🤖Taking the leap: how to get started with automating your user acquisition – with Colette Nataf(CEO at Lightning AI)🔄

Our guest today is Colette Nataf, CEO & Co-Founder at Lightning AI. Lightning AI helps advertisers find target audiences and advertise to them on Facebook and Google. Lightning AI has worked with more than 500 ad accounts and has managed more than $100M in ad spend.

I’m excited for today’s episode because Colette is among the few folks I know that’s transitioned from a user acquisition service to a user acquisition product. She’s not only taken what to many can be a significant leap – but has also grown her business at Lightning AI very substantially since then. Additionally, of course, she’s always been a source of great advice to me -> and I couldn’t be more thrilled to feature her today.

KEY HIGHLIGHTS

🔑 Some of the key triggers that inspired Colette to start automating user acquisition work.

🔄 How Colette mapped out her personal process as a precursor to automating it.

🕹️ What resources should a marketer find once they understand what to automate.

⏰ Figuring out Facebook’s documentation can sometimes be a very time consuming process.

🏆 The most common early win for most marketers who are looking at automation.

📈 Are products with very high CPAs just bad candidates for automation?

👨‍🏫 What a marketer can do to educate themselves to be ready for automation.

🛠️ Building the mindset of understanding processes is as important as building hard skills around analysis.

⚖️ Why scaling audiences via automation can be challenging – and how Lightning AI approaches this.

👩‍🏫 How learnings from audiences can inform creatives.

📊 How parts of creative iteration and execution can be automated.

KEY QUOTES

How to automate a process

The very first thing that I started doing was thinking about my own process. So for me, it was understanding what are the steps that I’m taking as a human and as a marketer every single day, and which of these are repetitive and which of these need to have some kind of actual human intelligence behind it.

Automation is the logical next step

And if you have a process, you know the process, you’ve done all the hard work at that point. And then you can really hand it over to a back end engineer and they should be able to give you something where you can basically like press a button or it just runs at a certain time during the day and it works.

The importance of LTV in optimization

If I were a marketer, and I had a magic ball of what is the one thing that I would want to fix that gonna make the biggest difference, it really in my mind is LTV predictions and sending back those lifetime values. I think it makes a huge, huge difference in optimization. And also it allows you to work with the Facebook algorithms instead of trying to run against it.

Use early signals to direct automation better

If you can predict potential conversion rates based on some kind of early signal on your website, or in your mobile app, and then send that value back to Facebook, you’re probably going to be a lot better off; you’ll be able to fake it with more events.

Determining lead scores off unconventional data

This is something that we do with B2B SaaS companies in particular – they’ll be able to predict lead scores based on the email address of a person. They look at your company name, and then they figure out what’s the potential lead score based upon the information that we can ascertain from the company. And so bigger companies tend to have higher lead scores, industry, etc.. And so they’re actually sending that value back at the time of email submission, because if we waited for purchase, it would sometimes be six months, so there’s just no way.

Data Science is the future

I think the best thing that you can do to start educating yourself is really understand how you think about data, how you think about numbers and how you think about these processes, because I don’t think it’s super important that marketers figure out how to avoid and circumvent the engineers which is what I did. I think it’s more important that you really understand the processes behind it and the way that these people think. Because the more that you understand how to communicate with engineers and how to communicate with data, the better off you’re going to be.

Creating correlations with data

And people interested in cats are interested in dogs. And so we can put those together into a data set. And then essentially, what’s happening is that we’re rapidly testing those audiences. And we’re trying to understand, are these things actually a match? Or was it kind of not? Was it a swing and a miss?

The real benefit of data correlations

It gets really exciting when people can take away something about their audiences and really know and learn from that. And we work with a skincare company and their best audiences are people interested in travel. But, you know, they were able to take that, they make different creative for people interested in travel. And they put that all together and build out a better brand from that. And now, not only, are you extending your audiences, finding new users, maintaining low CPAs, but you’re actually able to affect the final outcomes from that. So, I think it’s really exciting when people and computers can work together.

FULL TRANSCRIPT BELOW:

Shamanth: I’m very excited to welcome Colette Nataf to the Mobile User Acquisition Show. Colette, excited to have you on the show.

Colette: Excited to be here. Thanks for inviting me.

Shamanth: Absolutely. I’m excited to have you because you are like just a super happy person. Always a pleasure to talk to you. But more importantly you have, you’ve done something that I have a great admiration for, which is you’ve transitioned from being an agency to running a product. And in the process, you’ve automated a lot of user acquisition activity. And one of the things I hear about from a lot of marketers that I speak to is about automation and how they can automate the user acquisition better.

I’m definitely looking forward to getting your insights and pointers into how a marketer might think about automation.

To get started, when you transitioned from being service focused to automation focused, what were some of the key opportunities you noticed that inspired you to say: aha, maybe we should start to productize some of these things, maybe we should start to automate. What were the first opportunities you noticed as a marketer?

Colette: So when we started working on what eventually became Lightning AI, the things that we were automating were 100% different. So I think what’s been interesting to see both as a marketer, and as a business owner is that things really do change very quickly. When we started thinking about automation, it was because we were just doing the same things over and over again. And it was driving everybody crazy.

So the biggest problem that we were facing on Facebook was that we were working with companies who didn’t have enough data to optimize towards purchase events. They were optimizing towards, like ‘Add To Cart’ or whatever they had that was right beforehand. And they needed to have bid adjustments placed based on the conversion rate from add to cart to purchase for each of their individual audiences. And we were going in manually and doing this every day. And it was like, first of all, it was highly inaccurate, because we’re just doing this manually and people make mistakes all the time.

And it was honestly, it was just like a huge pain. It’s just not fun to go in and use Facebook manager and try to import and export and get random errors or like, who knows what was happening? Everything looked like it would go wrong all the time.

Shamanth: Yeah. Interesting. So this was very much scratching your own itch. You were like – it was time consuming, let’s see if there is a better solution. And what were the first steps you took to automate some of the stuff?

Colette: Yeah, absolutely. So I mean, really

Because there are some things that I think as marketers, we just use our gut and kind of feel like we know the answer, there is still a little bit of art behind it. But I think there’s also a lot of science. And so it’s really those things that feel very scientific, where you can kind of start by saying, like, what are the steps I’m taking: that becomes an algorithm. And that’s what we can teach computers how to replicate.

Shamanth: Right. So what you’re saying is, if there’s a marketer that is wondering where to begin, it might be a good idea for them to map their process out – and see what is it they’re doing multiple times. They are changing bids, 20 times a day, maybe there’s like, if you’re hitting spend caps, reduce bids by 10%. That’s an easy rule that can change, right, that can be automated.

Colette: Yeah, I actually have a friend here where he would wake up at like 4 in the morning to change bids and change budgets because he found out that there were better conversion rates in the morning for this product. So he would wake up at 4 in the morning, change bids, go back to sleep and then when he woke up, like for the day, he would set them back…. and I heard this story, I was just like, “you’re doing that every day, on weekends too?!”

Shamanth: Yeah.

Colette: So I think there’s a lot of things that, like marketers know the right choices and we look at all of the data. We know all of the data. So we know the right things to do. Yeah, sometimes you just need a little bit of help from computers. So you’re not the one actually executing it.

Shamanth: Indeed. So let’s just say, our hypothetical marketer has mapped out their process. They’re like look, I’m changing bids 20 times a day, changing budgets 20 times a day. I need to automate this. What are some of the resources they might look for? Let’s assume they’re working for a big company or for themselves, what would they do once they identify this is what they want to automate?

Colette: So you’re gonna need someone called a back-end engineer. I think a lot of people assume that you’re going to need some kind of like data scientist, AI something, and for probably like, 80 to 90% of the things that marketers do on a regular basis, you really don’t: you just need to be able to map out that process and hand that process over to an engineer.

Back when I wasn’t working in my own company, and I couldn’t just command people to do things for me, finding an engineer to do this work was really hard. And it’s hard I think for a lot of reasons. One is that marketers don’t usually have engineers on staff. And so you really have to get buy in from a product manager. And that’s always kind of surprised me about how hard it is because marketers typically spend so much money.

Shamanth: Right.

Colette: Even if you get a marginal improvement, like 5% improvement, it’s a huge amount of money for the company. But I very much know that that can be a challenge. One of the things that I do now all the time is just hire consultants and I find these people just like on Upwork, or from word of mouth from friends who have had good experiences – and honestly, I think that it’s a lot easier to build something simple than people expect it to be.

Shamanth: So certainly, it’s understandable that marketers need engineering resources to make a lot of this happen – and it can certainly be challenging to do that, even inside a bigger company, as I’ve had myself experienced as well. Let’s assume a marketer has a roadmap of things they want to execute – the lowest hanging fruit could just be bid changes. If you have a different low hanging fruit in mind, please express that. But let’s assume for the lowest hanging fruit, what sort of quantity of time and effort do you think it could reasonably expect for an engineer to take?

Colette: That’s a great question. I mean, I think that things like bid changes, really low effort. The hard part is usually getting the access required and setting up the connection between your engineer’s workflow and Facebook or whatever the system is that you’re looking to automate. So that’s usually complicated. I will say the Facebook documentation is awful. Apparently, there’s only one person who updates documentation for all of the engineers.

I don’t know if that’s still true. But when I heard that, I was like that, that is just the craziest thing I’ve ever heard in my life, like it’s insane. So they don’t always make it easy. And so I think there’s a pretty steep learning curve that exists there, I would assume the first project would take, you know, probably a few weeks, and then after that, it would be a lot faster. And especially if you’re already kind of mapping out what to do. I think one of the other things that can be challenging is like when you first start working with engineers, you really have to explain what the problem is: because oftentimes, they’ll have a different way of tackling that problem. And you know, no one’s right or wrong. But I think engineers generally like having the flexibility of having guidelines and then being able to do whatever they want in the middle, where they think marketers expect it to be like, “I told you to do this, you must do this.”

Shamanth: I get it. Yeah, totally, right. And so from what you’re saying, It sounds like one of the bigger challenges is that just the Facebook documentation just needs to be untangled. And then you need to make sure you’re on the same page with your engineer who really doesn’t have the background that a marketer would have. And out of curiosity, I’ve been taking the hypothetical example of bid changes. But what would you suggest for a Facebook marketer who’s fully reliant on ads manager these days? Is that a good place to begin? Or what else would you suggest could be a good starting point for them?

Colette: Yeah, That’s a great question. And usually, my answer to this question has only a tangential relation to Facebook, which is interesting. Probably the number one issue that I see is that people are optimizing towards something that’s wrong on Facebook. And that could be like they’re optimizing for ‘Add To Cart’ instead of ‘Purchase’ or they’re optimizing towards an email sign up instead of having a high quality user. But usually one of the biggest issues that I find, especially with a lot of subscription businesses or mobile gaming is that they’re not optimizing towards lifetime value.

So the Facebook users tend to have very high churn, or just much lower conversion rates than users who are coming in organically. And really, the reason that that’s happening is that we’re sending back through the mobile SDK: these events that are happening, and we’re not sending back predictions of future values. So I mean,

I think it can be really challenging to figure out like, what’s the next hack that I’m going to find in Facebook is, and then keep up with that. But if you’re sending back data to Facebook, that’s forever, and it’s something that’s gonna make a big difference right away.

So I think it’s a good win for the marketer, I think it’s easy for engineers, because there’s nothing kind of new for them to learn. And I know that engineers complain constantly about installing SDKs, but I promise you that once they’re installed, it really isn’t that hard. And I feel like here’s where they come at me with pitchforks for saying that, but like, it really isn’t. It’s honestly the same as installing a pixel on a website. And I know a lot of marketers know how to do that because we’ve had to learn where that goes and how to put it in. It’s just like copying a few lines.

Shamanth: Interesting. Marketers like, we need to optimize for LTVs. In terms of just automating some of that stuff, or just fixing that in their process or workflow, where might they begin?

Colette: Yes. First, you’re gonna need some kind of LTV prediction. So that’s probably the most complex part of all of that is actually being able to understand like, what is the future LTV? And what are the leading indicators of it because obviously, the sooner that you get those indicators, the better it’s going to be.

And usually, when I work with mobile gaming especially, they’re not able to predict LTV as soon as somebody makes their first purchase. So typically, what I recommend is to just count that first purchase as purchase no problem. And then as soon as you can estimate the LTV, just send that back through. So whatever event allows you to figure out the LTV, just send that back through also as a purchase, also with the value – and Facebook will figure it out. I mean value optimization works. You should be using these features that are already integrated into Facebook and work with the system instead of against it.

Shamanth: Certainly. Just to switch gears a bit. One of the things that comes up when I speak to marketers about potentially automating a lot of this stuff is they wonder if there are certain kinds of apps or products that are not good candidates for automation. And one frequent case I hear about is if the CPA is extremely high, $100-$200 and they’re like, look, I just hardly get any signals. In cases like that, would you say that’s just a bad candidate for automation? Or are there other ways that they could be thinking about this?

Colette: Yeah, I mean, I think it’s a great question. And we’ve really started off by working with those types of customers. I think there’s a few things that you can do. One is, again, try and find those early indicators and send them through. So

And that, you know, allows them to have CPAs that are much lower and have events that they can use for optimization. So, I think that’s usually the recommendation that I give is like, what can you do that’s higher up the funnel, where you can make predictions about the future, then send that back as a value to Facebook and then use value optimization.

But the core of what you’re saying is true, if you don’t have enough data your hands are tied. If there are things that you’re doing every day, you can automate those things. But if you have a low budget, if you have high CPAs, probably what you’re doing is just running broad audiences on Facebook and it’s probably working well enough.

Shamanth: Certainly – that does make sense. You could certainly optimize towards upstream events. That’s really all you could do potentially. And you did speak about how it’s important to have engineering resources to build out a lot of automation. What might be some of the things that a marketer could do in terms of educating themselves and becoming more ready to handle more automation?

Colette: Absolutely, I would say take a Data Analytics class. So, whether that’s called Data Science, or Stat 101, whatever you know, I think

Shamanth: Interesting. Is there an example that comes to mind about, what is something a marketer that is somewhat plugged into how data works would do, as compared to somebody who wouldn’t be as plugged in?

Colette: Yeah, absolutely. The biggest thing that I think, if you’re a person whose brain just kind of works in a way where you say, like, I’m going to make these sets of tasks and every time I do like, whatever budget changes, like these are the things that I look for. My very first job in marketing was at Expedia. I was always like how do I find rules that I can make decisions without clouding my own judgment with my own thought processes.

If the cost per conversion is within this range, and if we’re spending this much, and if we’re getting this many clicks, then increase budget, or, you know, conversely, like, decrease budget, and I think that those types of thoughts and that type of mentality is super, super important when you’re thinking about what can you automate – and how do you work with engineers in order to create that automation.

Shamanth: Interesting. It’s almost about cultivating that mindset, as much as getting any hard skills around data analysis.

Colette: Yeah, absolutely. I think it’s a lot about how you frame what you’re doing, how you think about what you’re doing. It can be very challenging to learn how to code. I did that, and it was not a good choice. I tell people all the time, like, you come up with the algorithm, you come up with the way that something should be done, and then hand it off to somebody who – like that’s their full time job.

Like they’re just gonna be better at it than you are and some people have a passion for coding and for those people, you know, like, good for you, you do that, like, go join Wish and be like a marketer, or just an engineer at this point and learn from them. But I think for most of us, coding is probably not going to be our number one skill. Probably, marketing is our number one skill. You can really stay in that, learn how to work with data, learn how to work with these algorithms, and then bring an engineer in, I think you can be a lot more effective.

Shamanth: Indeed, that completely makes sense. To switch gears a bit, again we’ve talked about a couple of things that marketers could automate or work with engineers to make more efficient. And one area that it would appear can be somewhat challenging to automate is audience prospecting. And I know your team at Lightning AI focuses quite a bit on automated audience expansion. So I imagine this is something you have perspective on. And this can be challenging just because Facebook constructs audiences under the hood. You don’t really know if this user – and this is an example I’m making up – let’s say this is a user who likes Game of Thrones and also likes something else. So that’s something you just don’t see how Facebook’s making that connection.

So let’s assume there’s a marketer who’s tackled a lot of the lower hanging automation, be it bid changes, they’ve taken care of a lot of these things. They are like, right, I want to see how I can use automation to scale, to unlock new audiences, to really prospect new audiences. How might they start thinking about this? What might be some of the first steps they might take in this regard?

Colette: Yeah, absolutely. I mean, I tell people all the time, like, our algorithm isn’t a secret. What’s hard about it is the execution and building up the data set that we have. So I mean, the way that we create audiences is that we look for those correlations. So usually people know something about their audience.

So they’ll know my audience likes Game of Thrones. And we search different social media sites for people who are talking Game of Thrones and then we’ll search their public profiles to see what else they’re talking about. So that allows us to make correlations and analyze those correlations to predict that somebody interested in Game of Thrones is also interested in whatever – cats.

And people interested in cats are interested in dogs. And so we can put those together into a data set. And then essentially, what’s happening is that we’re rapidly testing those audiences. And we’re trying to understand, are these things actually a match? Or was it kind of not? Was it a swing and a miss?

And that’s really our way of working with Facebook because like you’re saying, a lot of these things are under the hood. Like we don’t know for a fact that somebody who appears interested in Game of Thrones is even really interested in Game of Thrones, like they might have just gone to the website or watched a video about it, and then, BAM, now they’re seeing all of these ads for people who like Game of Thrones.

I think there’s a lot of that goes on Facebook. I mean, I’m a new mom and I constantly get ads for things like kids who are 10 years old. You’re just like, well, you kind of got it Facebook, but also you’ve completely missed. It’s just something that happens. But in terms of automation, now we have a pretty extensive data set on what those correlations really are between audiences. And we can build those out very quickly. And we think of those as tests. So every audience that we create is a new test within the system. And if it works, well, great, it stays running, and if it doesn’t, that gets turned off really quickly so that we’re not spending additional budget there.

Shamanth:Sure. So if a marketer is looking to build in more automation in their audience expansion, interest expansion, they might want to consider building as much learning internally as possible. If I’m a mobile game, if my users like Game of Thrones and cats, they are my best users, right, so the stronger the correlation they cultivate, then maybe people who like dogs are negatively correlated.

Colette: Yeah.

Shamanth: So I think so what you’re saying is: the larger set of interests is, and the more correlations they can build up within that, the better the predictions will be.

Colette: Yeah, absolutely. I mean, I also say like my favorite examples always are when people use the information not just to lower their CPAs or to expand out their marketing and maintain the same CPA, like numerical goals but actually when they take that information and apply it back to their product or back to their creative. That’s kind of when everything really starts to sing well together.

And I think that’s when it gets really exciting when people can take away something about their audiences and really know and learn from that. And we work with a skincare company and their best audiences are people interested in travel. But, you know, they were able to take that, they make different creative for people interested in travel. And they put that all together and build out a better brand from that. And now, not only, are you extending your audiences, finding new users, maintaining low CPAs, but you’re actually able to affect the final outcomes from that. So, I think it’s really exciting when people and computers can work together.

Shamanth: Yeah, that’s interesting. So you almost look at the audience and you make the creative based on that. Right?

Colette: Yeah, exactly.

Shamanth: Interesting. I can see why that would absolutely work. And speaking of creatives, some would think creatives are an area that can’t be automated. And obviously, that’s like a blanket statement. But yeah, are there elements of creative ideation or iteration that lend themselves to automation that marketers should be thinking about?

Colette: I think that one of the things that’s really interesting when people talk about creative automation it’s like, the idea that they don’t have a set template that they use all the time for all of their ads, because probably they do.

I know when we hired an entry level designer, and we basically gave her a template and sketch and we were like, okay, you know, find images, put the image into this template. Here’s like the five copy iterations that you can use. Here’s like, the three call to actions, you know, make a bajillion ads, and you know, honestly, like that can be done by a computer.

I mean, finding the images, definitely a person’s job, computers not going to excel at that. But you know, once you have those making the iterations off of them absolutely can be done by a computer. And so then uploading them and testing it. I think uploading ads to Facebook Ads Manager is probably like the bane of my existence right now. It’s so annoying, so slow. I hate that Facebook doesn’t default to “create a post,” Facebook know that this is the right thing to do. Just do it for me, you know?

Anyway, so luckily, I have my own company and so I can make my engineers build out these things. And so now I don’t have to upload the ad through Facebook Ads Manager. And it’s definitely something that I think can really make an impact for people. It’s just having a way to upload these ads and also a way to really scale them out. So you’re not just making like one or two at a time, but you make really a whole bunch and not drive someone crazy.

Shamanth: Indeed and especially iteration based on common templates, certainly led to automation fairly easily. Colette, this has been incredibly insightful as it is every time I do speak to you. Yeah, thank you so much for being in the Mobile User Acquisition Show. As we wrap, could you tell our listeners how they could find out more about you and everything you do?

Colette: Yeah, absolutely. So you have kind of two options. One is you can go to LinkedIn, search for me – Colette Nataf and if you message me, I will probably respond unless it’s a sales email in which case I will probably ignore you. But I’m there if you want to have a direct chat or you can go to my website lightningai.com.

And you can sign up if you want to see what the experience of LightningAI is like with automation for audience testing. Or you can chat with anyone for my team in our little chat window. And, you know, we’re here to talk about automation, we are here to talk about Facebook, here to talk about anything you want.

Shamanth: Perfect. You also do write fairly prolifically on Medium, which certainly I’m a fan of. So please check out Colette’s writings on Medium – certainly, something I find very inspiring.

Colette: Thank you so much.

Shamanth: Alright Colette, thank you again for being on the show.

Colette: Yeah, thanks for inviting me. This has been fun.

A REQUEST BEFORE YOU GO

I have a very important favor to ask, which as those of you who know me know I don’t do often. If you get any pleasure or inspiration from this episode, could you PLEASE leave a review on your favorite podcasting platform – be it iTunes, Overcast, Spotify or wherever you get your podcast fix. This podcast is very much a labor of love – and each episode takes many many hours to put together. When you write a review, it will not only be a great deal of encouragement to us, but it will also support getting the word out about the Mobile User Acquisition Show.

Constructive criticism and suggestions for improvement are welcome, whether on podcasting platforms – or by email to shamanth at rocketshiphq.com. We read all reviews & I want to make this podcast better.

Thank you – and I look forward to seeing you with the next episode!