Experience Optimizer

Overview

Algonomy offers a multitude of strategies (or “algorithms”) for tailoring the customer experience. Each strategy satisfies a specific merchandising objective and leverages the merchant’s catalog and customer behavior to generate recommendations. Algonomy ingests information about the specific customer and page context—and based upon that data, it shows a strategy that drives the highest engagement. This strategy optimization method is known as the “King of the Hill” concept.

The King of the Hill eliminates the guess work of finding the right algorithm by automating the testing of the strategies to pick up the best for the given situation.

Strategies are launched into a competitive environment to determine which are the most effective to display. The Experience Optimizer or the decision maker tests the strategies among a subset of customers; maximizing for customer engagement (for example, recommendation click-through rate), and the best performing strategies receive the greatest visibility.

  • Strategies/algorithms are informed by browse, search, and purchase behavior collected by Algonomy instrumentation on the client site.

  • Through the self-optimizing “King of the Hill” process, recommendations are tested to a subset of customers and the most successful for the given context (for example, page type and category) are displayed more prominently.

  • Recs can be tuned (for example, filtered, boosted, and overridden) with merchandising rules set up on the Omnichannel Personalization dashboard. (The optimization method will always respect merchandising rules).

How Experience Optimizer Works

Experience Optimizer respects the Strategy Configuration and Strategy Rules that are in place for the current Context.

  • Strategies are enabled on the Strategy Configuration page at the page type level. Strategies that are enabled for a given page type are eligible to play for any placement of that page type.

  • Strategy Rules can be used to indicate that there is a set of “preferred” strategies that should be attempted to play before the other strategies.

  • When a placement is requested, if there is a Strategy Rule in place, the Experience Optimizer will attempt to select from those strategies that are defined in the Strategy Rule. Which strategy is selected will be determined by the past performance of the strategies in the current Context for the optimization metric that is configured for that page type.

  • The current Context is defined as the combination of various input parameters, including channel, placement, category, referrer, segment.

For example, let’s say that the following strategies are enabled for the Item Page: ClickCP, ClickEV, ViewedPurchased, CategoryTopSellers, and CategoryTopProducts. There are 2 placements: the first has no Strategy Rule, and the second has a rule that has defined ClickCP, ClickEV, and ViewedPurchased as the preferred strategies.

For the first placement, Experience Optimizer will consider all 5 strategies. Whereas for the second placement, Experience Optimizer will first consider the 3 preferred strategies. In the case where none of the preferred strategies can play, it will then consider one of the remaining fallback strategies.

In both cases, the strategies that are in consideration will be selected based on the previous performance of the optimization metric that is configured for the item page. Let’s assume this is click through rate, which is the default metric.

Exceptions:

  • Exploration % is reserved for testing Experience Optimizer. Within this percentage of views, Experience Optimizer will attempt strategies other than the best performing one. This data is used for its learning. It is necessary for Experience Optimizer to be able to adapt to episodic changes in performance.

  • If the “use only these” option was selected in the Strategy Rule, then Experience Optimizer will only attempt to select from those preferred strategies. That is, it will not attempt to play any of the fallback strategies.

  • If the “specify order” option was selected in the Strategy Rule, then Experience Optimizer will select the strategies based on the defined order. That is, it will not use past performance as a way of determining which strategy to select.

You can change the optimization metric at the page type level and exclude specific parameters from the Visitor Input, which is used to define the Contexts. For more information on configuring experience optimizer, see Experience Optimizer Configuration.

Key Values

While engagement can be important, retailers are most interested in displaying strategies that generate the most Revenue Per Session lift, not necessarily those get the most clicks. The challenge is that, given number of strategies in play, it’s difficult to determine a specific strategy’s contribution to a retailer’s incremental revenue.

To that end, the King of the Hill optimization provides the following key values to our clients:

  • Reduce costly decision-making: Rely on the self-optimizing personalization engine to determine which strategies to display to a given customer.

  • The right strategy: No single strategy will consistently be the top-performer. The “King of the Hill” rotational system gives episodic winners the chance to surface in recommendations.

  • Display guarantee: Recommendations will continue to substitute to the next-best strategy if the top doesn’t qualify (unlike hard-coded recommendations).

  • Driving relevance through ensemble learning and frequent remodeling: Catching and responding to customer trends all the time - recommendation models rebuild multiple times a day.

  • 100+ algorithms address multiple dimensions of the customer with explicit and transparent messaging.