WHAT the Data?!

Jean de Bressy - How to leverage data for marketing automation? (Zalando) (#17)

• Lior Barak and Jean de Bressy • Episode 17

Share episode

Hello and welcome to another episode of WHAT the Data?! Podcast, and today we have for you a great episode with Jean de Bressy.
After growing as a product manager and marketeer at trivago, Jean moved to Zalando three years ago to work on expanding performance marketing activities through automation. A very keen reader, he now focuses on making the most of the available data in-house to optimize conversions.

In this episode, we got to know more about Jean's view on data-driven decisions, his experience with data while working for Zalando, and his view on the future of marketing automation.

If you want to get to know our guest, even more, you can follow Jean here: 
https://www.linkedin.com/in/jeandebressy/

-------

As always if you enjoyed this episode, please like, subscribe, and share it with your network so we can outreach new audiences.

Jean de Bressy  0:07  
Welcome to the what the data podcast with your hosts, Mitch and Leo,

Lior Barak  0:23  
you are listening to another episode of what the data podcast. I want to start with a small apology. In the past few weeks we have not released episodes in the manner that we were planning to so every second week, me Ken and I were a little bit overwhelmed with work and we needed to attend that we are back on track and I hope more episodes gonna come very soon. In today's episode, we will talk to Johnny brasserie, a product leader zalando biggest fashion ecommerce in Europe solving the marketing automation challenge to optimize each euro spent to maximize the return. With john, we have talked about marketing automation future with GDPR, what companies should do with the wish to start automation tomorrow, how to trust your models and data during a pandemic when everything goes like crazy out there. And I must say in a personal note, I really enjoyed this conversation with john, who I think is one of the leading minds when it coming to combining tech and business to drive better marketing automation. If you wish to learn more, tune in to this episode. Hi, I'm happy to have you on the show today. How's it going? Very good. Thanks for having me. You're in a long time. Thank you very much for having me. It's been a very long time. I'm so excited actually to have you here. Because I think that you are among the very few people who understand how to work with marketing automation today. 

Jean de Bressy  1:49  
Yeah. Okay. Thank you very much for this introduction, I think. Yes, thanks. 

Lior Barak  1:56  
So tell us a little bit. What are you doing today? And how do you drive automation in your organization?

Jean de Bressy  2:02  
 Yes. So I'm leading the Facebook Performance Team at Zalando. So I've been doing that for a few years. I'm actually a product manager. What we do in the team is we responsible for all of the performance marketing on Facebook, in zalando, globally, so all of the markets. And my team is composed of product managers and engineers. And what we do is we automate decisions and marketing at scale. And we try to make in parallel, we try to make the best investment decisions and build campaigns that will work for the entirety of Europe for us. 

Lior Barak  2:34  
As a product manager during performance, did you feel any gap when you started doing that? Or was it an easy intro for you? 

Jean de Bressy  2:44  
I will. So I come from a world where I was I was a product manager before working or you know, front end tools. And there was already I was working with marketing departments. But then you do different you look at different types of data. So you do a lot of user testing, qualitative data, you look at a hot jar, like heat maps and these kind of things. And you pretty much try to understand what's happening with the customers.And when you start doing marketing, you look at a lot more quantitative data, very big scales, very big numbers, you're looking at your impression numbers, and we're trying to build features and build products that are based on really bigger sets of data. And automation is the same. So the first thing that we're working on is automating decision making for investment. So how much should you invest in campaigns? And you know, there's no front end to it, you're just looking at data and making decisions based on what your data scientists say, which was a completely different exercises what I was doing before,

Lior Barak  3:40  
 so you're trusting your data scientists to actually tell you if your performance are good or not? What data actually do you receive on a daily basis? So what are you looking at daily that can help you to make the decision if you should invest more or less? 

Jean de Bressy  3:55  
Yes, so the data set that we look at, of course, you have all of the performance metrics from the channel. So regardless, from the channel that we're using, right, you have your impressions, cost per impressions, because per mil CPC cost per order and everything that the channel gives you. That's part one, you can add to this more signals that you look into because you can receive them via API. And then Part two is everything that we see on our side. And we have teams that are actually in charge of our attribution model and reallocating value to the campaigns, we add the notion of customer lifetime value on top of the retributive matrix that we see. We also have a forecast for prospective return on investment because once you know, if we sell something, we have to take into account the return rate of items before we can know what is actually left for us in terms of sales. And so we have to forecasts this return rate and the actual value of marketing today so that we can make a decision before the return actually occurs. So there's this on top of the attribution and the left hand value. And then finally, we reach a point where we can see a return on investment. And on top of this has less than we add correction for the level of incrementality of our campaigns. So it's a very long funnel, it's pretty complex, there's several teams maintaining it. And yes, on a daily basis, what we get as the output is the return on investment of each of our campaigns. And we use this as another input metric for our Moodle that decides how much to invest in your campaigns. I hope this is clear, I can also dive deeper.

Lior Barak  5:30  
If we looking at your return of investment, so ROI, is this a forecasted ROI? Or you're just taking the measures that you you just mentioned, and doing some kind of calculation on your orders from a certain day? 

Jean de Bressy  5:44  
Yes, it's because of course, you know, as I said, right, if you sell 100 euros worth of items as an equal, you have to take into consideration how much we'll come back to you, before you can know what's left. So the ROI for us is always forecasted. And then it takes a certain amount of time before we can look back and say, Okay, this is what we forecasted. And this is where the actual value is. And so we can, you know, our models adjust themselves, because we look at the difference between the actuals and the forecast to make sure our forecast is very accurate. One of the goals of their evaluation teams to make sure that the forecast is as close as possible to reality. 

Lior Barak  6:21  
Interesting. in the field that you're working today, you're creating quite a lot of creatives, or is it purely using data? using the products and based on it, it's promoting or not promoting stuff?

Jean de Bressy  6:35  
 We do we do both, right. So there's different types of automation that we that we work on, as a team. Because there's the one part is high level decision making, understanding what's the value of your campaigns, and then being able to say how much you should invest in a campaign. And that's, so there's a very long data infrastructure behind it very complex, it's also very big investment for even a company like zalando, it's been years that we're working on this and fine tuning the model just so that we can make the correct investment decisions, which is the key, it's really the key to understanding performance in big online players. And then in the team. On top of this, we have a layer, which is okay, what creatives do we run? What are the campaigns and what should we automate? I mean, there's a lot of it that is already done, because we it's an e commerce. So you can use advanced products that are also automated, like dynamic ads. But we do work also on automation for selecting items, making sure we can present ads that people will like and then optimizing them over time.

Lior Barak  7:35  
If we're looking today, at your learning curve in the past three years, has it shifted a lot from the way that you're looking at data, or the way that you're understanding the use of data, 

Jean de Bressy  7:46  
Many things have changed. And I think what I mentioned before is that you stop looking at quantitative and you look at quantitative, very high level, lots of numbers. But I think the main revolution for marketing or my thinking on data is that you can't trust what the channel is telling you. So having only the story of the channel that tells you Okay, here's your return on advertising, spand means nothing, until you have checked it for incrementality. And along those big differentiating proposition in performance marketing is to look at incrementality. That means checking what the causal impact of marketing is, you need other tools to actually check it. 

Lior Barak  8:24  
That's, that's super interesting. Because if we're looking at most companies out there that at least at Facebook, and Google today is a big channels, they trying to close the market, right? They're trying to close the opportunity to marketers, to actually have control over their marketing. What is your What is your view about it, actually. 

Jean de Bressy  8:44  
And so, I mean, personally, I tend to agree with that statement. But I had, the more I had had a closer view to what you're saying maybe two years ago, because the features on Facebook are really going in the way of mass market and mass automation by default. So if you take, you know, all of the creative optimization features that they're doing, they're going in this in the direction of GDN as well, which is saying, okay, just give us the assets, and then we'll figure out what performs best and we won't even tell you how that works. And we're just gonna optimize it on our Facebook is also very big proponent of saying, only do broad targeting Don't you know, don't play with the targeting don't do anything, just target broad. That's what works best. For my view on this was there's no sir. It's a way for them to limit your capacity to tailor ads and do things they're refined. But I have to say in the past two years, like we still can do quite a lot of things with Facebook's API, I think more than with Google. So we're looking at things like in display, Facebook still allows you to do quite a lot of things. There's a very healthy ecosystem of third parties. And there's also places now it kind of moves where automation is going because if you can rely on Facebook to do ad set budget allocation for instance, and do that automatically, you can check that it's okay, there is one less decision you have to make. And it opens an opportunity somewhere else. So for instance, creative decision making process to understand what's going to be the performance of creatives. There's a lot of new companies being built here in that very space, just to understand, Okay, can we build models to see the future prospective performance, you know, estimate the performance of the creatives before you upload them, and then use that to just take better creatives to your ads, on average, right, so that all of the creatives you put into your campaigns are better, and I think so can move where you can automate. That's how I see it. Now, there is always kind of a space where there can be innovation, it's just that it gets reintegrated later on in the big channels, 

Lior Barak  10:44  
Are you afraid that at some point, you will just drop a budget like Google tried to do it couple of years ago that you just give them the budget and they optimizing everything for you, and you don't have any control over the campaigns? Are you afraid of

the direction they're working at that's, that's for sure. And the way to understand it for me is, they do that because they have millions of customers, right? So at our ends, or if you're very advanced in marketing, this is annoying, because they're removing your decision making power. And they're saying, don't do anything, we're automating it for you. But if you're thinking of the small shop, this is the story of Facebook, they also always say that if you if you think about the small shop that's opening a Facebook store, and wants to spend a bit of money on the business manager, then you're 50 years old shop owner doesn't know how to maximize, you know, its revenue. So a feature that are everything for him, or her is extremely good. For us, it's a different story. And I think, I mean, these companies are aware of it, they know that if they do a feature that helps the SMBs a lot, it's not going to necessarily help the big businesses. And I think they're going to need to have kind of tiered approach where they can serve big businesses and companies that are very advanced with data a bit differently than what they do SMBs, which, you know, the feelings that they kind of have a one size fits all solution. And they just ship something so that it makes it easier for the long tail of advertisers but still having to be prayers. 

If we talking more about marketing automation, do you think this is something that small companies should go into this direction? Should they try to even automate their processes? Is it worth it?

Jean de Bressy  12:21  
Yes. I mean, the simple answer is that it's always worth it. But it's also mitigated. And so I'd be measured in what I would say, because it really depends what type of data the company have, or companies have, what type of marketing are they doing, you know, is it automatable? Is it easy to automate, because a small company oftentimes doesn't have enough resources to have a full time Python back anymore full stack engineer working not on their product, but working on their attribution, for instance. And it's a very big decision to make when you're a small company. But I think the earlier you make the decision to automate your marketing do better. And I will start always, by having a good stack from your data, that is the most important. So if you run your attribution once you you get to the size where it's okay to do that. Or if you clean your data properly, and you can look into it and make correct decisions. That's the most important I think most companies even medium size, don't always do that correctly. Like they do that. Or they do that very late. And I think that, you know, if you do that five years after you were created in we already have a big marketing team. It's a bit too late, like you're slowing down the work of marketers by not having the proper data. 

Lior Barak  13:37  
I like to to touch the Python developer for for starting creating automation. Can you maybe describe how a marketing automation team should look like? if somebody wants to start it tomorrow? What does he need to have in his checklist?

Jean de Bressy  13:49  
So if you're doing if you're doing that for all of your channels with different than if you're specializing it for our channel specific, but I would say you just start by looking at what are your top channels, what their API's look like? And what is it you know, what are the main buckets where you spend your money, and then you can prioritize decisions. So if you have creatives that are similar overtime, if you have processes that you can repeat, this is what you get, these are the things that you can automate first. And it's pretty easy to understand that you can just look at what are the decisions that marketing managers make on a weekly basis, and then what is repeated all the time, that's where you have to make first and then there will be decisions or decisions that are that have bigger impact, such as setting budgets and steering campaigns on daily basis, a lot of marketers do that they set their own budgets. And they have to check that all the time. And I think that's, you know, most companies automate this pretty early because you don't want to have to make those decisions manually all the time. It's just controlling it with a dashboard is enough or low dashboard. By the way, you're I know you're a big proponent of that. But I think that's enough like that's, you know, you can just take the biggest most error prone decisions the most manual parts CMD can be automated, oftentimes decision that involve just you know, if it's zero or one, if it's increasing or decreasing a budget if those decisions are easy to automate, creative, much more difficult than even with several engineers, there's a lot of things that are difficult to automate with creatives, because a video is super hard to automate creative has so many aspects to it, that understanding the performance of a creative based on just the image itself is not enough. So it can become extremely hairy, extremely complex problems very fast. So focus on the simple problems First, the ones that really like nice wins, the easy wins. And that's where you make the most money for your investment. 

Lior Barak  15:40  
If we if we look in on a team itself, so do you need to have a Python developer Do you need to a data scientist in 

Jean de Bressy  15:48  
Their data centers the same, it will depend on how much data you have, right? So you start with an engineer, because most of the things that you're going to do, an engineer can tackle, it can be, you know, it can be start with a data engineer, full stack, engineer yourself for a long time. Because before you need data sense to build advanced models on your steering, there's a few years of time when also you need a big stack of data, that it just makes sense to have a data scientist straight. And as I discovered it, just like for marketing decisions, you're just on a daily basis during your campaigns, then rule based models will get you very far, you don't really need to have a data scientist from the beginning. 

Lior Barak  16:27  
That's cool. Because I think that a lot of our listeners struggling with marketing automation, I know that a lot of people want to do marketing automation, as they find it something that can save them a lot of time. Would you say that when you looking at the path that is alanda have done? Or in general, that companies that went into marketing automation, if they actually could reduce the amount of complexity of their campaigns? Or was it actually increasing the complexity?

Jean de Bressy  16:58  
That's a very interesting question. It's difficult to give it was just like strict answer, more simple or more complicated, because in some things, some things got more complicated. And some things got simpler. So I look back, like the investment decision, for instance, are not extremely simple. And but they rely on complex models. So you use a lot of data to end up to something, the end result is something simple, I said here is the amount of money you're putting on that can be right, it can not be easier. You have a few 100 campaigns, you know how much you're spending those campaigns. But it only looks simple, because when you look at the, the path to get there, and how much data goes there. And then we have like three iterations on data science models and the timelines. That's the thing that you need to take in consideration if you want to automate the timeline to automate something that is just an engineering work or data engineer. And the timeline to the data science projects are completely different, like automating a rule based system for investment is, it can be one of project, data science, doing the same with data science, it's got to be six months to one year, again, you need to understand that you're looking at very low shots. So in to be ready for this. So to answer the question, because I'm veering off, I would say a lot of things got simpler because we started thinking in terms of Okay, what type of decisions are we making? How do we repeat them? And then how can we automate those things, but there's also layers, we're going to become more complicated, because we discovered, okay, if you want to make creative, then it gets. Yeah, as I said, it's just very complex. So you have to break it down in different steps. And then you have to do a lot of testing. And that's where we're at right now, where this part where automating creative creation, and all of the ads is something we're interested in. But it's still far ahead of us. It's a very complex problem.

Lior Barak  18:43  
 Interesting. If we looking now, we had COVID-19 for the past year, how trustful was your data, actually? or How could you stay doing automation? Was it reflecting in your results in any way?

Jean de Bressy  19:00  
So I mean, I don't know if you know a lot about your situation now, but COVID did not affect us negatively wer'e ecomm. And we thought it was going to be very bad for us. Because most companies had to cut marketing, for instance, and change everything that we're doing. Because the situation got so bad for most businesses, but we're extremely lucky. ECAM grew throughout the pandemic. And we are in that position where we actually increased our marketing budget and had more work. Last half year, then we had the beginning of the year, despite pendimic on the data that we have is extremely solid, very reliable, reliable, so it didn't move so much to pipelines that we'll look at. They'll pretty much unaffected because they only look at aggregated behavioral data of our customers. And that's stable. What did change this two things and it's not related to COVID actually, the two things that most affected us is having to work from home because Then you're isolated. So it's a bit more difficult to adjust what you're doing, prioritize and work together in the sense that things get a bit lost. And it's, I think, tiring overload time period. So she was really tired by the end of the year. And then the second thing is GDPR compliance. Actually, this was a very big piece, because last year, we released constant banners on app and web. And so you lose a lot of data. And since the new understanding, so the current understanding of GDPR is that you also lose for your OPT outs, you lose all of the Google analytics data for people who opt out from tracking then had a big impact. There are pipelines because they rely on the entire community, we see that that actually was a very big change.

Lior Barak  20:42  
I see. Interesting, but ecommerce in general is growing. And so we had some interviews in in the podcast that people were actually mentioning that due to COVID, the user behaviors changed. And they saw a lot of impulsive buying, which wasn't there at the beginning, especially in the gaming industry, do you feel it's happening the same at zalando, or this is not the case? 

Jean de Bressy  21:09  
The answer, as a company, I can tell you, this is what we see. I can definitely tell you what people are doing around me. And I think, you know, if you're a guy and you're bored, you tend to have played more video games than before, if you're in your 20s, and make this kind of impulse buying that make you happy. And the women I know have turned to online shopping and impulse buying or online shopping. Really, there's actually a lot of people that were not using zalando around me or lounge that started using it, and they started using it heavy, because they had more time on their hands. as a marketer, what I can tell you is for sure, people started using Facebook, way more in terms of behavior, social media group are lots of gaming room, social media, as well. And what he did is there was lots of traction early in, in half on last year. So right after, depending kids and all of the lockdown started in q2, roughly. There was a big drop, everybody stopped spending on social media. But at the same time, a lot more people went on social media. So the advertise, there was like this very specific moment in time when it was cheaper to advertise. And there was more funerals like less competition, more screen time, some more spaces for advertising. So basically, it all became cheaper. And yes, there were those a huge opportunity. 

And I think nobody could just like predict that it was going to go that way was what really was really interesting to to notice it. And then as I said, like as he comes along, it was so lucky that we're in this position where people transition from offline to online because nothing was open. And what we've benefited from this. 

Lior Barak  22:45  
That's That's cool. It's It's It's actually what we seen many of the interviews that we have, it's it's a returning pattern that the assortment at some point became very cheap. And everybody went in like crazy. And then it became expensive again, because everybody realized how the price went down.

Jean de Bressy  23:03  
Yeah, totally true. Yes, there was a kind of gold rush I had a moment it was.

Lior Barak  23:08  
So before we go, can you tell me a little bit about if you could have changed or keep one KPI, what will be the KPI that you will, for sure, keep for the rest of the life. And you think everybody needs to start using it in marketing or in general, in general, KPI for life, happiness.

But it keeps for marketing. I mean, it's very advanced. But for us, we have correction factors for incrementality. So you know, we look at ROI, I would say the main KPI is ROI. But the main thing that gets into building it is how incremental your campaigns are. So you know, the lift value, if you do Facebook, that's what you should be looking at CPAs all of these very important metrics go into building it. But incrementality is the end goal like that. That's where it's at. If your campaigns are not incremental, you shouldn't be advertising. And from the KPIs that you're using today, Which one will you kill?

Jean de Bressy  24:05  
The KPI is amazing, they will kill 8%

Lior Barak  24:10  
of the KPIs. When you look at, you know, you know this this theory about vanity metrics and companies and people look within a very happy to report on something and say this, this went that direction and that direction, and we have more users and it's, it's never something that you have done yourself like Zalando was an ecomm and then COVID hit and we're so we're very humbled by the fact that Okay, we're still growing, but we're trying to take the opportunity to be a good company and a good player in the ecosystem. But we know it's not because we're the best at what we do. It's because the nature of where we stand that it happened. There's a lot of luck in there. So I think for me, like spend, you know, we look a lot at the spend and then people talk about how much we spend and our spend is entirely linked to how much the ROI is and I don't control it like 3d printing.

As it increases, if it decreases, it's not my decision, it's automatically decided. So instead, I think we'll have a bit of a fake KPI to say, hey, how much was spent, and always look at this, like, this is the golden standard of your activities. Because then year over year, you need to define how you're progressing against your spend of last year. And it doesn't really make sense because it's opportunity based, you know, if like, next year, everybody goes out, nobody shops online, and the numbers decrease on zalando. And we're going to look at year over year performance. And segment spending enough, is just you know, given the environment that owners like, I can't, I can change that. So I think spend some time it's a bit of a fake metric.

I love it. I love it. I completely agree spending sometimes is a is a horrible metric to look at, and has no connection.

It also ends up being a war like people are okay, we'll have that much budgets and so let's spend the money because next year, they're not going to give us the budget if we don't spend it. So then you rush for solutions and say, okay, probably not the main, you know, the best use of, of money to just spend it because it was located debt, 

I completely with you. before we finishing this episode, would you tell us where can we find you, 

Jean de Bressy  26:16  
Where you can find me, you can find me, you can find me on LinkedIn probate. That's the main place that Jean de Bressy and I don't have that I work with Facebook, on Facebook, and I'm doing Facebook marketing, so very difficult to find I get on Twitter. And I use Facebook as well. Because the people as I've seen is the people who work the most in social media are the people that are less active on social media, because they're very careful with those things. 

That's cool. So thank you very much for your time. It's been really pleasure. Any last words you want to tell to the listeners about marketing automation? Before we go?

I think here, everything is said just focus on your data first look at what decisions are the most important to automate and focus on this is the main thing. And yeah, keep it the main thing for one because it doesn't change like even 10 years into into doing marketing automation, still looking back at the same principles. 

Lior Barak  27:12  
Amazing. So thank you very much for joining us. It's been really, really pleasure. And Until the next episode, 

Jean de Bressy  27:19  
thanks for having me by your by. 

Transcribed by https://otter.ai