Amazon directly attribute 30% of their total group revenue to product recommendations (Forrester). If you’re not already using product recommendations, that’s a lot of potential revenue to be missing out on. We speak to Senior Product Manager Ian Pollard and Global Support Manager Mark Ritchie to learn more about how customers can get started with product recommendations, and what makes dotdigital’s tools so unique.
Understand the value of product recommendations
dotdigital offers different categories of product recommendations. Could you give an overview of the differences between them?
Mark Ritchie: When it comes to product recommendations, most people are familiar with best sellers. They’re also perhaps familiar with most-viewed which is a good way of surfacing some different products that may not necessarily be best sellers, but they’re beginning to trend on the site.
In dotdigital Engagement Cloud, we also have trending which is unique as it’s a hybrid between the other two, and custom where you can curate your own recommendations. These four belong to a category known as non-personalized product recommendations, although this name can be misleading, as they are static lists but they are lists you can filter on.
Nearly all of our clients know and love our audience segment builder. They understand the power it gives them and they understand how granular they can get. One of the nicest aspects about the recommendation builder is that it uses that same technology as our audience segment builder, it just focuses on a different data set. So when we’re talking about a non-personalized recommendations, they’re not going to be completely unique to every individual, but you can still use that builder to narrow the results down to show a specific set of products.
I love coffee, so let’s say I now sell coffee, along with makers and accessories. If I use the static best-sellers list, I would have a mixed bag of products. I would have some beans, I would have some accessories and maybe I’d have a single-serve coffee maker included. It’s just a list of products and if I use that list and insert it into a campaign, I’m just hoping it catches the recipient’s eye. However, if I use the builder and add a bit more logic to the filter, I’m now customizing that recommendation to meet my segment’s demand. This means I can build a best-seller list just for single-serve coffee makers; or let’s say the segment I’m targeting is looking at people who have purchased espresso beans in the last 90 days — I can narrow my results down on that best-seller or most-viewed list to match that criteria too.
We also have personalized product recommendations and predictive product recommendations that use machine learning, so they change dynamically based on data stored against individual contacts. In these categories we have also-bought, best-next, and lookalikes.
Does this mean some product recommendation types are better than others?
Ian Pollard: Each recommendation type has a different role in the marketing mix. Custom, best-sellers, most-viewed, and trending are great product recommendations, but they are going to amplify the most visible trends, which are things you’re probably already aware of. Best sellers are your best sellers for a reason. But as Mark mentions, every recommendation type has advanced filtering rules. This means if you did want to amplify a particular brand or subset of products, you can create recommendations that give you exactly that kind of coverage.
On the machine learning side, the personalized and AI-powered predictive product recommendations are going to find more nuanced relationships between customers and products that you perhaps aren’t aware of and reveal implicit trends in the data that aren’t as obvious. That’s where the machine learning is able to come in and shine some new light on customer-product relationships.
We give you a full toolkit so you can deploy these recommendations in their entirety, because they’re not necessarily meant to be used in isolation. They form a menu of recommendations that give you full coverage, for your BAU emails, your post-purchase emails, your re-engagement emails, your abandoned cart emails — each of these is a different canvas for building different kinds of product recommendations. By giving you such a large selection and all the power of the recommendation builder, you get full coverage for revenue uplift.
What kind of businesses use product recommendations, and what are they looking to achieve with them?
Mark Ritchie: A common question we hear around product recommendations is how to fine-tune strategy for different customer types and products. I’ve seen it across a diverse set of industries, from apparel, food and beverage, skin and beauty products, to care services and more. Some clients have a very simple set-up. They want to market their best sellers or most-viewed products to make product discovery quicker, which in turn will drive more sales and revenue. Others have a more sophisticated set-up, utilizing our predictive recommendations.
Earlier in the week I came across a client in the apparel industry who we assisted in setting up a ‘complete the look’ post-purchase email. They were using the predictive recommendations to say, “here’s what you’ve bought, but did you know, you could buy this shirt and these pants to complete the entire look of what you’ve purchased?”. Of course, this is just one example of taking it to the next level using our product recommendations tool. At its core, it really allows you to market to shoppers based on their natural, instinctive behavior and show them the things that are relevant to them.
The way that a lot of clients were doing this before was just taking a guess. They had some data and bits of pieces from web behavior tracking or things they stored in data fields and they were making the best from what they had. But with the product recommendation tool, specifically with the predictive stuff, you really get a sense of what your customers want to see, and you’re able to market to them much more efficiently.
Ian Pollard: Because the AI-powered recommendations find much more discrete product relationships, the way I like to think about this is: If you were able to go through as a marketer and look at each contact and their order history, you could figure out what they like. You understand your products, and you could probably create micro-segments that have very high conversion rates. But you simply can’t do that at scale.
AI recommendations can surface those products that are a much tighter fit. It does this in two ways: One, by looking at lookalike products, so if someone has a certain affinity for certain brands or types of products, the machine learning looks for similar products that they have bought previously. The other way we surface is through best-next. That takes a more considered view of the customer journey. It looks at people with a similar purchase history and predicts the next product in that journey. If someone buys a tent, it’s able to suggest camping stoves, chairs, sleeping bags, and other things that other similar shoppers have purchased. As a marketer, if you had infinite time, you could build these. But AI gives you the ability to build that capability for very low effort, and much higher return.
Can B2B brands use product recommendations?
Ian Pollard: Absolutely. With our latest release, we became even stronger in our support for B2B brands. There’s nothing particular about B2B commerce that stops you using the full power of dotdigital product recommendations – the main difference is that you would expect there to be multiple product catalogs that have different prices for different types of businesses, and that’s absolutely fine. We support as many product catalogs as you like. So when you choose to create a product recommendation you would create it against a particular catalog which would therefore bring in all the of the appropriate product details and pricing for a B2B marketer. We’ve also recently released some additional functionality to our Magento 2 connector to capture specialized B2B data too.
Are there typical stumbling blocks that brands often come up against when looking to set up product recommendations?
Mark Ritchie: Product recommendations rely on data – that is accurate and contains everything required for the recommendation builder to display correctly, or at all. One of the challenges with ecommerce is that data evolves over time. When a story is passed down from person to person it changes a bit. This is why historical data typically degrades.
A product name or even SKU can change multiple times. For some product recommendations types, we have to marry different sets of data together and the most trivial difference can affect the outcome. Let’s say you have an order collection (which is all your orders) and you have a catalog collection (which is all your products) and you know your best seller line is pinot noir. In your order collection, it’s listed as pinot noir. But in your catalog, it’s just listed as pinot. Even though you know these are the same, the AI doesn’t know because it’s acting on the data. Fortunately, we get it. We understand the complexities of data and the challenges that brings to many companies.
Your data may not be in a perfect state, but we’ll help you get it there.
Firstly, we have connectors for an array of ecommerce platforms which most of our customers are already using. All the required information that dotdigital Engagement Cloud needs to power the product recommendations is already brought in with those, in the required format.
Ian and his team have also done a lot to minimize data errors. They have looked at matching other things like SKUs so the model knows these are the same items. They have also created automated tools that let people know where the data might be mismatched.
My team is also here to conduct data audits for customers, where we look at the structure of your catalog and order data and piece together what adjustments need to be made. You just need to reach out to your account manager to get that service from us.
Finally, we’re currently looking at building our product recommendations for customers to get the ball rolling and demonstrate the true value and power of product recommendations. We want to do this because we know that once people have the core elements there, they take it to the next level. We try to provide as much of a base for people to work with so they can build on it because this tool is extremely powerful. If you’re not using it, then you’re losing out on potential revenue.
Ian Pollard: Once you have the data sorted, the first step would be to make sure you’re using our ROI tracking so you can track the total revenue attributed to that product recommendation. You get some nice over-time reporting too. You need to be able to measure the effectiveness of what you’ve built and if it isn’t giving you the return you want, or the return dips, you can then go in and adjust the rules to re-optimize the recommendation.
Secondly, a little bit of forethought is needed by a customer to think about global rules – these are general rules that should apply to all of your recommendations. These might be as simple as a minimum price to a product that you want to recommend, or certain types of products like gift certificates that you never want to recommend. You can set these up at catalog level and any new or existing recommendations that use that catalog will inherit those global rules. That can be a big time saver, as you don’t want to duplicate that time and time again.
Our product recommendations are also multichannel. You only have to create them once but then it can be used in an email, on a landing page, or it can be deployed on your website as well. There’s a lot of recommendations right now and we’re working on more. It’s going to be different for each company where the best place to use those recommendations are. Considering them all and their multichannel reach in your strategy will put you in great stead, and our team are on hand should you need anything.
What can customers expect after implementing product recommendations in their campaigns?
Ian Pollard: There’s examples I’ve seen of customers reporting the average web session length increasing by as much as 50% for email clicks on our AI-powered recommendations vs. other email links. That tells me that the machine learning is working; it’s able to find these discrete customer/product relationships, and they are definitely engaging.
Mark Ritchie: We hope customers are also going to see how easy it is to use and how much time they’ll save using it. If they use it strategically and users take advantage of how powerful that recommendation builder can be, they will see an increase in statistics across the board. From engagements in campaigns to average order value, conversions, repeat purchases, ROI, brand awareness, the list goes on and on…
How you get started
If you’re a dotdigital customer, simply contact your account manager. Or, if you’d like to find out more about dotdigital Engagement Cloud, request a demo here.
Are you a dotdigital customer using Magento? We are offering product recommendation set-up completely free of charge for a select group of customers. Click below to register your interest and learn more.