Avail’s R&D team is just about to roll out all the features in the latest product release. Still, Avail CTO and co-founder Henrik Schinzel was kind enough to answer some questions about what the new features mean to merchandisers.
1. The HTML Control Panel
Q: Java or HTML – does it matter to a merchandiser?A: Yes. The reasoning behind why we move to a web application (in HTML, CSS and Javascript) is the same as for why we once chose Java: to create the best user experience, regardless of platform. Java then promised the same things web applications deliver today – really, Java failed to live up to its promise. For example, there are several combinations of Java and operating systems that are not 100% compatible. Some large companies also have policies that block Java applications.
Web technology enables us to release faster and more frequently, to continuously enhance the usability. There are also less access problems, and since more and more applications are web-based these days, the user interface should be recognizable and easy to learn. That’s important to a merchandiser, especially someone who is new to our product.
Q: Have you made any updates to the user interface itself?A: We tried to stay close to the old Control Panel interface initially, so customers wouldn’t get lost. However, we did take the chance to make some small changes. For example, there are tool-tip previews of template settings and you can scroll your graphs just like in Google Analytics.
Q: Why a Control Panel – isn’t the recommendation engine automatic?A: In the early days of recommendation engines, they were considered black boxes. The Control Panel was like a “fuse box” – only to be opened in case of emergency.
But solutions have matured. Today, a better metaphor would be a “cockpit” – rich metrics to aid in the decision-making, and controls in case you wish to make changes, but still a very high degree of automation. The reason is we’re no longer just developing a recommendation engine – Avail Behavioral Merchandising is a merchandising platform, where the recommendation engine is a key component. A merchandising platform should be a toolbox that empowers merchandisers to work more effectively and efficiently in as many situations as possible.
2. “EDU” – Enhanced Discovery & Upselling
Q: Could you describe what EDU means?A: Discovery and upselling are steps in the merchandising cycle – first you need to help the user find what they are looking for (the reason they went to the shop in the first place) and then you have the chance to upsell them to a higher-margin alternative. Historically, recommendation engines have been more focused on what happens once you have decided to buy something, cross-selling.
We poured over the problem, and came up with a new algorithm for this purpose, primarily based on click data. I won’t get too detailed, but what’s interesting is that it understands sequences and is self-learning. Sequences means the computer knows that the order in which things occur matter. Self-learning means it constantly learns by its mistakes and successes, to become better and better over time.
Q: What is an algorithm – and does a merchandiser need to know?A: Algorithms are “recipes” for how to solve a problem. Listing what “Other people who viewed this item also viewed” to find other products that interest a shopper is one algorithm. The way you solve a task like “Find the largest number in this list” is another. You may not be able to explain what your algorithm is – but you used one.
Of course, the whole idea with what we do is to prevent that merchandisers need a PhD in Mathematics. Avail takes care of developing the best algorithms for you. However, a merchandiser can benefit from understanding the limits and opportunities of various algorithms. For example, in book retailing, if two different books were bought together or in separate orders may not really matter. But in e.g. consumer electronics, it’s different – if a lot of people who bought a computer also bought a TV, that doesn’t mean that the best recommendation to someone who’s just added a computer to his order is a TV. So as a consumer electronics merchandiser, it really helps to understand order data is essential information.
Q: Is it really that hard to find out what “Other people also viewed?”A: No. However, what you need to realize is that is just how it’s worded to the consumer. If you look at what people who viewed a certain product most often also viewed, it is quite likely a best seller. People who viewed books by Stieg Larsson may also have viewed Barrack Obama’s biography, but it’s not a good alternative to recommend. It’s not specific.
What Avail trains our algorithms to do is find the unusual recommendations, which makes them relevant and interesting. People who view Stieg Larsson books, unusually, also view books by e.g. Sjöwall/Wahlöö (another Swedish crime author). That’s a helpful, personal product recommendation. Finding the algorithm to accomplish that, without losing performance at peak traffic, is not trivial.
3. Other functionality
Q: Any other new features you’d like to highlight?A: We are in the process of also rolling out a shared API (program interface) for all recommendations, replacing the several different functions we have today. This allows you to change how recommendations are applied without having to bother the IT department. As part of that, we’re also adding the possibility to design recommendation logic using very natural language, e.g. “People who bought this item also bought”.
