How Schneider Electric increased opportunities for upselling and improved the adoption rate of new products using Amazon Personalize

Intro

Schneider Electric has an extensive offering for its customers and partners to help build the best value, with more than 100,000 products in its catalog. Electricians in the residential segment need to identify the best combinations, with the right compatibilities, and an overall consistency when they design electrical and wiring device installations for their customers. Schneider Electric’s pace of innovation and consistency in new product releases result in rapid growth of new products and opportunities for electricians to integrate new technologies into their projects. Keeping up with the pace of innovation requires a significant time investment for those wishing to stay on top of the latest technology.

This is an opportunity where artificial intelligence (AI) and machine learning (ML) can facilitate electricians in building their quotations and selecting the right items from the Schneider Electric range of products. “Our electricians and partners are facing a wide variety of choices when it comes to electrical job quotations, starting with switches and sockets selections to electrical panels design, as well as EV charging installations. We wanted to create a solution to help them be more efficient in their daily jobs, using AI/ML in a natural way,” says Uyanga Ganbold, Software Director at Home and Distribution (H&D) at Schneider Electric, anticipating a digital transformation before the beginning of this journey.

Following the discussion with Uyanga, they identified the context as similar to that of customers looking for the best movie streaming recommendations that are based on personal history, context, and ideas of users with similar content preferences. User experience based on such metrics is often referred to as
personalization.

But how to identify, describe, define, refine all these rules, and prioritize the best-fit product selection for the specific job at hand? Who will be ready to catalog all these details and maintain them to keep it appropriate with changes coming every month, in many countries at a time?

Amazon Personalize—which allows developers to quickly build and deploy curated recommendations—is a fully managed ML service that goes beyond rigid, static, rule-based recommendation systems and trains, tunes, and deploys custom ML models to deliver highly customized recommendations to customers across industries such as retail and media and entertainment; it can bring this innovation into the electrician landscape. Building an application that uses previous electrician actions, preferences, and experiences to suggest best product options was identified as a quick way of improving end-user experience and facilitating decision-making.

In situations where multiple options are available with hard-to-define selection criteria and rules, using ML to build a simplified end-user solution for electricians provides a positive approach. Alongside the Schneider Electric Home & Distribution Software team, Amazon Web Services (AWS) set out to help electricians accelerate product selection and product integration through personalized recommendations,
ultimately reducing the time for bill of material / quotation setup by presenting more recent and innovative products.

Solution

The AWS Prototyping team and Schneider Electric started with analyzing existing user journeys and defining target user experience, then worked backward to iteratively design a prototype that would help electricians choose the right products while preparing a quote for their residential projects in the company’s mobile app, mySchneider app electrician experience. The primary goal of the prototype was to demonstrate how a recommendation system built using Amazon Personalize would help Schneider Electric solve two of the major pain points hindering its product adoption rate and revenue growth in the residential space:

• Electricians preferring to use outdated Schneider Electric products, despite new and improved Schneider Electric products available on the market
• Missed opportunities for product upselling, because electricians are reluctant at times to propose premium product offerings with better aesthetic and design for consumers

After initial discussions with the electricians, Schneider Electric found a couple of reasons for the above pain points: 1) electricians are comfortable with using products that they are already familiar with; and 2) electricians are not staying up to date with all the latest product offerings from Schneider Electric. To help alleviate these pain points, a recommendation system was built in the prototype using Amazon Personalize to provide near-real-time recommendations to electricians, satisfying the following criteria:

• Most viewed products based on housing project size, location, and room type
• Product suggestions based on neighborhood characteristics, such as total number of electric vehicle (EV) charging stations
• Smart-home product suggestions (upselling) based on the range and/or color selection
• Product suggestions personalized for each electrician based on their purchase history

Amazon Personalize is a fully managed ML service that offers prebuilt ML
models, called recipes, that were trained using historical electrician-product interactions datasets from Schneider Electric. The trained models, called
solutions, provided a basis for receiving product recommendations that are filtered by custom attributes, such as project size, location, and room type. Once deployed, the model automatically learns from new interactions made by the electricians and retrains itself to help improve personalized recommendations. Another feature of Amazon Personalize, called exploration, allows the extending of personalized recommendations to include products with little to no prior interactions with electricians. This is helpful for Schneider Electric because the company constantly releases new products and expects its electricians to quickly adopt these products, instead of reusing existing or older products over and over again.

Creating a dataset group

The first step in creating a recommendation system using Amazon Personalize is to create a dataset group. A dataset group is a collection of items, including:

• Datasets such as users, items, and interactions
• • These are required to create solutions that are then used to generate recommendations.
• Event trackers for recording near-real-time user interactions to maintain the recommendation relevance over time
• Solutions that are models trained from existing Amazon Personalize recipes
• Campaigns that are the deployed solutions (trained models) that actually generate near-real-time recommendations for users
• Filters for filtering the recommendations based on custom criteria
• Batch inference jobs to get batch recommendations for a list of users

For the prototype, all three datasets (users, items, and interactions) were required to generate personalized product recommendations for electricians. Because Schneider Electric was using a third-party service to store the raw data collected from the mySchneider application, an automated data-preprocessing pipeline was set up to clean the raw data and use it to prepare the users, items, and interactions datasets with appropriate schema for the initial training of the recommendation system.

Choosing the right recipe and configuring solutions in Amazon Personalize

In Amazon Personalize, recipes are algorithms that are prepared to target specific use cases. Although Amazon Personalize offers different recipe types, the focus of the recommendation system created in the prototype was to provide product suggestions that are personalized for each of the electricians. For this reason, we chose the user-personalization recipe as a basis for solutions covering personalized recommendation scenarios. In addition, two other solutions were created based on popularity-count and similar-items recipes, respectively, to recommend the most popular products and products similar to ones recently viewed by the electrician. Solutions were then trained with the datasets that we created previously and finally deployed in campaigns to generate near-real-time recommendations.

Most viewed recommendations by electricians

When creating a quote for residential projects, Schneider Electric electricians typically deal with a number of project types (such as house or apartment) and room types (such as bedroom, kitchen, and living space). In an ideal world, when preparing a quotation, electricians always include the right set of products that are popular and relevant for a given project and room type. A typical example would be to include a socket with USB charger in a kitchen island to make it convenient for residents to charge their phone or laptop. However, Schneider Electric identified that the majority of the electricians included only products that they were already familiar with, in most cases ending with basic design and white color. Often, these products were less popular among the consumers, and opportunities were missed for getting the best out of what is available from Schneider Electric, such as more functionality of smart/connected devices, better design with new colors and new textures of glass or metal, or even sustainable products made from recycled plastic.

To help electricians make informed decisions in choosing the right products for a given project and room type, most viewed recommendations were added to the prototype. When an electrician chooses a project and room type in the quotation journey, the prototype will display a list of products that are viewed the most by other electricians for a given project and room type. As the project and room types change, so do the most viewed recommendations. This is achieved by including the project and room types as contexts when getting the recommendations from Amazon Personalize.

Location-based recommendations

To suggest products based on neighborhood (location) characteristics, the prototype used open-source external data providers to gather information on EV charging station count for all the cities across the country of France (which was the country of focus for the prototype) where Schneider Electric electricians had presence. Based on the information gathered, the solution would then suggest products from certain product categories (provided by Schneider Electric) whenever electricians specify the location of the residential project for which they are preparing a quote. For instance, if the project location is in a city where the specific rate is “moderate to high,” the solution would suggest products from the Motion Detection Sensor product category. Similarly, if a city has a high EV charging station count, this could be an indicator of an increasing popularity of EVs among the residents. So, the so

lution would then suggest products from the EV Charger product category.

Because both the specific neighborhood rate and EV charging station count are subject to change over time, the prototype includes a refresh job that gets initiated periodically by another feature of Amazon Personalize, called the dataset export. Amazon Personalize allows exporting interactions, items, or users dataset to a bucket in Amazon Simple Storage Service (Amazon S3), an object storage service offering industry-leading scalability, data availability, security, and performance. The export can include data imported in bulk (through the dataset import job) or data imported manually (through PutEvents, PutItems, or PutUsers API operations) or both. In the prototype, only the users dataset was configured to be exported periodically (with the help of Amazon EventBridge, a serverless event bus that helps you receive, filter, transform, route, and deliver events) because it contained a list of different cities in France where the electricians would typically undertake residential projects. The refresh job then retrieves and updates the latest specific neighborhood rate and EV charging station count data for these cities from the external data providers.

Recommending high-end products based on premium range and color selection

Oftentimes, customers are looking to provide personalized product suggestions that align with a business metric of their choice. Some examples of these business metrics include, but are not limited to, profit margin, increased adoption of newly released products, video watch time, and more. In addition to providing product suggestions based on the historical and near-real-time interactions data, Amazon Personalize can further optimize these recommendations to include products that satisfy a business metric.

For Schneider Electric, that business metric was to increase the adoption rate of new products by having the electricians choose products from its premium offerings wherever electricians saw fit. To align closely with the business metric, Schneider Electric wanted the majority of the personalized product suggestions from the recommendation system to contain premium products if an electrician chooses a premium product range and/or color while preparing a quote for residential projects. In Amazon Personalize, this is made possible by choosing the product price from the items dataset as an objective, while creating a solution. Once options are chosen, Amazon Personalize optimizes the recommendations to include products from the Schneider Electric premium offerings. In addition to choosing an objective, Amazon Personalize also allows controlling the sensitivity (off, low, medium, high) of the objective in order to balance product recommendations based on the business metric versus relevance through the interactions data.

Keeping the recommendations up to date

Product recommendations from Amazon Personalize could lose their relevance over time, unless the datasets and solutions powering these recommendations are updated periodically. In the prototype, all the new interactions made by electricians are captured and sent to Amazon Personalize in near-real-time using the
PutEvents API. This makes Amazon Personalize learn from the electricians’ most recent activity and maintain the relevance of the product suggestions over time.

Because the solutions used in the prototype to serve personalized product suggestions are based on the user-personalization recipe, Amazon Personalize automatically updates (retrains) these solutions every 2 hours, after which new interactions start to influence the recommendations. However, solutions created from other recipes need to be updated manually (by creating a new solution version and updating the existing campaign with it) to have them retrained with the new interactions and start influencing the recommendations.

Overall architecture

Though Amazon Personalize provides an API (GetRecommendations) for getting the personalized product suggestions, we built a web app prototype to showcase the real end-user experience: guiding electricians through the journey of selecting the right products with the help of the recommendation system for their residential projects. On top of that, we enhanced the recommendations by enriching the product suggestions with additional data (such as the product images) and implemented advanced filtering.

Impact

Using AWS, Schneider Electric has evaluated and uncovered opportunities for upselling and revenue growth by increasing the product adoption rate
during the project quotation eQuote process. The prototype demonstrated how a recommendation system can be used to provide personalized product suggestions in near real time based on product category, project geo-location, project type, and past user interactions.

Conclusion

“Using AWS Personalize service, we can test and validate an idea within a matter of weeks, instead of months or years. Such solutions will help companies like us to onboard and quickly integrate high value-add services into our software and apps,” says Shweta Singh, Software Business Owner at Home & Distribution at Schneider Electric.

Innovating in the energy management landscape, the same way that Amazon Personalize does for recommendations in the video streaming service, is for Schneider Electric the right approach to impact the way employees and partners improve customer experience.