In this video, Autron Co-founder & CTO Julian Timings discusses machine learning in Amazon advertising, the importance of retargeting, balancing inventory and ads in Q4, and much more.

Transcript below.

Julian’s background

Rachel Go: First off, thank you so much, Julian, for coming on to chat with me for the MyFBAPrep interview series. To kick off, can you share a little bit about your background and how it led to where you are today?

Julian: Cool. Yeah. Excited to be here. Yeah, so, I come from a pretty technical background, actually. I ended up—so I’m originally from the UK (hence the accent). I did a PhD from Cambridge University in engineering and particularly, like, machine learning and AI. So, I have this kind of strong technical background. And after that, I actually went into work for—into Formula One for a couple of years. So, as a child, I loved motor racing, and that was always my dream, and my PhD was somewhat related to that. And did that for a few years, super exciting, learned a lot on the technical side, but kind of realized that what really excited me and engaged me was the idea of building a business and building technology to kind of power that business, I would say.

So, you know, after F1, I moved into my first company, which I built, which was Cloud Kitchen, and that was a, yeah, great experience. We kind of built the company from nothing to something and we built all the tech for that. Just so people are aware of what Cloud Kitchen is, it’s a business that not only sells food but also produces all the food, delivers the food, like, the full-stack food delivery company, when it was kind of popular back in 2014, I would say, there was a lot in the U.S. too.

So, we built that and built it to a decent size and actually got acquired by a Y Combinator-backed start-up. It was also in the same space. So, that was, you know, a good learning and experience and had a successful exit from that business. And then after that, this is when I first kind of, like, started to have some touch points with Amazon. I kind of realized that there was some great opportunities just on the eComm side, to be honest. Like, as an Amazon seller, this is a huge opportunity. It is great platform.

So, I started a company to sell products—physical products on Amazon, built a brand and a website and, you know, the full thing that’s required to build a business on Amazon. And quite quickly, I realized that, you know, advertising is such a crucial part of being successful on Amazon, right? I think you see how much of the top-of-fold content is effectively ads, right? And realize that, well, this is actually super, super important. So, like many sellers, I started doing the ads myself manually and kind of learning the basics.

But, given my background, I realized that all of this actually looks really well suited to kind of machine learning basically. And [with] some of the theoretical understanding that I had, I thought we could, you know, apply that to the concept of bringing results and cost-effective results on Amazon advertising. So, I built something basically from my own business to start with, right? As many entrepreneurs do, they solve their own problems. And then I shared this with a friend who was also selling on Amazon, and it worked really well on his account, too, and, you know, one thing led to another.

So, this was Sean, who is actually my co-founder now at Autron. You know, we decided to commercialize the product. You know, we spoke to sellers, realized that there was a pain point here still around this gap between people doing it themselves, right? And then—or just a hundred percent outsourcing it to an agency, you know, which is still a fundamentally, a human running the ads. We thought this was, you know—we thought we could do better with—on all the bid optimization and all the automation and so forth with the tool, and this is what it started out as, and we’ve kind of developed that idea ever since, and today, you know, working with hundreds of sellers across the globe on all of Amazon’s marketplaces to help them with their ads and still with this idea that AI and automation is a great fit for this model, I would say.

How Autron leverages machine learning to maximize Amazon advertising

Rachel Go: Can you tell me more about Autron and what it does?

Julian: Yeah, cool. So, Autron is an advertising tool for Amazon sellers and brands, and it’s backed by some pretty sophisticated machine learning that helps to optimize your bids fundamentally on your campaigns and also will automate the entire campaign creation process and management of campaigns. It optimizes the bids on an hourly basis, leverages all of the data that Amazon makes available to us to be able to maximize your returns fundamentally. And you can visit us at autron.io—autron.ai, sorry, or .io. We have two domains. And, yeah, [I’m] excited for people to try it out.

Rachel Go: You mentioned machine learning, like you noticed that it was a good fit for machine learning. What aspects in particular did you pick up on that you thought this would be a good idea to apply this to?

Julian: Well, I think initially, there’s a vast amount of data obviously that’s associated with Amazon advertising, right? They do provide pretty decent tooling in terms of being able to get access to that data, understand all the different factors that might influence sales, you know, whether that’s ad placement, whether that’s seasonality effects, whether that’s audiences. So, by having all those different kind of levers, I guess, this kind of made you think, “Okay, well, machine learning is good at understanding which lever to pull to give a particular action.”

And also, on Amazon, the way it currently is dominated, I would say, is still kind of by search-based intent. So, similar to Google Ads, right, it’s more about people with intent searching for products. So, that is less about the creative, I would say, right? So, Meta advertising is very much kind of creative, and TikTok is very much creative-based.

So, machine learning back—you know, a few years ago, before we had kind of, you know, image generation and so forth, we’re still, you know, more suited to numbers and words and that type of building models around that type of data. So, this is why I think it was particularly suited at the time for machine learning and still is, but Amazon is pushing more in the creative direction and—but, of course, we have, you know, more modern tools to help us with those problems and challenges, too.

Lean into automation and strive for high data accuracy

Rachel Go: I noticed you had a really diverse background going from, you know, F1 to Cloud Kitchens, and very different industries. So, what lessons have you picked up from your past career that has helped you with your Amazon career?

Julian: Yeah, I think obviously, like, the Cloud Kitchen was kind of my first business in a way. So, that was all the lessons that you get running your first business around hiring and how to build out a product and brand and customer acquisition. But, you know, learnings, I would say that are highly relevant to Amazon sellers and sellers in general is basically leveraging two things.

One is kind of automation, I guess. You know, typically in the beginning, there’s only a few of you, and you could be a solopreneur and you haven’t got unlimited amount of time and how do you kind of, like, leverage tools and automation to be able to have more impact, right? So, I think, you know, leaning into automation is something which has been a valuable lesson for me.

And the second thing is around, I guess, data-driven decision-making. So, again, like, you have a lot of signals coming at you as a business and you’ve got to decide which—where to focus, what to prioritize, and you generally have to make decisions on incomplete information, right? Maybe 70 percent is the rule of thumb; you make a decision on 70 percent of the information. So, you want that information to be as high quality as possible, right? So, having data kind of assist those decisions, whether that be around, again, you know, customer acquisition, where to put your budget in terms of growing your business, whether that be around, you know, the unit economics of particular products and how to, you know, focus on the products that have the best margin, or things of that nature, where do your customers come from, you know? I think having these inputs and using these data to help you with your decisions is a really important lesson, I would say.

Rachel Go: I know a lot of merchants struggle with verifying how accurate their data is. So, with that in mind, can you share some tips that you’ve picked up on how to confirm that your data is at least 70 percent accurate?

Julian: Yeah, I think there’s a few things. I mean, on the Amazon ecosystem explicitly, you know, they do a pretty good job of attribution, right? Just on Amazon ads, right? So, you know, if you’ve got a particular campaign setup, you should be able to, with pretty high confidence, understand, like, what are the keywords that, you know, drove particular sales or were responsible for a certain amount of spend, you know, which ones of the ad groups, you know, are the important ones.

So, I would say, on that side, it’s fairly well defined. On—you know, something which has become more readily available over the last year or two is around Amazon Attribution. So, this would be, “Okay, how do I understand—if I drive traffic from external sources to Amazon, how do I measure that and the impact of that?” And until fairly recently, there wasn’t particularly great methods for that. But Amazon came out with something called Amazon Attribution, which effectively allows you to, you know, track the clicks and conversions that come from external sources and result in purchases on Amazon. So, I think, you know, having a both on-Amazon and off-Amazon tracking solution is pretty—is the fundamentals, I would say, right?

And then I think what you can do on top of that is just try to get those data sources into a unified place. And there are a bunch of kind of third-party tools out there to help you with this so that you can, you know, put it into some kind of BI tool where you can look at the data and try to understand through graphing, I guess? And plots and charts tend to be a bit more easy for us to digest, you know, “What are the things that are important?” And “Where should I invest more time and money?” I would say.

Autron’s emphasis on AI and ad sales

Rachel Go: Amazing. Going back to Autron, I did take a look at your website and I noticed there is, in addition to machine learning, some AI aspects. So, where does AI play a role in creating and optimizing campaigns in Autron? And why do you choose those areas for AI enablement?

Julian: Yeah, great question. You know, I think—so one of the—we use it in a bunch of areas within Autron, right? I would, you know—the places where it’s particularly well suited right now are around natural language, right? And where does this come up in Amazon, right? It’s all of the kind of listing copy, right? So, you have a title and you have a description and you have all of that ACoS content and you have all the back-end keywords.

So, this is kind of one of the ways that we can inform Autron around the linguistics related to our product is using AI, right? Because it can read all that and it can pick out, you know, the important keywords and the important phrases and give Autron a view, not just from the data—we have all the data, of course, that we get from the APIs when Autron connects to your Central and Ad account. But then also what we can do is inform Autron, like, of how—and also the images too, like, what are the images that describe this product? What are the keywords? And all that type of thing. That’s one area where, you know, the recent advancements have been able to supplement Autron’s understanding.

Another one is around, you know, keyword generation, of course. You know, I think, like, keywords and opportunities of where to—what to bid on, you know, there’s a bunch of data sources we have for that. And, again, turning those into the right type of keywords, I would say, for Amazon is another important area. And then something which we’re exploring too, which, you know, this is not out yet, but we’re building this out for customers is a way for them to be able to, like, interact with Autron and their data in a more natural way, right? I think a lot of people don’t necessarily, you know—it can be overwhelming with all the graphs and plots and how to understand that data and what are the, you know—it’s all very well presenting data to people but, like, what they really want is, like, actionable insights, right? Or recommendations or things they can take action with.

And, so we’re looking at ways to pull all that data together and then communicate that back to our customers and clients in a concise and actionable way. And that’s something that’s really, I think, really the direction that the client-agency relationship might shift in the future, right? And I think at the moment, sometimes the tools are not, you know—you quickly run out, you quickly get to the end of what they can—how they can help you, right? Because they don’t—they’re not really powered by your own data. But I think if you can kind of power them by your own data or the seller’s data, then, you know, it could be much more useful, and that’s something we’re actively working on at the moment.

Rachel Go: I saw that Autron also charges based on ad sales rather than ad spend. Why did you decide to go with this unique monetization approach?

Julian: Yeah, I think it seemed rather obvious to us, to be honest. We really want to, like, align our incentives with that of brands and advertisers, right? And I think, you know, if we’re aligned on that, you know, you got sellers and brands [who] want to grow their sales, and we want you to grow your sales too. And I think that just made a lot of sense, right? And I think, moving forward, I expect just more people to kind of adopt that approach, I would say, like, it just seems to make a lot more sense and means that the relationship, I think, that we foster with our clients is more transparent and more aligned with, you know, what they want to achieve and hopefully, you know, longer lasting, I would say.

How to optimize audiences (and avoid degrading them)

Rachel Go: Can you tell me about a couple of different ways to optimize audiences that a lot of sellers overlook?

Julian: Yeah. So, I think some of the fundamentals around building an audience and optimizing that comes around to a few things: One is, like, really understanding the demographics and the interests of that audience, right? So that you can kind of better tailor ads to them.

In terms of being able to, I guess, “squeeze” (for [lack of] a better word) more out of that audience, you really want to participate in retargeting efforts, right? So, this would be, you know, don’t just try to communicate with the audience once, right? Or, if they’ve showed some initial interest in your product, they could have, you know, viewed a page or, even better, clicked on it, or, you know—or another, or they could have even converted, right? So, you can retarget these in various ways, depending on the touch point that they’ve had, and I think that’s super important.

And it kind of never ends in a sense like, okay, you have to draw—you have to balance that line between being too pushy, I guess, versus making sure that you’re reengaging with the customer at the appropriate point in the future so that they’re interested again. So, that would be about understanding their buying behavior and and what is the frequency of that buying behavior and making sure that you present yourself at the right time based on that, so that there’s higher conversion rates.

Another way that’s super important is really just to kind of build out personas, I would say, around your audience. You’ve really got to understand, like, who they are and build out theoretical profiles of, you know, John and what his purchasing behavior is and what drives him to make his particular decisions and things related to that, I would say, is another great way to optimize the audience.

And then, yeah, finally, don’t just focus on, like, purchasing behavior. Also, think about, like, wider audiences, I would say, that, you know, that have overlapping segments of customers and where you might be able to find customers who aren’t necessarily directly looking for your product or service, but can—but when they see it, you know, it grabs their attention. So, I think that’s another useful tip, I would say.

Rachel Go: What would your response be to sellers’ concerns when it comes to ads about degrading audiences? Because the more efficiently you reach your target audience, the less ideal those audiences are going to be moving down because you’ll have reached everyone who’s your perfect fit and then have to essentially dig deeper and make sacrifices in quality.

Julian: Yeah, I mean, I think that’s relevant if you have an audience that’s finite in size, I guess. But I think on—particularly on Amazon, there’s, you know—you’ve got this constant flow of new customers, and I think the sizes of the audience for most sellers is reasonably substantial, I would say.

So, yeah, I would—I can’t think of an example right now of a product that’s got a very constrained audience size that doesn’t grow over time, right? And I think that, for most sellers, that that audience is always pretty dynamic, I would say. And I think there’s—you have to be conscious of how much you communicate with them or how much you engage with them or how much you target them. But, in terms of the flow, I think that’s the beauty of Amazon as a platform, right? It does provide the people on the platform for you to sell your products to. It’s a little bit different off Amazon, where there isn’t such a robust platform and you have to drive those sales channels yourself. But for Amazon, yeah, not—I would—yeah, I wouldn’t be too concerned about that, I would say, in general.

A focus on Amazon Attribution and brand analytics

Rachel Go: Great. Well, as a seller yourself and someone who knows the Amazon ecosystem intimately, what are some tips you have for sellers looking to build full-funnel attribution with a good overview of both organic and paid efforts?

Julian: Yeah. I mean, I think just touching back on the two areas I mentioned earlier around, you know, the paid efforts on Amazon — you know, Amazon ads is obviously the place to do it, and there’s good attribution there, reliable and pretty detailed. Off Amazon, driving traffic to Amazon, there’s Amazon Attribution, which is a, you know, free offering from Amazon, which allows you to generate links, tracking links effectively, which you can place off—on your off-Amazon eComm sites and social media or wherever you want, wherever you drive traffic from. So, they’re the fundamentals of measuring the Amazon paid efforts.

And then on the organic side, you know, there is some pretty decent tooling within Amazon—within Seller Central around brand analytics and search query performance and products, which people view to then—before purchasing your product. So, I would definitely recommend sellers kind of explore some of the, you know, the free tools that Amazon provides via Seller Central, particularly around brand analytics. And they seem to be adding new features there on a, you know, quarterly or so basis, so always keep an eye on those things.

And then completely off Amazon, on the organic side, right? The fundamentals around Google Analytics and those kinds of tools that allow you to understand, you know, the search traffic that is coming, [and] ends up on your Shopify store, wherever you sell your products off Amazon too. That’s pretty crucial, I would say.

Mind your numbers going into Q4

Rachel Go: Thank you. And my last question is, what is your advice for merchants coming into Q4?

Julian: Yeah, I mean, Q4 is huge for many, many, many merchants. And I think what we see, at least at Autron, is — it’s quite a simple advice in some ways, but, like, make sure you’re in stock. Like, this has such a big impact on your success on Amazon and just the continuity of your sales, your ads, and just making sure everything I would say stays primed on the Amazon side. So, you know, really make sure you’ve got your product in stock for, you know, the entirety of the period.

Beyond that, you know, in terms of, like, leveraging some of the data, I would suggest kind of looking at the trends from last year in a similar period, just to understand, like, when the peaks and troughs are. Obviously, you know, you’ve got Black Friday and then the lead-up to the holiday season and so forth. So, just kind of look explicitly at this—at your sales volume on a week-by-week basis historically, and try to understand what the trends are.

Like, you know, something that people often overlook is kind of—you know, they think that, you know, sales and conversion rate are going to be spectacular all the way up to Christmas. Yeah, what we see is that they are spectacular up until around the 21st, and then — or the 22nd — and then they literally drop off a cliff. So, you just have to be mindful of those kind of, like, you know—the last couple of days, you know, conversion rates come down and you need to be, you know—if you’ve been aggressive with your ads, for example, you need to wind that down beforehand.

So, that’s some of the two things. And then, you know, I would just touch on making sure you’ve got your promotions, like, planned well in advance, right? I think the timelines there from Amazon are always pretty extended. So, you know, the deadlines are fairly early in Q4, I would say. So, just make sure you’re aware of those and plan them. Plan them and your participation in them accordingly, I would say.