Guillaume Cabane is a B2B marketing expert. He currently works full time as the VP of Growth at Segment, a company that makes it possible to engage in personalized marketing through the granular use of data from multiple sources.
Guillaume is especially passionate about tracking and attribution across multiple channels to deliver strong returns on investment.
His experience includes marketing leadership roles with organizations across a number of SaaS verticals including four years with Apple running business development for B2B online direct sales in the European market.
Here are some of the topics discussed in this episode:
Avoiding Predictive Touches
Single touches and specific actions of on user should not be treated as predictive indicators for the future actions of another, especially if they fall into different personas.
An engineer actually looking at the implementation dock is very good sign. And a CEO looking at the implementation dock, probably not a good sign. It’s not predictive of anything. We have to use that information in the context of who the user is.
With the right segmentation, it’s possible to score the touch points appropriately instead of attaching an arbitrary standard value to the action of an undefined user.
Segmentation is an essential part of the advertising and the personalization experience, Even after conversion the initial post-click landing page conversion personalization should be maintained throughout the rest of the funnel including the purchase as well as within the product itself.
The rest of the funnel is not optimized. You’ve pushed the conversion problem right to the bottom of your funnel because you created a personalized experience at the beginning. A dream, something fantastic. It was exactly what the customer wanted.
The product is where you spend the most time with the user. Using in-app copy that reinforcing the value proposition which triggered the original purchase helps maintain this level of personalization and improve the customer journey experience.
Solving Multi-Touch Attribution
It’s rare to find a company that is 100% confident in their multi-touch attribution model. More often than not, it’s not possible to identify the most important touch-point with a user that comes from retargeting during the top-of-funnel acquisition process.
You need to understand who is seeing your brand, how they’re seeing it, how they’re perceiving it, and where it is in relationship to their lifecycle with your product. You can’t just, randomly retarget folks hoping for the best. Most marketers just hope for the best.
One of the reasons multi-touch attribution is difficult is because of the legal constraints surrounding Personally Identifiable Information (PII). Guillaume advocates opening up advertising network data to third party analysts for independent verification of impressions.
He also thinks ad networks should providing marketers with an anonymous identifier for specific users that see an ad instead of providing the location where the ad impression occurred. Keeping the ad location confidential would still operate within the confines of PII.
Note: This interview has been lightly edited for clarity.
Ander: One of my favorite things about doing these interviews and going out to offices of other companies in the Bay area or wherever I may be, is getting to see the environment and see where marketing thought leaders and advertising thought leaders, work on a day-to-day basis.
And that’s where I am, here right now, in the Segment office with Guillaume Cabane, VP of Growth at Segment. How are you doing, Guillaume?
Guillaume: Hey Ander, I’m doing great today. People at the office usually call me ‘G’.
Ander: G, Ok! What should I call you for this interview?
Guillaume: G is great.
Ander: Alright, G. That’s great. Why don’t you tell us in a thirty-second nutshell how you got to what you’re doing here at Segment.
Guillaume: Yeah, so if you put myself as a product, right, I’ll give you a pitch.
Guillaume: I started working a very, very long time ago by today’s standards, in 1996 on a website which was about Mac games in France. And I got hired by Apple in 2004 and I worked at Apple in online B2B sales from 2004 to 2009. And let me tell you, it was a very different world. It was a world where people did not want to pay with their credit cards online. They wanted to call us and give that credit card.
Ander: Yeah. That’s pre-iPhone.
Guillaume: Yeah. And think how it has changed today if you asked for credit card numbers over the phone. It’s pretty rare. After that, I left and went into IT security as the CMO of a consulting company in France. Super interesting. That’s where I learned how to work with engineers. Banking crisis, I left. I created my own startup for about two years, raised some seed money, had lots of fun. I then moved on and I worked at Mention, which was my previous adventure in Paris, as the Head of Marketing and Growth. Super interesting. Super successful start-up with 500,000 users.
And I started using Segment at Mention. I started doing some talks and some meet-ups at Segment and got hired! And we are here. I arrived in the Bay area six months ago for Segment.
Ander: Isn’t it great when you get hired for a company that creates a product that you already know and love?
Guillaume: It’s amazing. I mean, I couldn’t dream of something better. First and foremost, I love the product. I have the unique chance of working with my product and enjoying that every single day.
And second, it’s also about the people, right? Because I sent Segment, when I was a customer, about one ticket every two days.
Ander: So they knew you quite well?
Guillaume: Oh, yes! Maybe they recruited me because it was too expensive to handle all those tickets.
Guillaume: But it also means I interacted with the people. I knew them and I knew how reactive and how kind they were. And the kind of tribe feeling that they expressed when they answered. I said, ‘Those are great people!’
It’s difficult to make the move from across the pond, going from Paris to San Francisco with your family.
Ander: Yeah, no kidding!
Guillaume: It’s a move that you think a lot about and if not for Segment, I would probably not have done it for anyone else. You need to have a lot of trust in the people who are hiring you, you need to make sure that they actually really want you, that you are going to be able to contribute a lot to their bottom line and be super helpful and contributive.
Ander: Of course. That you’re set up for success.
Guillaume: Exactly. So a lot things needed to align well. And they did.
Ander: That is awesome. And, as the VP of Growth here and as someone who’s used the product before, how would you describe the product very simply? The short sales pitch?
Guillaume: So, I’m not a salesperson but I can tell you my version of Segment. It’s a data bus. We take the data from where it happens, a property that you own, a marketing website, your mobile app, or your cloud sources, and we bus that data to where it’s valuable to you – so your e-mail provider, your advertising tools, your marketing automation tools, your analytics, your warehouse. That’s it! We just bus data around.
Ander: Perhaps you can give us a quick use case for how you might do that with AdWords or Facebook advertising.
Guillaume: Sure. There’s, like, a few interesting use cases of Segment for advertising. There’s the push and then there’s the pull.
So let’s take the ‘push’: You have people coming to your site, your mobile app, and you want to retarget them. You could just drop the cookie of the advertising platform. It’s actually way more efficient to go through Segment because you’re going be able to drop the cookies that you want, but you’re also and going to be able to push the user data to those platforms.
Job title, industry, and more. All those user properties you’re going be able to push to AdWords, to AdRoll, to Facebook and for a much better segmentation. And segmentation is everything when you think about performance.
Ander: It certainly is. And segmentation also is a tremendous part of providing a personalized advertising experience, which we Instapage know is important, and so do you.
Guillaume: Completely. The world has changed in the past few years where marketers are shifting from having the best message for everyone – which is very hard if you’re not Apple – to the perfect message for a small group – which is personas – to the perfect message for one individual based on what we know of that person and what the person is doing right now.
Ander: The ‘one size fits all’ approach doesn’t work anymore, at least in many contexts.
Guillaume: Totally. When I do some presentations about this topic, I like think about shoe size in the US. The median male shoe size median is 11, right? That’s only 20% of the population. That means that if you had to find the ‘one size fits all’ and you had that on your website, you would only show the shoe size 11 and all the others would leave. You would have an 80% person bounce rate – which is what people have on their websites. They have between 60-80% bounce rates.
Ander: And that also contributes to these ridiculously low conversion rates that you see across AdWords. The median conversion rate for AdWords is 2.35% or something close to that.
Guillaume: Exactly. And Google has been very good at explaining and forcing marketers at improving their quality with what they call the quality score, or how relevant are you with the intent of the user.
Google looks at bounce rates, they look at the data behind it, they look at the behavior of the user and as a result, they force us to be better. But one thing that’s very important, which I think most marketers don’t get, is we work mostly in silos. So you’ve got the person in charge of advertising and then you’ve got the person in charge of, like, CRO on the websites.
Ander: Channel specific, in other words.
Guillaume: And so the person in charge of advertising is going to build granular campaigns, get their conversion rates up, and their quality score up. So granular campaigns and dedicated post-click landing pages. And that’s great, but what comes after? It’s a generic product, right? You just dropped them on the product in the generic payment and generic emails.
So the rest of the funnel is not optimized. You’ve pushed the conversion problem right to the bottom of your funnel because you created a personalized experience at the beginning. A dream, something fantastic. It’s exactly what the customer wanted. And then, just after the customer gave you his personal details, you said, ‘Hey, now it’s going to be generic. Too bad.’
Ander: I totally understand what you’re saying and I could not agree with you more. But, is that more about actually customizing the product itself to meet their needs or is it customizing how you articulate the value proposition of the product?
Guillaume: Great feedback. A bit of both. Ideally, you change the product. But that’s very difficult.
Guillaume: Most companies don’t even change how the product looks or how you articulate the value proposition within the product. And it’s very easy. Think about it – you know the acquisition source, which campaign ad group and which creatives made the customer tick, and you know which post-click landing page they saw, right?
If it worked, why not use that data inside your product to show banners that reinforce the message that made the customer tick? It could be, ‘Sign up for free.’ Or it could be, ‘We’re the best post-click landing page tool for CEOs’. It could be anything.
I did that in the past. We actually had a reserved space in an app which reinforced the message from the acquisition, from the ad, within the product and down to the payments. And that increased the conversion rate 3X.
Ander: I am not surprised by that at all. Towards the top of the funnel a lot of the interaction happens on post-click landing pages as a natural extension of an ad. At Instapage, we’ve created a solution that allows you to scale that part of personalization. But how do you do that as they move down the funnel and into the product itself?
Guillaume: There are various tools you could use. It’s not that hard, once you have a system to capture that – and I don’t want to preach from my chapel – but honestly to take a shameless plug…
Ander: Shameless plugs are great!
Guillaume: If you use Segment, you will have the data from the acquisition in Segment. It’s going to be available and it’s going to push that data to all the tools, like A/B testing or personalization tools.
You need a personalization tool. The same way that you need a tool to generate your post-click landing page, you need a tool that accepts that data and can customize the experience. You could use a tool for your web experience – Optimizely is a good example. You could use a tool to customize the e-mail experience like Customer.io. You could do the same thing through notifications through ads too.
You’ve got to find the right tool for each channel that accepts the rich customer data. And then you’ve got to reserve some space in the app. We’re just talking of changing the value proposition, not changing the app itself. I have also changed the app itself in some cases. It requires a lot of work but it’s also possible.
One example of that… In the past, we had an app and we changed the payment page based on acquisition. We saw that in some cases we could derive the technology profile of the customer based on the ad that they clicked on, which showed us if they were more better at technology or not as good. If they’re better at technology, actually pushing PayPal helps conversion.
Guillaume: If they’re not as good at technology, pushing PayPal creates an issue. And so you’re way better off featuring different credit card possibilities or, like, other payment systems.
Ander: Very, very interesting. So, this is all kind of easy said and done when someone clicks on an ad, they get onto a post-click landing page, they download the ebook or access a lead magnet, whatever it is, and then they move throughout the funnel.
How do you do this with multi-touch attribution when there are many different touch points before you even have necessarily a qualified lead?
Guillaume: There’s multiple things in that question. I’m going to start with the end, which is qualifying the lead and that’s a problem mostly happening in B2B, which is, the world I know most because obviously I’ve been working B2B for the past 14 years now. So I know nothing about B2C to be very honest.
Ander: There’s nothing wrong with that.
Guillaume: Qualifying a lead requires you to build a model of ‘what does a good lead look like?’ It also means getting a lot of data points. We have built such a model and I’m going to tell you how it works.
We have connected Segment’s sign-up form with Clearbit, a company that provides an API. You send an email address and they send back between 20 and 120 different user traits and company traits, about company size, company funding, industry, and all of that? We have applied that to the past two years of sign-ups.
And we have another company come in, Kudu, which does some predictive modeling, built a regressive engine that tells us, ‘Hey, all of those who ended up paying Segment, what do they look like? What attributes do they have in common? How do they differ from the others?’ And they have built a scoring model and a user scoring model. That model is so good that it is able to find 60% of Segment’s revenue within 10% of all signups.
Guillaume: It reduces the noise by 90%? And that’s now a qualified lead. So I encourage everyone to do something similar in B2B.
Back to multi-touch attribution… Now that you know what a good customer looks like, you need to be able to track the user across multiple platforms and make sure that you understand that it’s the same user. That’s number one problem first, right?
Ander: Of course.
Guillaume: And then you need to create a strategy on how you communicate to that customer. Tracking is the hardest part. If you think of multi-touch as changing device, it’s actually very hard to track the same user if you don’t have an app, moving from one webpage on a computer to another webpage on a mobile phone. Unless you do fingerprinting, it’s impossible to track both together. It’s easier, actually, for people in B2C where they actually have affiliate links and they can pass the cookie through that.
Guillaume: Most of the time, the problem in B2B comes from long sale cycles – sale cycles that range from, like, two months to six months. And, you know customers are gonna go through multiple channels on the web. They’re going to discover you in SEO, then they’re going to get an emails, then back back through SEM, maybe retargeting, maybe they will attend a webinar, and you have to create an attribution model for that.
So if you use Segment – and, again, that’s a shameless plug – you’re going to be able, as long as they’re on the same device, to track all of those touches.
Guillaume: Which is great. And then you can understand which touch is important, which one actually is significant.
Ander: Which is significant, generally speaking? As far as that persona is concerned? Or for that specific individual?
Guillaume: You cannot apply the general rule. You cannot say, ‘Oh, I know this event, this touch is predictive and so I’m going to apply it to everyone.’ Based on personas, CEOs might react differently than engineers.
An engineer actually looking at the implementation dock is very good sign. And a CEO looking at the implementation dock is probably not a good sign.
It’s not predictive of anything. We have to use that information in the context of who the user is.
Ander: What’s something that you’ve observed with this multi-touch attribution specific to digital advertising?
Guillaume: No one does it correctly.
Ander: So what are they doing and how do they do it correctly?
Guillaume: I am yet to see a company that comes to me and says, ‘Multi-touch attribution, we got it covered. And we are very sure about our model and it makes a lot of sense.’
They all come to me and say, ‘Oh, we tried to do it and we failed.’ Or, ‘We did something. It works ok, but it has those caveats.’ Always.
For example, here at Segment we’re working on the U-shaped attribution model where we give 40% of the value of the first touch, 40% to the last touch, and we distribute the 20% in between all of the middle touches.
Guillaume: Usual in B2B, very usual.
Ander: In that case, what percentage of the touches towards the top of funnel come from ads?
Guillaume: A very low number, for Segment. Ads is not a good driver of revenue for Segment for contacts.
We stopped doing, like, 30 or 40% of our ads last month. Didn’t change anything. Usually when you want to test your attribution model, it’s useful to have control groups, to see if a model accurately describes or predicts what I think it should be. It’s hard to do control groups. I know a company that actually excludes Austin from their advertising completely. The city of Austin.
Ander: Why are they doing that?
Guillaume: Because there’s a lot of variables in control groups and it’s actually very hard to ensure that you’re excluding an audience from all marketing. It’s very hard. And you get what is the raw, organic interests for our product or our company? Because when you think of organic like SEO, it’s always influenced in some way by the other marketing impacts. It’s like how do you really measure PR or wellness, right? It’s very hard.
And so they actually excluded the location of Austin from any of their digital advertising. And so they know that Austin is only, like, organic. It’s only the brand that drives that. And they can see the market within Austin and compare that with other cities.
Ander: So, that is primarily a testing strategy.
Guillaume: Correct. Testing is important in the context of building a model that you trust.
Most of the time people don’t trust their model. And there’s a good reason for that. If you go back to the U-shaped model that I talked about, the 40-20-40 that we have at Segment, it has a huge power. We send emails, drip emails like most SaaS companies, to our customers. And of course there’s a high chance that the last event that was triggered before conversion is in an e-mail.
But, It does not mean that that e-mail is predictive or that it contributes. It’s just that it happens to be sent in an e-mail. For example, when a customer becomes active on Segment, we send a ‘thank you’ and a ‘congrats’ e-mail. But that has no impact. We could not send it and it would not change anything. It’s just that means we’re trying to be nice.
Guillaume: The fact that they opened that email did not help them towards becoming a paid customer. So if you count all email the same way you’re doing something wrong.
Ander: The same as if you count all ads the same way.
Guillaume: Exactly. Exactly. If you aggressively re-target all your audience, well, you’re gonna count in that ad’s traffic, a lot of people who would have come anyways. And there’s a huge problem behind that is…
Ander: And are you referring to ad impressions?
Guillaume: Impressions are a huge problem. I do not trust view-through impressions because I cannot measure them and I can’t control them.
View-through impressions means that you need to trust the ad platform and the numbers they are reporting that they are creating for you to pay them more. That’s what it is. They’re saying, ‘Yeah show us click-through, like, what you can measure is only 10% of what we actually help you to do. And all the rest, trust us! Continue to play the odds.’ I’m not very good at trusting.
I’ve already been there too long. But, I do recognize it exists and it has some importance.
When I was working at Apple in 2007, we had a whole business unit in charge of doing email. It’s a big business at Apple, especially back then.
I was in charge of other channels, and I remember that we had a meeting with Tim Cook and we took all the numbers together. I think we had something like 3 or 4X the total revenue that Apple Europe did. So we reported individually three times more and of course we had the attribution problem. So we started to debate, who should claim the revenue? How do we split it? I remember the email folks said ‘People who get the email, even though they don’t open it, there is a contribution to the sale.’
They actually had an experiment running where they had a control group of people who did not receive the email and then, on the other side, people who received the email but did not open it. They said the fact of seeing the title in your inbox, even though you don’t open it, creates a brain impression and that can have an impact. And they did show a very, very small uptake. But it was there.
Guillaume: Over a very large number, sure, it works and you have people who don’t click the images and stuff like that. And that’s the same thing. You have some kind of a metric which you have a hard time trusting. But you know it has some kind of an impact. So that’s the problem.
Ander: How much of using your gut or your instinct comes into all of this?
Guillaume: I’m not a big fan of view-through and all that. I hate making gut-based decisions. I don’t trust that. I really prefer hard numbers and anything that I can cannot measure, I usually discard.
Guillaume: If it exists, fine. But it’s just like organic and in some parts, SEO, you know? You can try to do things. It’s not scientific. It’s not something that you can just push a lever on.
Ander: We’ve identified throughout this conversation a few interesting problems that there are in this space, in marketing as a whole, as well as in digital advertising – one of them being multi-touch attribution issue.
Ander: And the other being view-through metrics and looking at how well they actually work. Given what you know and your experience with technology products and with marketing, do you think that there’s going to be a solution to this? Do you think someone’s gonna come up with that?
Guillaume: Sure. Two weeks ago, Google announced a huge change in their advertising platform. I don’t know if you saw that?
Ander: I just might have.
Guillaume: They actually changed their terms and conditions so that there is no longer a wall between their data collection business unit, which is Gmail, and their advertising business units, and that they will use the personal data from one into the other.
Guillaume: That’s a huge change if you think about the whole industry. The advertising industry, over these past ten years, has been very clear about what constitutes private individual information (PII) and what does not. Even though you take into account all the bad things we can say about the advertising industry, right now it’s pretty well protected, right? You cannot use PII or push PII in ads and hence the point of personalization.
Ad platforms like Facebook, which know everything about you, are starting to change the game, and the game could change more. These ad platforms could open to advertisers or to a third-party, the data on view-throughs so that I could have, as a marketing manager, a third-party that I trust that ensures me that the conversion did happen.
Ander: Interesting. In that case, you’re exclusively not relying on the ad platforms to provide you with accurate data. You have verification from that third-party.
Guillaume: Exactly. And as we move more and more into advertising marketplaces with real-time bidding, what you have already, those are the parties who do the bidding for brands. They could have access to the exact raw data that would enable marketers to trust the view-through conversion. You could go even further than that – and I see how that could cause problems – but you could give access of some of that information to the marketers. For example, if you, Ander, see one of my ads on a website, I’m not going to know it’s you.
Guillaume: I’m just going to see that I paid for one impression on one website. That’s it. However, it is possible that the ad platform tells me the opposite – not where you were without telling me who you are, but telling me who you are without telling me where you were. Because that would be PII.
Ander: Hmm. Ok.
Guillaume: So it would tell me, ‘Hey, you served an impression to Ander without telling me where, because that’s obviously sensitive. But then I would know, ‘Hey, this person, this cookie, who I got originally through this SEM campaign, and has viewed my pricing page and my product page, and did not or did sign up. Then I have their first name, last name, and emails… now I’m retargeting and I have served Ander ten ads, ten ad impressions, and you have not reacted. I should probably stop ads and change channels, right?
Ander: For me, specifically?
Guillaume: For you specifically. At the user level. At that point, I’m actually doing something which is detrimental to my brand. And you see that all the time. All the time you see those sites where you have the same brand buying all the ad space. It’s horrible. Right?
I can relate that to email marketing. How many times did you sign up for a service and you got bombarded with emails and you did not open any? It’s called ‘death by email.’ It’s not good.
Ander: No, it’s not.
Guillaume: We shouldn’t do that as marketers. After the third email that we send you that you do not open, we should stop doing emails. It’s not the right channel. It’s not working. And it’s not that after you receive five or six that you’re going to start opening them, right? You decided consciously that you did not want to open my emails. So I should stop email and I should change channels and go to ads. And I should serve you three, four, five, six impressions, maybe. Maybe ten, right? There a limit.
At a hundred impressions it’s worthless and just wasting money. And if I know that. I can then change channels again. When ads don’t work, I can go to push. When push does not work, I can go to text. When text does not work, I can go to calling.
Ander: Is it fair to say that you advocate for this personalization approach to not just the content that someone sees and how they’re and how they’re engaging with the platform, but when they’re seeing specific types of content?
Guillaume: Yeah, I advocate for personalization to understand the user lifecycle across all the channels, not only on the properties that you own including your website, your email.
You need to understand who is seeing your brand, how they’re seeing it, how they’re perceiving it, and where it is in relationship to their lifecycle with your product. You can’t just, randomly retarget folks hoping for the best. Most marketers just hope for the best.
Ander: Well, it’s the easiest thing to do.
Guillaume: Because costs are down. Costs are cheap. But what we’re not seeing, is that we’re actually doing something which is detrimental to the perception of the brand. You have to change channels and be smart about it. So I advocate for opening up all those channels in terms of data collection so that we can create better marketing. Because, even though the costs are low, there is a huge value to each individual and in handling that individual in the best possible way. That creates performance and that’s what we want.
Ander: And, that also humanizes us as individuals and personalizes our own experience.
Guillaume: Exactly. Each time that I have created a personalized message on a channel, I have seen a huge uptake in conversion. And usually people see that in email but if you can do that in ads you can serve personalized ads to one person. It’s more difficult but it’s possible.
You can totally personalize for the exact product which is what we see in the retargeting, when you get the product that you saw on Amazon, on the New York Times. But you also can do that for SaaS and B2B. If people selected a specific plan, if people declared that they are a CEO, or that they are an engineer, you should show them different ads. And if they are in the free trial, they should see a different ad.
You can combine all of that together to have very, very personalized retargeting ads to nudge them into the next step in your conversion funnel. Just like you do with email. It’s a discovery process. You’re not just trying to bring them back to your product saying, ‘Hey, we’re here. Don’t forget us.’
Ander: I completely agree with you. And it’s really exciting during these interviews to hear people validate what we’re saying, how much personalization is important, especially with the degree to which we can do it now, and how not enough marketers, not enough digital advertisers, are taking an advantage of this granularity.
Guillaume: And it’s mostly because most of the ad spending in done in B2C and they don’t have as many data points as we have in B2B. It’s still very cheap, honestly, just to blast a lot of ads, right?
In B2B, our world is very different because the market is limited, the costs per lead are way up and so that’s why you’re seeing that come from the B2B world because we’re trying, we’re struggling to optimize the return on investment on each lead. We need to improve the conversions a lot and that’s why we are doing that before B2C.
Ander: Awesome. Well, G, it’s been a pleasure talking with you. This has been fascinating conversation. I have personally learned a lot and I believe that members of our audience are going to absolutely love this as well. What’s the best way for, as a member of this audience, to learn more about Segment?
Guillaume: It’s an interesting question. We’re not very good yet at telling the world what we do in terms of advanced marketing. We do a lot of very advanced marketing and the problem is that there is a cost to sharing that – and the cost is the more you share, the less it works.
Ander: Of course. The secret sauce is secret for a reason.
Guillaume: We will, however, open some of it in the coming months to try and enlighten some of the digital marketers with the best things possible. So we will create recipes, growth hacking recipes that you can re-use and implement on your own site. The first recipe will be frictionless sign-up. We’re gonna publish that and if you use it, it will increase your sign-up rate by 30%.
Ander: Awesome. I’m going look forward to that, I’m sure that the rest of the world wide listening audience here will look forward to that as well.
Again, thank you so much for your time, G. It’s been a pleasure coming to your office.
Guillaume: Me too. Thanks, Ander.
Note: This interview has been lightly edited for clarity.