Best practices for widgets

Widgets in Optimizely Product Recommendations let you personalize the products you recommend to each customer on your website. To ensure your widgets give the most accurate recommendations and to avoid your widgets not returning any recommendations, use the following best practices.

The feed plays a very important role in widget recommendations and is often a key factor when an issue with your recommendations occurs. Use the following best practices for your product feed to optimize your recommendations.

Verify that the product feed contains all the products you want to recommend. If you remove a product from the feed, Optimizely Product Recommendations marks it as inactive, even if that product is available on your site, resulting in that product not being available for recommendations. To recommend a product, it must exist in the latest feed.

Verify that any attributes you use in rules for widget algorithms or merchandising campaigns exist in your product feed, have matching names, and are assigned to the relevant products.

You may have a series of attributes you pass against your products, like material or collection. You may use some of these attributes directly in your recommendation rules, like “match on material.” For that rule to work, the material attribute must exist on products in the feed, and the attribute names must match.

For example, if you rename material to productMaterial, that breaks the "match on material" widget algorithm rule. You must make that same name update to any widget algorithms or merchandising campaigns that use that attribute for them to continue recommending the correct products. Similarly, if you remove attributes from your feed, ensure you are not using them in any widget algorithms or merchandising campaigns, as that can affect the recommendations those widget algorithms or merchandising campaigns return.

The system will not recommend products with a stock quantity of zero. If you want to recommend products that have zero stock, like made-to-order, use a placeholder stock value to keep them active in recommendations. A product's stock quantity must be greater than or equal to one (1) to be included in recommendations.

Ensure all products in the feed have images attached to them. If a product that is missing an image gets recommended, that can confuse customers and decrease engagement.

Consider implementing logic to exclude products without images from recommendations. Alternatively, you can update your feed to include an image attribute with a y (yes) or n (no) value and assign that attribute to each product in the feed. You can then add an r.image="y" rule to all of your widget algorithms to ensure only products you have designated as having an image can be recommended.

In your widget algorithm rules, you can apply rules to recommend unique products, like unique-category or unique-brand, letting you provide a variety of different recommendations.

For example, on a home page, you may want to provide a wide variety of recommendations for new customers to increase the chance of helping them find a product directly relevant to them. You can achieve this by using unique-category in your widget algorithms.

However, there are some edge cases you should consider to ensure recommendations are not impacted. There may be times when you apply a new category to a large portion of your catalog (or entire catalog), like a Black Friday or other sale category. This would make the unique-category rule redundant as now everything belongs to the same category. There is no uniqueness left, and you would end up with 1 product being recommended and no more.

If you plan to add an attribute value across a large portion (or the entirety) of your product feed, update your widget algorithms to remove any reference to the corresponding unique-<attribute> filter.

When configuring widgets and their algorithms, you can add quick and advanced filters to further personalize the products the algorithm recommends for the widget.

Be careful when applying filters, as you could make the rules too specific that they do not match any products and do not return any product recommendations. If you apply too specific of a filter to a single algorithm, only that algorithm is affected. But if you apply it to multiple or all algorithms, you could cause the most relevant algorithms or even the entire widget to not return any product recommendations.

For example, you may have a filter on algorithms 1-3 but no filter on algorithms 4-11. If your filters on algorithms 1-3 do not work as expected, this would result in the top 3 algorithms being skipped and the first set of products coming from algorithm 4 onwards.

You will likely still get a full set of recommendations being returned to your widget, but the results are less effective than intended. The algorithms stack is set up so that the top algorithms will return the most efficient recommendations for the strategy you are trying to achieve. Skipping these algorithms can result in a less efficient set of recommendations that may result in less engagement from your customers.

Use the following best practices to ensure your filters work as expected.

When adding filters, ensure that there are no conflicts. For instance, you may add the Same parent category quick filter and then add an advanced filter of category=123. This rule works perfectly if you are viewing a product from category 123, but for any other category, this rule does not work.

In another example, you may add the Downsale or Upsale quick filters. These filters recommend products below or above the price of the product currently being viewed. Additionally, you may have a price minimum or maximum threshold advanced filter, such as only ever recommending products greater than $30. If the algorithm aims to recommend products over $30 and down sale, and a customer views a product for $25, the algorithm tries to return products over $30 and under $25 at the same time. These conflict with each other and cause the algorithm not to return any product recommendations.

You may want to add a very restrictive or niche rule to your recommendations, like on category 123, only recommend products from category ABC that have a specific color, material, and style. Depending on how much behavior these products have and how large those categories are, this rule could be too niche to return a full set of product recommendations.

If this is the case, you should ensure this rule is not attached to each algorithm in the stack but rather to have a staggered approach to the rules. Consider having the niche rule on algorithms 1 through 3 and then removing one rule at a time. For example, on algorithms 4 through 6, remove the requirement of style. On algorithms 7 and 8, also remove color. Continue this trend until only the most important rule is attached to the algorithm. This will massively increase your chance of having a full set of product recommendations in your widget.

You should always use at least two of the following fallback methods for each widget:

If you add both the fallback algorithms, ensure Fallback recommendations (with no filters) is the last one as it is the least restrictive.

These fallback methods ensure you have the best chance of receiving a full set of product recommendations in case the rest of the widget algorithms return nothing. Without them, there is a chance that you may encounter fewer recommendations than desired or even zero recommendations.

Merchandising campaigns are a tool available within Product Recommendations that lets you add an additional filter to a specific widget.

Widgets without merchandising campaigns can have up to three levels of rules before returning a product. 

With a merchandising campaign, you can add a fourth level of rules. While that can improve the accuracy of your recommendations, it can also cause potential complications. The main two concerns are:

Within a merchandising campaign, you have the option to preview the rules you are creating on a given widget, so you can ensure your rules are working before making them live. You should preview multiple products to ensure consistency in your rules and to help you potentially find any edge cases you need to clean up.

Assume you are using a simple widget with one Bestsellers by units sold algorithm, and you set the widget to return 10 recommendations.

If the algorithm returns 100 products to start, since you specified to only return 10 recommendations, the system sorts those 100 products by their algorithm score. In this case, that is done by taking the top 10 bestselling products across the site.

To improve your recommendations, you add the Same parent category quick filter. This may reduce the original 100 products down to 75. The system then takes the top 10 of those 75 using the same ordering process.

Finally, you add an advanced filter for sale price over $50 (r.saleprice["USD"]>"50"). This may reduce the matching products down to 50, and again, the system takes the top 10 of that 50.

If there is no merchandising campaign attached to this widget, then you have your 10 recommendations that this widget's algorithm will display on a given product page.

A merchandising campaign lets you add a fourth level of rules, letting you further refine the recommendations. Merchandising campaigns are best used when trying to achieve a rule on a specific portion of a widget. This is because you have the option to define a master rule, where you can use the attribution in your feed to target a specific product or group of products.

For example, you may want to recommend products over $50 at the widget level, but for category XYZ, which has higher-priced products, you want to add a rule to recommend products over $100. In your merchandising campaign, you could add a master rule of category="XYZ" and then add a recommendation rule of sale price>=100. You have now added a fourth level of rules to this area of your widget.