Chief Global Supply Chain Officer, Wayfair
Keynote: Beyond Transactions at Wayfair
Personalizing transactional email has become table stakes for e-commerce retailers of Wayfair's scope and experience. But as it ...
About Sascha Hower
In a September 2022 keynote at the Email Insider Summit, Sascha Hower, Chief Global Supply Chain Officer at Wayfair, discussed the company's approach to email personalization and its limitations. Hower described Wayfair as a platform serving the home goods market with over 33 million products and 23,000 suppliers, and noted that the company sends billions of messages to 27 million customers who made a purchase in the previous year. He stated that Wayfair builds and develops most of its technology and data science in-house, including algorithms for personalization and recommendations. Hower said the company uses machine learning models to predict customer behavior, such as clicks and unsubscribes, to determine optimal email frequency. Hower highlighted challenges with stale personalization, explaining that signals can become irrelevant if not decayed based on product consideration cycles and price points. He noted that during supply chain disruptions, Wayfair pauses notifications for customers experiencing poor post-order experiences until resolution and an apology. Hower also said the company is intentionally shifting from aggressive subject lines to more conversational tones for long-term brand accretion, and is building models to improve omni-channel messaging by centering the customer rather than the channel. He acknowledged that this omni-channel integration was still in early stages.
Transcript (43 segments)
S
Sascha Hower0:07
Good morning. Thanks, Steve. I promise we're not actually listening to what you and your wife are talking about in your home. That's not how it works. But I am super excited. This is my first Email Insider Summit. What a great group. I'm super excited to talk to you a bit about how we think about email and personalization and the limits of personalization at Wayfair.
So in terms of what's on the docket, I'm going to go through and give you a crash course in how we think about managing our email program at Wayfair, and then we're going to go far deeper into how we think about personalization and what the limits of personalization are when we're trying to build a story or a narrative within our email.
Now, one of the things as I've been hearing all the stories from different people over the past few days is that we all are operating in highly different contexts. And I think it's important here to level set on what context we're operating in at Wayfair, because that plays a really heavy hand in driving how we think about our strategy for the program and how we think about personalization.
So Wayfair is a platform that serves the home goods market. When I say platform, what I mean is we have suppliers who come to our site and upload products, then we have customers who come to the site, they buy the products, we deliver it to them. This presents a few different challenges for us. The first is we aim to be the destination for all things home, and home is a ginormous category. For us, we have over 33 million different products. One of the largest issues we need to solve is narrowing down that catalog so you're not going through this endless scroll of millions of products. We need to help you very quickly find that needle in the haystack.
Second, we have over 23,000 different suppliers who come to our site. As you can imagine, there are varying degrees of standardization across these suppliers, different ways they approach imagery, different tone and copy as they're describing their products. All of that stuff gets hydrated into my email, and it becomes very challenging to maintain a cohesive experience within the email when you lose control of all the little bits and pieces.
And then last but not least, we had 27 million customers place a purchase on Wayfair last year. We send billions of messages. It's mission critical for us that everything from our tech stack to how we approach solutions around personalization all has to be scalable. So that's our context. Let's talk a little bit about email at Wayfair. Our mission is to maximize the customer lifetime value of our lists by building deep and enduring one-to-one relationships with our customers. We do this by personalizing the entire portfolio of communications and then wrapping that in an on-brand, personable, enjoyable experience.
There are three different pillars that we're racing against to achieve this mission. The first is the foundational. We need a solid, scalable foundation. We need it to work on time as intended to get messages out the door. That's very basic but super important to us. Next is the experience piece. Admittedly, we are earlier in our evolution around crafting the right experience, but as we think about this channel serving as an extension of the brand, this is where we're placing a lot of attention these days. And then last, certainly not least, we think about personalization. When we talk about personalization, we think about it in terms of personalizing across the four rights. We want to get customers the right information at the right time when they're primed to receive and act on that information. We want to send it at the right frequency, so number of emails per week. And then we want to make sure if I've got a message to get out to the customer, I want to get it to you through the right channel for you.
All right, I'm going to talk a bit through the behind-the-scenes sausage making in terms of how we manage our program, because I think that helps it really come to life. The first question we ask ourselves is who are we going to send to? We have our customer list, and then we tend to follow industry best practices in terms of customers that we're going to pull and suppress because they've not been engaging with our email, so we pull them out. Second, especially in this COVID period, we have had massive supply chain disruptions. What that means is you went to buy a sofa, it's backordered, and it finally gets to you months later. That's a bad post-order experience. What we've learned is that for customers who are facing this bad post-order experience, it is not the right time to reach out to them and try to earn more business. So what we do is we snooze down notifications for customers who are currently in the midst of this bad experience. We wait until that has come to resolution, until we've had the chance to give them an empathetic apology, before we start sending to them again.
So that's who we send to. Next, when do we send? The first step there is how many emails should I be sending you in any given week? Here we have a series of machine learning models that will essentially predict whether any given customer is going to click and make a purchase or whether they are going to click and unsubscribe. Then we take that to determine the optimal frequency to maximize our revenue and minimize the costs. So if you are a customer who always eagerly awaits your Wayfair email, clicking around and opening it, I'm going to be sending you seven, eight, nine, ten emails every week. If on the other hand you've been lessening your engagement, you're not as likely to click and engage with the email, you might only be receiving one email a week.
So that's how frequently we send emails. Next is timing. We want to make sure that we are at the top of the inbox, front and center, when a customer is primed to act on that information. We personalize here based on your browse time. So if you are someone who tends to browse the site at 10 PM before you go to bed, we are going to get you that email top of your inbox at 10 PM before you go to bed so you're primed to act on it then.
So that's when we send. The next step is what do we actually go and send to the customer? First is the template. Here, form follows function. We want to make sure that the template we're serving will get you the best mix of information that you need based on where you're at in your purchase journey and your own individual shopping styles. There are three different components that determine the mix of stuff that we put in front of you: groupings of products, individual products, and then inspirational or editorial pieces of content. I'm going to go deeper into what we send on the next slide.
So there are five basic components that dictate what we send to a customer. First is the subject line. This is the front door of our email experience. We could have the most magical experience within our email, but if that subject line doesn't resonate, doesn't pop out for you, you're never going to open that email, you're never going to see the magic. So subject lines are very important to us. We personalize subject lines using a templated approach. Here the template is product class dash our top picks. Then we hydrate that subject line with the first module in the email, which here is an air fryer. So that template reads 'Air Fryers: Our Top Picks.' That's the subject line.
Next, we have individual products. If you have been actively searching for different air fryers, we again have this large catalog problem. We want to skinny that down and put in front of you the exact right one or two products that are going to help you close the deal and convert. Next is groupings of products or events. If we don't necessarily know you've been searching a ton, but we know you're looking for some place to sit, we would serve you a grouping of places to sit, so sofas and armchairs and stools, and then we'd aim to get you back to site so we can pick up better signal off of which to personalize. And then last here is banners or editorial content. The goal here is to provide an enjoyable experience. We tend to serve these primarily to customers who we don't have a ton of signal on, and the whole intent there is we get you to click, knock around our site, then we pick up more signal and we will send you more personalized emails going forward.
All three — SKUs, events, and banners — are powered by their own set of algorithms. But generally speaking, what they all aim to accomplish is they personalize what we're sending you based on your browse context or based on things that are generally popular across our list or to customers like you.
The last component that we personalize here is the template. If you are very in-market, very high intent, you've been actively searching for air fryers or beds, we're going to choose a template that gets specific product recommendations front and center. On the other end of the spectrum, you've not been shopping at Wayfair, we know far less about you, we're going to send you a more inspirational or editorial block of content up front.
So this we are personalizing based on customer segment or your level of intent. We're actually halfway through an initiative where we will start to personalize based on one-to-one signals. So if I know what your level of intent is and how you like to shop, I'm going to change that template to better suit your needs.
So that's how we think about personalization at Wayfair. Now let's talk about where things can go sideways. There are a few areas I'm going to talk about specifically. First, subject lines work really well until they don't. Second, stale personalization can do more harm than good. You have to be very careful in terms of how you're decaying your signal. And then finally, when you have a computer generating more and more parts of your program, you can very quickly lose control of the narrative and the experience.
All right, so we're going to take a look at some subject line examples in the wild. Here the subject line template is 'The [product class] Everyone's Buying.' I am searching for area rugs, so that is what's hydrated into the template. That's perfect. I am in fact searching for area rugs. I see that in my inbox, that's relevant to me, I click, I engage. Now here's where it can go sideways. Remember, we have a large catalog. In this instance, somebody is searching for a part of a toilet. So yes, the subject line is personalized and relevant to them, but this is more of an eyebrow raiser, not something we would necessarily as marketers choose to put out into the world: 'The Fixture Parts and Components Everyone's Buying.'
Next, this subject line is super straightforward: 'Inside This Email: Solid Wood Nightstands.' That's as straightforward as it gets. But again, I've got my toilet problem here: 'Inside: Toilets.' That's not what we would choose to put out there.
So what do we do with this? First, we craft the subject line templates quite carefully, paying attention to all of the different edge cases that we might bump up against when we think about the context in which we're operating. Second, you're never going to get all the edge cases, so we lean on business rules to restrict certain product and subject line combinations. And then last, we dog food our own subject lines. Everybody on my team has many, many email burner accounts, as I'm sure most of the folks here do. So we actively look in the wild to try to find weird, janky experiences and then solve for them with business rules. We also have 16,000 different employees at Wayfair who have no qualms about sending me any weird subject lines that they have received. So we welcome that and use that information as well.
All right, I'm going to talk quickly about stale personalization. The point of personalization is to be relevant. If I am searching for area rugs, I'm going to send area rugs to that customer. Now, if I just bought an area rug, what is the thing that is least relevant to me tomorrow? It's probably an area rug. So here is an example. This is my inbox. I was searching for area rugs on Wayfair, and over the next three months, Wayfair sent me 14, 15, 16 emails about area rugs. Now, if I had purchased that area rug at Wayfair, we would have immediately ingested that signal and stopped talking to you about area rugs. We would have started talking about complementary products like rug mats or side tables. But if I purchase that area rug somewhere else, we don't have a deterministic signal anymore, and so we have to rely on business rules to decay that signal.
So what do we do here? Here we decay the signal based on consideration cycle. If you are searching for something like a piece of decor, like a vase, that's a pretty quick consideration cycle, so we are going to decay that signal and stop talking to you about vases very quickly, in a matter of days. Contrast that with a vanity for your bathroom. That's a far bigger purchase. You probably have to take some measurements, you might have to contact a plumber or a contractor, so we're going to decay that signal far more slowly over time. One other thing we think about: if we don't always know the consideration cycle, you can use price as a proxy as well. You tend to make decisions for smaller price points far quicker than large price points.
All right, last but not least, controlling the narrative. This is an example of an email that is highly personalized to me. In the course of the same week, I was in fact searching for towel bars, air fryers, outdoor events, and dressers. Very personalized to me. What's happening here though is this is not a brand-forward experience. I'm going to close the transaction today, but this does nothing to create a deeper, more enduring, loyal relationship over time, which as I said at the beginning is part of our mission. We want to build a relationship with a person. We don't want the person building a relationship with a computer. Also, I'll call out here the subject line. 'Pro Tip' is the template: 'Check Out Our Components: 12-Inch Bar.' Also a bit of a bizarre subject line, and more importantly, that only speaks to a really tiny portion of what is in the email.
So what do we do with that? I think there are elegant technical solutions we could implement, and we are starting to work towards those. In the meantime, for some of our brands like AllModern, this is a far more pressing issue. They're in the middle of a specialty brand evolution. It's very important to them that everything they put out there upholds their brand, tells a narrative, builds a relationship with that customer. So what we've done here is we've created themed templates, and all of the themes tie back to the brand. For example, AllModern wants to be the authority in all things modern. So we create every quarter a number of different templates that are about authority. We've got top-rated products, we have buyer's picks, we have things that are community approved. We create this set of themes, and then we hydrate that with personalized information. So for that top-rated email, this person seems to be in market for things to sit on, and so we are hydrating this with top-rated seats. If you were, for example, in market for outdoor furniture, this email would be hydrated with outdoor furniture things that are also top picks. So this is an example of how we started to think about keeping the best of both worlds: having personalization and being hyper-relevant to a customer while also creating a cohesive, coherent, brand-forward experience.
And that is all I've got for you. I think we have time for questions.
A
Audience Member15:46
Thank you, Jess. That was great.
B
Ben15:59
Hi, my name is Ben. I'm from Braze. This is my first Email Insider Summit, but it's been a really lovely, intimate experience. I really appreciated the level of depth you went into in granularity when you talked about hydrating profiles to ensure that there was really deep-seated personalization in each touch point and being able to test and iterate on when things felt a little stale or outdated or things were decayed. I know one of the largest problems for high-volume senders is when you reach a certain level of scale, making sure that you're narrowing down to a particular SKU or product category. How do you guys deal with issues of scale as probably the largest spender in the home decor space? I'd love to hear a little bit more about that.
S
Sascha Hower16:49
Good question. We're on a journey there. We have a team of data scientists who are focused on product recommendations and being able to get you that exact right product. I will say we're pretty good compared to our peers, but we're probably not great if you compare us to someplace like Netflix. So what we do today is we think about you're in market for sofas, I'm going to get you a sofa. We don't have a way easily to think about what style might be particularly interesting to you, what price point. So over the next 12, 18, 24 months, we're starting to focus far more on getting that exact right product for you, not just a type of product that is top-rated or popular. It's not an easy problem to solve with a catalog of our size though.
J
Jeff Foley17:34
Hi, Jess. Jeff Foley, Validity. Love the presentation. Fascinating. I'm wondering how you split up the work then with all the stuff procedurally generated. Do you find that the work is less the traditional 'I need to plan this campaign' and more 'I need to think about templates'? It seems like a different breed of email marketer that you need.
S
Sascha Hower17:57
I think that's right. So we've got across our entire team about a hundred people, and they do everything from our data science to the creative to the typical email marketing. Across the core email marketers, nobody on my team actually came from an email marketing experience, which is interesting. Part of the reason why I love coming here is because I get to meet with people who do this as and have been doing this for years. Most of the people on my team have very technical backgrounds, very product forward, very good at digging into how should I be building this template, what will that mean in terms of the customer experience. I think as we start to focus more on building a more typical customer experience, we're going to have more creative marketers come in and start to sort of sand off the edges a bit, and we will shift from being very technical focused to being far more brand forward, if that makes sense.
C
Chester Bullock18:53
Chester Bullock with RentPath. My question is more around how you're achieving this personalization and these recommendations. I think you just alluded to it. You said you got a team of data scientists. So are your algorithms developed in-house? Are you guys using an outside AI tool? How are you achieving all this?
S
Sascha Hower19:13
Good question. It's in-house. Wayfair has a heritage of building and doing most of our tech and our data science in-house. This one, on the data science side, it's absolutely done in-house.
J
Josh19:26
Hi, Josh from HBR. You mentioned segmenting out people who have a topic affinity versus people who are engaged but don't have a specific affinity. What have you found about relative conversion rates and order value from those two segments? How much of a benefit is it to really know specifically someone's area of interest?
S
Sascha Hower19:48
It's a huge benefit. When we're able to get very specific and start recommending individual products to a customer, we see conversion and revenue and engagement, frankly everything, is far, far greater for those segments. Once you get less engaged, our signals aren't as strong. It's more like getting more shots on goal, like am I going to surface something that is relevant to you and get you to interact? So we actually see conversion rates are far lower with editorial content.
A
Audience Member20:19
Jess, one of the things that I noticed in some of the templates that you're using and in the subject lines is that you seem to be trying to, and I'm wondering if you're doing this self-consciously, mimic editorial. So that you're sounding more like a buyer's guide and you're sounding less like an e-commerce company. I noticed in some of the templates where you're putting content up, it's feeling more like an independent buyer's guide than simply another recommended push from Wayfair. I'm wondering if that's being done self-consciously.
S
Sascha Hower20:51
Yeah, that is. I think it depends on the brand. Wayfair is our main brand, but we have four specialty retail and luxury brands: Perigold, AllModern, Joss & Main, and Birch Lane. Especially for the specialty programs, we find it's mission critical to be more editorial, to tell more of a story. That's what our competitors are doing in that space, and so that is far more intentional. On the Wayfair side, we are doing it intentionally, but part of what we're also doing is trying to tone down our subject lines, which have historically been quite loud and in your face. That's been very performant in the short term, but over the long haul, we think that might actually be less accretive to the brand. So part of what we're doing is just trying to be more conversational and less loud.
A
Audience Member21:37
Are there new and different metrics that you're applying as you're engaging these different approaches? Do you have different numbers and metrics that you're looking for to see if they're actually working?
S
Sascha Hower21:48
The hardest thing for us right now is customer lifetime value and being able to measure that. If I put a transactional email forward, I'm looking to get you to convert and close the deal the next day. That experience is always going to win out versus a longer-term play where I am building a brand and hoping to inspire you and generate loyalty over a long period of time. So we have to rely on some type of customer lifetime value indicator to prove out whether this longer experience is actually more accretive than the short-term experience. But it's a hot topic. We are actively trying to think about ways we can determine lifetime value for our space, which is people purchase far less frequently, so it's a bit of a harder problem to solve.
A
Audience Member22:33
And I want to lean forward into our next presentation by asking you about how this integrates with your app strategy. Since a lot of us use the Wayfair app and we're getting on sort of an alert cadence as well, how are you thinking about how email integrates within your app users and how you coordinate all that?
S
Sascha Hower22:53
So from a transactional sense, if you've purchased something and we are messaging you about that, we put in place business rules according to how urgent that message is, how quickly we need to get in front of you. We don't actually use push a ton for that, but you might use SMS if I need you to be home and accept the delivery, whereas if I'm just telling you something shipped, we'll more likely use email. So we're starting to do that in the transactional space. We are building models to get better at omni-channel messaging, putting the customer at the center versus the channel at the center, which is historically how we've done it. So you may or may not see the exact same message get blasted to you through all different channels. We want to flip that around and make sure that the channels work together in concert and we're able to send you the right message through the right channel for you as an individual or based on our business needs. But we're in early innings there. I think we'll make a lot of progress in the next year.
A
Audience Member23:50
So you're still working... I mean, the emails are taking account of shopping patterns and shopping preferences, but the integration of mobile not quite yet. You're not quite to the point where you're knowing whether people prefer to use their app as opposed to...?
S
Sascha Hower24:03
Correct. And we're personalizing everything off of the same signal, so you could end up with the same message through many channels.
A
Audience Member24:13
Jess, that was fantastic. Thank you so much.