Our Terms of Use & Cookie Policy

StackAdapt uses cookies to improve your online experience. Cookies are used to help us better understand where our visitors are coming from, recognize you when you sign in to the platform, or better personalize pages you visit. Cookies are placed on your computer automatically but if you chose to opt out from any tracking, you can change your cookie settings through your browser.


AI in Advertising 101: The State of the Industry

StackAdapt’s CTO, Yang Han had the pleasure of presenting at NewCo Toronto this past week on the current state of artificial intelligence (AI) in advertising and his predictions for the near future. 

How Does AI in Advertising Work?

Programmatic native advertising is fundamentally rooted in AI, a complex and often misunderstood process called Real Time Bidding (RTB). So, how does RTB actually work? Han begins:

“Say I’m a user and I go to the automotive section of CNN.com. CNN would request an ad from what’s called an advertising exchange. The ad exchange’s job is to work with thousands of publishers online and monetize for them. And so what they do is, they forward this request to all the buying platforms that plug into this exchange. [StackAdapt] would be one of them, and we would use this data to predict “do we have a client who actually wants to pay for this user?”

He continues:

“Let’s say we’re running an ad campaign on behalf of a car company like Ford. Ford wants to target people in Toronto. And then we see a user is currently on the automotive section or previously they have browsed content about cars, very likely this would be the right user. So then, we would submit a bid for this user with our ad, and the ad exchange would determine who would win this auction and place the ad onto the site.”

ai-in-advertisingAI in Advertising is Generating Smarter Marketers

With the rise of AI, more and more marketers are expecting concrete results from their advertising efforts. Han explains:
“Many marketers, their end goal is to get the user to actually buy something. That action can be attributed to a value. Then you can determine, “Am I actually making money by spending money on advertising?” You want to get a positive ROI. That is essentially the whole purpose of advertising.”

Predicting Conversions

Return on investment is predicated on converting prospects into customers through advertising budgets. Han describes how major companies currently use artificial intelligence to predict conversions and generate value for clients:

The top two companies today that can predict conversion really well are Google and Facebook and that’s because they know a lot of details on you as a user. The companies that do well in conversion outside of these companies are typically what are called retargeting companies.

[This is when] you target a person that has already been to your website. This doesn’t require much machine intelligence at all and the reason why this works is because a lot of these people probably would have signed up or bought your product anyways… and these companies take credit for it. They do add value by reminding users, but it can get annoying.But there are billion dollar companies built on top of just this simple concept such as Criteo and AdRoll.

What Facebook is able to do, they have what is called a “lookalike audience”. So, given the traffic on your website, they can look at these users and they can look at their attributes and look at similar users on Facebook and target them. They can do this because they have massive amounts of user data.


“At StackAdapt, we also have massive amounts of browsing data. It may sound creepy but we have a proprietary way of gaining thousands of action points every second and knowing about the specifics of what people are reading. This is what allows us to also prospect new customers that haven’t been to your site when it comes to these goals.

In the case of Google, they are somewhat behind Facebook but they’re using a lot more AI as well. Historically, Google has used AdWords where you bid on keywords users have searched for. This obviously works well because the person is essentially providing you with their intent data. However, it doesn’t require a lot of AI to do this. Only this year has Google really been able to incorporate user data from all their other properties as well and I’d say they are still playing catch up compared to Facebook when it comes to utilizing this well.

For [StackAdapt], we really started these initiatives over a year ago. We’re starting to see good results in particular verticals, such as travel. For some clients we’ve been able to actually take budgets away from Google and Facebook. And that’s because we are able to access all this data from the beginning.”

Messaging Automation

As Gary Vaynerchuk says, “creative is the variable of success”. In his speech, Han explores the current state of AI in advertising messaging:

“For most marketing platforms you have to upload your message manually, or you have to come up with a manual A/B test. This particular aspect of the industry I’d say is very primitive.”

He gives the following example:

“[Let’s say] I go online and I try to buy a trampoline. If I go to other sites, I’m going to see that exact trampoline or see trampolines from other brands. Which is good, but in order to do this, you don’t need a lot of complicated algorithms. At StackAdapt, we’re starting to solve this problem using some business based algorithms as well, before we move on to the smarter algorithms.

What we’ve been able to do is, we’ve been able to understand what people are actually reading about online. Let’s say a user browses a flight ticket from Toronto to New York, we can automatically pull that from the page and we can promote an ad to the user and say, are you flying from Toronto to New York? Here’s the price, and that’s based on what they’ve done before.”

The Future of AI in Advertising

Much like the state of messaging automation, AI in advertising hasn’t yet hit its stride. Han explains:

“All the unique aspects of what people do, in terms of predicting what they want next, still isn’t there in the industry. These limitations are mainly because there are so many users out there, billions of users, and to run a lot of the more complex machine learning models out there requires a lot of computing power and time, but when we have to show an ad to a user, we only have 100 milliseconds to do it.

So a lot of the more complex stuff has to be trained offline in a very fast period of time for each user which has obviously been a very big challenge. These are the next steps and where we seeing the industry going.”

StackAdapt’s Approach

So, where does that leave digital advertising solutions like StackAdapt? Han states:

“Today as a company, like most companies, we’re still focused on doing very well in today’s world, but hopefully in the next year, we’re going to be working towards more complicated predictions that really add value to the user.

I think the end goal of this is: we want to get the industry in a way where people will be able to find things online that they may have not known about before but they’ll be able to see this and say, “Wow, this is really great.” We all know there’s a lot of amazing content, there are a lot of amazing products out there that have yet to be discovered by people.

Right now, the industry is utilizing this technique where you’re being reminded of stuff you know about already. And it can get annoying for some people. But it doesn’t have to be this way because I do believe that the data is there available for this to happen… You should be able to predict what someone will want next and some interesting things they don’t know about. Hopefully, in the next couple of years, that’s where we’ll see the whole industry move toward, where ads get smarter and start to enrich people’s lives.”

You may also like: