In this series of articles, Crakmedia’s teams will discuss their own experience with the AI-based tools they use every day. These tools allow the Quebec City performance marketing company to quickly adapt to the changing conditions of this highly dynamic type of marketing, requiring the ability to adapt in near real-time.
The year 2024 has come to en end and Crakmedia has made giant strides in the last year in the use of the AI tools at its disposal. In October, we discussed how our designers’ work was accelerated and facilitated by the multitude of tools at their disposal. A platform like Stable Diffusion allows the team to train artificial intelligence with the creations of Crakmedia designers in order to generate material that complies with the style guides of our brands under management.
A few months ago, Crakmedia’s automated ad placement optimizer project won the Mercure for Productivity Increase at the 2024 Mercuriades Gala, organized by the Fédération des Chambres de Commerce du Québec. It is now well established in the daily life of the company’s media buyers and several other tools are being developed or planned to support their work. AI tools have the potential to revolutionize the daily lives of corporate media buyers, and our communications specialist François Tremblay spoke to Francis Jobin, Director of Media Buying Operations, and Gabriel Savoie, Media Buyer, to get their take on the subject.
What is media buying?
Media buying is an essential strategy in both digital and traditional marketing. It is the process of finding, negotiating, and buying advertising space on different mediums to reach a targeted audience. Media can include television, radio, newspapers, magazines, and billboards in the case of traditional marketing, while digital marketing is more interested in social media, websites, or mobile apps. In the case of Crakmedia, it is obviously the digital platforms that interest us!
The main objective of media buying is to obtain maximum visibility and impact for an advertising campaign, while optimizing the return on investment. This requires a deep understanding of consumer behaviors, market trends, and the performance of different advertising platforms.
And in an increasingly competitive world, the use of AI-based technologies has become essential in order to maintain the edge over other companies like Crakmedia.
AI as a valuable support to operations
FT: How has the job of media buyer changed in recent years with the arrival of tools based on artificial intelligence?
GS: The arrival of AI in our operations is very recent, so we are experiencing that change right now. What will change rapidly for us in the coming months or years is the increase in our ability to optimize advertising campaigns. Time-consuming and repetitive tasks in particular will be replaced in part or completely by automated tools. They will allow media buyers to greatly increase their responsiveness and speed.
FJ: And it’s not just the repetitive tasks that will be optimized, we’re also talking about those that are extremely complex and difficult to execute in a reasonable amount of time.
We work with advertising networks that have extremely varied sources of web traffic, and it would be practically impossible to detect at a glance very small variation. An AI model can absorb all this data within a few minutes and give us the answers we need.
FT: What are the new challenges of the media buyer profession?
FJ: One of the biggest challenges is the multiplication of data. If we go back to 2011, all of Crakmedia’s advertising operations (and even administration and accounting in fact!) lived in Excel spreadsheets that were connected to our data sources.
Today, thanks to new management tools that give a lot of flexibility in our management of advertising space, we target cities, states, even times of the day much more precisely… It’s powerful and convenient, but it has increased tenfold the amount of information we collect about our traffic and therefore the difficulty of data analysis.
We now generate terabytes of data every quarter and it’s completely indigestible by a human being. This is where AI becomes very interesting, because it can help us process this astronomical amount of data with pretty much the same team.
FT: Do you have an example of work that AI is helping you do that would have been difficult to do before?
FJ: Recently, Gab had to identify which sources of quality traffic disappeared, when and what was the impact on revenue. It takes days for the team of traffic acquisition analysts to consume the necessary data and give an answer. Why? They have to receive the need, understand the work to be done, acquire and clean the data, process the data, interpret it and then bring out something actionable. And that’s completely normal, it does take time.
GS: I got an answer in a few [prompt] questions after having the data analyzed by an AI model. The analysis cross-referenced data from more than 7000 websites, over several months and the model was able to find that at a specific time, about forty sites had stopped sending traffic and that they represented a loss of X dollars.
FJ: We eliminate a lot of back-and-forth and downtime by using AI. We don’t need to plan a needs study meeting, and then follow up on the preliminary results, reorient everything because it’s not exactly what we wanted… In the end it takes a week to get an answer and it’s too late to react, the momentum is gone. In performance marketing, you often need to adjust in near real-time and it’s just impossible to wait a week for a response.
FT: What have these tools changed in the way you do media buying at Crakmedia?
GS: All the analyses that before should have gone through analysts or data scientists, but which require quick answers, we can either take care of them ourselves or at least predigest the data so that the analysis can be done in a fraction of the time.
FJ: The fact that our media buyers can take care of certain analyses thanks to AI, it allows analysts to look at hyper-complex analyses that have repercussions on our entire operations and whose benefits are major for the entire company. It’s important for a media buyer to understand why a given ad space performs worse, but it’s not necessarily worth it for an analyst to spend hours on it when he could have spent that time on a project that has 10 or 20 times more impact on the company.
GS: It really allows us to make moderately complex analyses profitable: the simple ones, we were already taking care of them, and the very complex ones, the analysts and data scientists were taking care of them. The middle ones, we couldn’t necessarily justify them time-wise or cost-wise, and now we can.
FT: Can Crakmedia consider itself a pioneer in AI-powered performance marketing?
FJ: Yes and no. No, because open-source models have existed for a long time and it took us a while before we gave ourselves the right to use them. There was a whole legal, compliance, and data privacy aspect to consider before we could even consider injecting sensitive data into an open source platform. All the data we collect is sensitive data that we need to use responsibly.
Where I think we are leaders is in giving ourselves the right to use this data and to put in place the infrastructure to be able to do so entirely securely. We still have a lot of work to do to use AI to its full potential, but we’ve been making great strides since we had access to ChatGPT at the end of October. When I look at the trend, I believe that by the end of 2025, we will be light years away from where we are today. I even think that we will surprise ourselves by completely exceeding our objectives and expectations.
FT: Do you encounter any obstacles or limitations when working with AI?
GS: Yes, but I would say the problem is that we limit ourselves because we don’t understand what the tool can actually do. Through trial and error with AI, we realize that the tasks it can help us accomplish are practically infinite.
FJ: At the beginning, when you start with AI, you have the false impression that you know how to use it… And the more you work with it, the more you realize the immensity of these tools and the possibilities… and a little bit of their “stupidity” too.
GS: Actually, it’s that AI is extremely literal so you have to explain to it like you would explain to a child. Your AI model at the beginning, it doesn’t know anything, and you have to teach it everything.
FJ: AI is “smart” in its ability to learn and its ability to work extremely quickly. But if you don’t explain to it what you want and how to interpret the data, it’s going to improvise something to satisfy you. It’s like asking a 3-year-old child to multiply fractions: they lack the knowledge and concepts to be able to do it. On the other hand, AI can understand extremely quickly.
GS: And the mistake that people often make when they’re starting out with AI is to assume that the tool is actually smart and will make complex connections on its own.
FJ: AI will be able to assimilate within an hour what you have learned in a year… provided that you teach it and that you understand what it does not understand. When it hallucinates stuff, you must have the ability to understand that it’s a hallucination and that it doesn’t make sense, and you can then ask it to explain its steps so you can correct its interpretation.
GS : And AI-based tools are usually really good at doing a specific task that they’ve been trained on… but not not very good at everything else.
FJ: I think it’s going to take a long time before we have a single Media Buy AI that can do everything and interpret any data. Instead, we’re going to have 10 tools trained on a specific aspect.
FT: So a human being still have a place at the heart of media buy operations despite AI?
GS: You still have to supervise its work a minimum, especially at the beginning and make sure that it’s quality before letting it work alone.
FJ: Every AI tool we develop, I consider it a new employee who has to prove themselves before I can let them work independently. In thier first year they are in apprenticeship, in years 2 to 4 they acquire his autonomy and from his 5th year there are able to train other people. It’s a bit the same principle for AI, but on an accelerated scale, within a few weeks.
GS: And the usefulness of AI depends a lot on the human using it: you have to know what to ask it, how to ask it, and have a rough idea of what you want to achieve. And for that, you have to master the subject and know how to interpret the answers. If you ask the AI to write you some PHP code, you have to be a pro yourself to be able to know if what the AI gives you is good or if it’s nonsense before injecting the code into your program.
FJ : People tend to have way too much confidence in AI right away and this can lead them to make a bad decision based on a misinterpretation. Hence the importance of being an expert yourself in order to be critical of the AI’s responses and detect nonsense.
FT: What is the main advantage for Crakmedia to invest in AI?
FJ: There is a big question of human resources. Traffic acquisition analysts are a rare commodity not only at Crak, but also on the job market in general: all companies need them, and there is not enough to meet the demand.
Media buyers are also a rare commodity on the market because formal training simply does not exist. We hire people who have backgrounds in sales, marketing, administration and teach them in-house how to become media buyers. We very rarely have someone who arrives with 10 years of experience and who can be profitable on day 1.
For 2025, it is one of our main goals to find a way to automate many of our most time-consuming operations precisely to compensate for this labour shortage to which there is no real solution in the short or medium term.
FT: What would you like AI to help you do in the future?
GS: Our tools are not able to make decisions autonomously at the moment. The dream would be for AI to be able to make decisions automatically based on real-time data without being asked. That the tool essentially learns to trigger itself with minimal supervision.
It would also be convenient if it was able to identify its own shortcomings, as a human being would, and essentially self-suggest things to learn in order to do its job better. But we’re not there at all.
FJ: We would like our AI tools to be able to create a complete sales flow almost autonomously without consuming too many human and financial resources. This would allow us to launch campaigns to validate hypotheses in the space of a few hours. Then, once the hypothesis has been validated, we could assign the resources to create a complete campaign… And if the proof of concept fails, we won’t have spent weeks on the launch of a campaign doomed to failure.
FT: I already suspect the answer but I’ll still ask you: do you think AI could completely replace a media buyer?
GS: No. AI will be a tool that will play an important role in supporting our work, taking care of repetitive or time-consuming tasks, but it will not be able to replace us. It’s going to change the way we work, definitely, and I think it’s going to be for the better. It will increase our capacity.
FJ: Has the computer replaced the human? Obviously, no. Has it allowed us to push our limits? Yes, absolutely! Before, it took 100 engineers with drafting tables to make plans for a machine or a building. Now it takes 5 with computers and the same 100 engineers can work on 20 projects in parallel. We are accelerating our development capacity, and everyone will be able to find their place in this new work organization despite the technological revolutions.
_________
Want to explore the possibilities of digital performance marketing? Come and discover the positions available at Crakmedia on our website: crakmedia.com/carreers
Explore our previous articles on the topic of artificial intelligence:
Share
