This post has been republished via RSS; it originally appeared at: Cloud Perspectives Blog.
Ever shopped online? Watched streaming video? Listened to a music service? If so, youve experienced recommendation engines.
These fascinating digital tools use statistics and algorithms to analyze your use history and digital footprint to figure out what you like and give you more of it. Before this technology became a standard feature of e-commerce, recommendation engines were custom-built, internal toolsand often experimental. They were only accessible to companies with dedicated data science teams and technical specialists.
Thats no longer the case. Data science advancements and cloud technology now make it possible for almost any company to adopt recommendation engines. Using mobile and Internet of Things (IoT) technology, recommendations can be integrated with in-store shopping, giving customers a seamless, omnichannel experience. Heres how, along with a few tips on jump starting this technology at your business.
Rapid rollout, real results
As customer demand for personalized experiences continues to grow, data-driven tools such as recommendation engines are becoming essential to maintaining a competitive edge. As Satya Nadella, CEO of Microsoft, put it, Its not enough to know whats happening now in your businessyou have to anticipate what will happen, then be prepared to capitalize on that insight.
Fashion retailer ASOS exemplifies this type of thinking. To fulfill its mission to be the worlds number-one fashion destination for twentysomethings, the company has created a recommendation engine to provide customized suggestions to its 15.4 million customers.Using cloud-based data storage and analytics, it has created a scalable solution with transformative impact on the customer experience. In a world where we have 85,000 products on the site and 5,000 products going live each week, we need to make sure that the right subset of those products is in front of our consumers, says Bob Strudwick, Chief Technology Officer at ASOS. Now, the products and content will be more relevant to you as a shopper.
You dont have to be a global enterprise to achieve great results with a recommendation engine. Here are four pointers that apply regardless of your level of ambition.
1) Model: Choosing an approach
The idea behind a recommendation engine is to filter a large set of options down to those with maximum relevance to a given user. There are several ways to do this: content-based filtering, collaborative filtering, or a mix of the two.
Content-based filtering recommends items based on their similarity to what an individual user purchases or likes. This approach works well when you have detailed data about the features of the items you want to recommend. It relies on specific information about the product more than insights about users. Upside: Recommendations dont rely as much on customer ratings as on the product descriptions, so new items added to store inventory can easily rise into a recommended products list. Downside: Filtering is based on a limited user profile, so conclusions about user preferences are drawn from fewer actions. This makes targeting potentially less accurate.
Collaborative filtering makes recommendations based on similarities between users and how they rate products. This is a good approach when you have data on both the users and similar shoppers behaviors and interests. It requires more data to start with but can be expanded more easily to new scenarios. Upside: Collaborative filtering provides suggestions that are more likely to reflect the customers tastes. Downsides: It takes more data-crunching capacity than content-based filtering; suggestions made to new shoppers with little e-commerce history may not be as accurate; and new items added to inventory may not make a recommendation list, since they need a critical mass of ratings by users first.
Bottom line: If you have robust data about users and how they rate products, you might consider collaborative filtering first. If most of your information is about products themselves, content-based filtering might be the better choice.
2) Data: go big
Regardless of your filtering approach, more high-quality data is better when it comes to machine learning. If you run an e-commerce site, you may already have detailed information about product features and user behavior. Loyalty programs, customer relationship management (CRM) systems, and enterprise resource planning (ERP) software can also provide the data needed to train recommendation engines.
Combining data about users and products with additional sources can increase the accuracy of recommendations. For example, demographic data or information from social media can be analyzed to identify user preferences more accurately. Just remember: Your choices here depend on what users are willing to share, and your ability to meet security and privacy regulations.
3) Platform: build on the cloud
The math behind product recommendations requires a lot of computing horsepower, especially when youre looking at large numbers of customers and items. Cloud-based systems are the logical choice because they can scale instantly to meet varying demand. Theres no need for you deploy and manage your own high-performance computing infrastructure.
Another efficiency is to use powerful, fully managed machine learning services from the cloud. Many offer drag-and-drop simplicity, so you can move your rec engine from idea to implementation much faster. If youre running a team thats short on data science expertise, this can jump-start your project and keep it from stalling.
4) Channel: improving experiences everywhere
Do you fully understand how your customers shop? What are they looking for? Where do they get their information? What data are they willing to give you, and how can you build trust with them? All of these questions are important to ask as you begin the journey of automating product suggestions.
Although many recommendation engines work on single platforms, they have the potential to impact the entire customer journey. JJ Foods, the United Kingdoms largest independent food distributor, succeeded by integrating recommendations into all customer touchpointsincluding its contact center and CRM system. Other companies are incorporating recommendations into their brick-and-mortar stores through kiosks, location-aware consumer apps, and point-of-sale devices used by salespeople.
Getting started is easier than you think
Theres no need to start from scratch in building your new recommendation system. Companies worldwide are taking advantage of prebuilt cloud solutions to accelerate results. And thats a pretty strong recommendation.
To stay up to date on the latest news about Microsofts work in the cloud, bookmark this blog and follow us on Twitter, Facebook, and LinkedIn.
The post Smarter shopping with AI appeared first on Cloud Perspectives Blog.