Contents

Frequently Asked Questions

  • Can Predict combine online and offline data to improve recommendations?
    Yes. On the Historical Sales Data page you can jumpstart the recommender by importing historical data, but also regularly import sales data from offline sources.

  • Can Predict data collection scripts be implemented via Google Tag Manager?
    Yes. If you are comfortable using Google Tag Manager, you can do the integration this way. Having said that, you will still need to review the Predict documentation to make sure you are sending all the required information. You will also have to check that all the correct scripts have been added in relation to various events and the required input variables included.

  • Does Predict support mobile applications?
    Yes, in most cases. For mobile apps using web technologies (e.g. HTML/JavaScript in a browser or embedded in a native shell) our JavaScript API is directly applicable for collecting data and delivering recommendations. For fully native apps, you could use our web service API directly.
    However, in this case there are additional details to cover relating to user identification, tracking, etc. Before committing to such a project, please discuss the specifications with the Emarsys Predict team.

  • Does Predict use machine learning from other Emarsys client accounts?
    No. Each Emarsys client’s data and analysis results are stored separately. Even crowd behavior from similar industry verticals is treated confidentially. We take data security very seriously and mean it when we promise not to use your data for other than your own benefit.

  • Email Personal Recommender vs. Post-Purchase Recommender? When should I use which?
    Unlike the Email Personal Recommender, which will show products that a contact might be interested in buying depending on their online behavior, the Post-Purchase Recommender widget will show products which are specifically correlated to their last purchase. As an example, let’s suppose they just browsed your website and ended up purchasing a digital camera. While Email Personal will show more digital cameras (according to your online behavior, you’re interested in cameras), the Post-Purchase Recommender will show memory cards, camera cases, etc.
    Therefore, as a best practice:

    • Email Personal should be part of routine newsletters and email retargeting (as shown by the Email Retargeting blueprint in the Automation Center).
    • Post-Purchase should be included in emails which are going out around a day or so after a purchase was made (which can also be triggered by an Automation Center program).
  • Who designs the Email Recommendations?
    You do. If you are using the visual CMS, you can include great-looking recommendations as part of the email creation process. If you prefer to use your own HTML code, you can use the Email Widget Designer to integrate the recommendations in your email.

  • Email vs. Web Personal recommendations? Will both show the same products? And if not, why not?
    By default they are using mostly the same signals/models, but on a different schedule. On the website these recommendations are displayed in real time, meaning that they change as the session progresses and each new page is loaded. In contrast, email recommendations are calculated nightly, so that the email content is stable and does not change while the recipient is reading it.

  • How can I recommend only items that are in stock?
    This is a built-in feature of Predict: stock availability is controlled through the catalog. Items that are temporarily out of stock or no longer offered should still be included in the catalog, but the field Available should be marked as FALSE. These products will not be displayed in the recommender widget, but will be included in the analysis.

  • Is Predict SEO-friendly. Or, more precisely, are the recommender widgets displayed on the my website SEO-friendly?
    Yes. The links served up in the Predict widgets are the links that the you have provided in the catalog, i.e. they are your own links. Concerning search engines, Google have been executing Javascript and crawling our links since at least 2011.

  • Is visitor data safe with Predict?
    Yes. Predict only collects behavior data using anonymized customer IDs, and it and anonymizes visitor cookies. No personally identifiable information is managed or handled within Predict.

  • My product image does not display within the Email Widget Designer – is there a limitation on the size of the image?
    Yes. There is a limit of 450Kb for product images in the Email Widget Designer; larger images will not be displayed. The Live Event View shows all images the same way as a browser does, so please make sure the images look good in the Email Widget Designer, too. The Email Widget Designer shows the actual email recommender image, exactly as it will render in an email.

  • Are multiple Predict accounts with one Emarsys account possible?
    No. In a single Emarsys account we can only put one Predict dashboard and can only connect the CMS to one Predict account. With Website Recommender there is less of an issue with multiple predict accounts, obviously.

  • Will the Predict scripts slow my website?
    All the Web Extend and Predict JavaScript snippets are asynchronous. No element of the Predict infrastructure can slow your website or affect user experience in any negative way, even in the case that the Predict service is completely unreachable for the visitor (e.g. due to network issues). The worst case scenario is that Predict widgets will not be displayed, without affecting the rest of the page.

  • Can I promote specific products in the widget (i.e. whitelist/blacklist items)?
    This is possible – using the product catalog, you can flag specific products that we can push to our models according to specific rules, and you can blacklist specific items by flagging their availability as “false”.
    However, in normal circumstances we would recommend not to try to influence the recommender with whitelists. There are two reasons for this: – We believe (and have in fact measured) that our behavioral recommendations are always better than any manually selected list of products, so introducing manually selected products into the behavioral widgets can only decrease conversions.
    We also believe that it’s always best to keep special promotions what they are – special, in the sense that they are displayed separately from normal products. Think of Google searches, where promoted items are never mixed with regular search results.
    If you do want to use whitelists, please contact the Emarsys Predict team to discuss your requirements. Blacklists are more common – we do use them to flag out-of-stock items and items that cannot be shown in widgets (e.g. adult content, brand collision). Again, please discuss your needs with the Emarsys Predict team.

  • I have an international website. Visitors from different countries buy different products. What should be done in order to be sure I recommend the right products to each audience?
    Nothing! If there really are such regional tendencies, since all visitors are interacting with the same recommender engine, the engine will learn from this and reflect these differences automatically. And because individual customers get the PERSONAL recommender widget, which uses personal browsing history, this whole issue is not relevant for the personal email and website recommendations.In other words, this is precisely how Predict is designed to work and there is no need to customize the engine or the data-collection in this matter. All localization options in Predict are related to managing languages, and differences in stock/availability.

  • What is Predict web recommendations response time?
    This depends on what you mean by ‘response time’. Our server-side latency for answering a web recommendation request is below 1 millisecond at the 99th percentile. The client-side response time will be determined by network latency – as our servers are located in Ireland, this works out to be around 10-40 milliseconds for most of Europe.
    The actual user experience will be determined by all the other elements in your website (i.e. when they start calling the recommender and how long it takes to load the product images from your server). An additional note is that our Javascript API is fully asynchronous, which means it will in no way hold up the loading of your website, even if there is some network connectivity problem between the visitor and our servers.

  • What is the level of Predict uptime?
    Our uptime on the current AWS infrastructure was 99.998% over the past year (12 mins. downtime).

  • What degree of language support do you offer?
    Recommender widgets display information from the product catalog – whatever is passed on from your database is displayed in the widget. Therefore there are no language issues here. Even if your online business is international, and offered to users in different countries and under different domains (e.g. international mystore.com site, and localized mystore.co.uk), this is not a problem.
    Multi-domain integration uses a single, aggregated product catalog that contains all the localized variants of the catalog data for each item ID. Localized data (product names, prices, etc.) need to be added to your main store catalog export. The Emarsys Predict Dashboard and supporting documentation are currently only available in English but will be translated into all Emarsys languages soon.

  • How and when do you judge a cart to be abandoned?
    When a visitor browses a site we track them and check the content of their shopping cart. If there are items in the cart and no purchase has occurred by the end of the day, the cart is judged to be abandoned (in a future release we plan to offer the option to reduce this period to one hour). If multiple items are added to the cart and only some of them are purchased, this is not counted as an abandoned cart. However, the remaining items will be again regarded as abandoned the next time the contact browses the site without making a purchase.