can be minimized with optimized loading.
Q. How does backend integration work?
A. We provide generic SDKs and APIs for data ingestion and recommendations.
Q. Do Rich Relevance and Dynamic Yield show recommendations on the listing page?
A.No, they do not.
Q. Do we have stress test capability?
A.Yes
Q. How is product feed data passed, and how often?
A. Configurable, partially covered in Q17.
Q. What formats do you accept for product feed?
A. We accept Product XML feed as per industry standards.
Q. Do you need the latest product feed for category changes?
A. Yes
Q. How do recommendations work with filters?
A. Sort by filtering: Recommendations won't work.
Shop by filtering: Recommendations work if attributes are present in the feed.
Q. What is a personal boutique page, and how does it work?
A. A curated, personalized product listing page based on the user's interactions (views, ATCs, purchases).
Updated 4 months ago