GPT Party 3.0. Galina Shubik: Practical Application of AI in Performance Marketing GPT Party 3.0, the largest Russian-language networking event dedicated to artificial intelligence, took place in Silicon Valley on March 9-10. More than 450 people gathered at Plug and Play to meet leading experts, entrepreneurs, and investors, discuss the latest trends in artificial intelligence, and gain practical knowledge. At GPT Party, Galina Shubik, co-founder of Affect Group, spoke on the topic “Practical application of AI in performance marketing.” Galina shared her company’s experience and talked about what tasks in marketing artificial intelligence already copes with better than humans, and where AI tools are still useless. Galina Shubik, co-founder of Affect Group It is clear that AI is having a significant impact on the marketing field. Every day, startups are popping up promising to automate all the work of marketing teams. I will use our agency as an example, how we use different tools every day. They save us a lot of time and simplify processes, although not all of them are fully functional. This is a really useful slide. You can take it with you and discuss it with your marketers. This is a process that we see every day in our company: six steps that correctly bring in customers from digital channels. It is clear that AI is having a significant impact on the marketing field. Every day, startups are popping up promising to automate all the work of marketing teams. I will use our agency as an example, how we use different tools every day. They save us a lot of time and simplify processes, although not all of them are fully functional. This is a really useful slide. You can take it with you and discuss it with your marketers. This is a process that we see every day in our company: six steps that correctly bring in customers from digital channels. Step 1: access The first step is to determine whether brands have customers in digital channels. This is simple: if the audience is broadly specialized, then there are customers in digital channels. If the specialization is narrow, we need help. We buy data from providers, Metadata has proven to be the best for us. We load this data into our digital channels, find profiles of real people and target them. Now the interesting part begins. The first three steps are what used to be done manually. Marketers could find the right people by socio-demographic characteristics or by average purchase frequency. Or you could buy the right data. And recently, a black box has appeared. This is what lives under the hood, in the brains of digital platforms. I will often talk about this using Meta as an example, but it works the same for TikTok, LinkedIn, Google, and so on. This black box is called predictive targeting. How does it work? We load a broad audience into Meta and say, “We need people who will buy our product.” Meta starts running an ad, a person goes from the ad to the site and buys, fills out a form or registers. What happens next? Meta knows perfectly well what this person did in the previous two weeks, what he liked, reposted, what sites he visited, what applications on his phone he used. And Meta starts looking for people from its large database that we gave it who are now behaving like a person who made a purchase two weeks ago. It finds them and shows our creative to different people at different stages of the journey. This is a part of artificial intelligence that is beyond the control of the human brain. This is where a real revolution occurs, because a person cannot analyze such a quantity of data. Of course, this does not work for brands with a narrow audience, for example, for small businesses. If you have a limited number of customers, this approach will be less effective and will require more manual management. Step 2: forecast You need to focus on your message because marketing success does not depend on pretty pictures. We need to push people to take the actions we want. Before we start focusing and launching campaigns, we need to calculate whether we should attract all these people through digital channels. Artificial intelligence is helpless here. A person analyzes their funnels, knows the audience capacity and expected conversions. A business knows how much money each user brings in per year and what ROI they want to get from the company. Our job is to test this environment and understand whether we can attract enough conversions through digital channels to get a positive ROI. Step 3: focus Focus is where AI significantly improves human performance. What is a focused message? For a successful marketing campaign, this means not just showing a beautiful brand, but using a number of life hacks. For example, it is necessary to offer a killer offer, a special offer that will make a person choose your brand, not a competitor. The use of biases, as studied by Google, plays a key role here. There are clear biases that influence a person’s decision to make a purchase. Now let’s move on to how AI helps us. First, there is the “blank slate” problem. When a strategist or creative has to start from scratch, it is very difficult. Then you can ask AI to create something, and then you rework it. Writing becomes much easier when you already have a basis. However, there are also things that AI does not help with. We tried various AI tools for writing marketing texts – from Jasper to ChatGPT. Unfortunately, they are not up to the task of writing short marketing texts. The texts created by AI are very boring and ineffective for our purposes. In addition, AI does not help in generating ideas either. We asked the AI for creative ideas for an organization helping immigrants at Stanford, but the options we received were unsuccessful. The idea proposed by the AI did not have the liveliness and sharpness that real creatives would have come up with. Another area where AI excels is in marketing focus groups. We ask it to pose as a group of 10 marketers, evaluate an idea, and give us its pitch. Surprisingly, what ChatGPT did matched the results of the actual focus group we ran — meaning they all picked the same idea. Another important aspect in which artificial intelligence helps us is image creation. Of course, the designer still has to make his own edits: add plates, change the position of the hands, adjust the colors. But the base can really be generated using artificial intelligence. This allows you to avoid the need to buy images from photo stocks and saves a lot of time on creating a basic image. This is a huge advantage for advertising agencies. In addition, this is how creatives created by a person look like, and this is how they are created by artificial intelligence. It seems that very expensive 3D is used, high detail, and at the same time, production is much cheaper with the help of artificial intelligence. It is important to note that artificial intelligence does not work without human participation. Recently, the founder told me at a demo: “Guys, don’t worry, if artificial intelligence draws badly, our designers will fix it.” Step 4: experiment Before launching a marketing campaign, we must test it on a small budget, check all the hypotheses to understand which one works for the brand. What is a hypothesis? These are the various targeting channels, messages, creatives, formats that we choose. There are also macro components, which include the choice of platforms, targeting and formats, as well as micro hypotheses, which include various banner backgrounds, texts, and so on. We must test each hypothesis to understand its effectiveness. It is important not to skimp on testing, as this can lead to underestimation of the hypothesis. Now we turn to artificial intelligence for help. Our task is to generate the maximum number of hypotheses. For example, 2056 variations were created for one of the real clients. If a person tested this, it would have taken us $ 45,000 and about a year of time. But artificial intelligence created space for us under the hood of digital platforms. How does it work now? We used to find entrepreneurs by interests, behavior, and database to test each group. But now we say to Meta: “Here are 10 million people, find us entrepreneurs who will take the desired action, and we will not limit you.” Meta uses its knowledge of each of us to show creatives only to those people who are likely to take the action. We upload any number of sets of messages, creatives, backgrounds, etc. as a content package and pass them to the AI. Meta starts testing combinations, optimizing them for the best results. It also tests different ad placement formats and chooses the most effective combinations of all parameters, which leads to the desired result. The AI manages targeting, creative formats and placement, and a human is responsible for preparation. Step 5: control Evaluating the effectiveness of an advertising campaign is a key task, and we use various tools for this. If the deal cycle is simple and happens online, we use GA4, which can build attribution models with maximum accuracy, determining where a person most likely came from, even if it happened offline. If the deal cycle is complex and long, we use ELLIE and Improvado. After launching a campaign, you can’t just stop it after the first few days, because the platforms need time to “tweak” their algorithms. A real-life example is a company that started advertising on three platforms: Google, Facebook, and LinkedIn. Initially, Facebook brought very few clicks, and it seemed that it should be turned off. However, after waiting for the deal cycle to end, it turned out that Facebook attracted a quality audience, and the final contract concluded through this platform became the cheapest and most profitable. Therefore, it is important to wait until the deal cycle ends before drawing conclusions about the effectiveness of the campaign. Step 6: thrive After testing, we now know which channels attract which users and at what cost. This allows us to move on to the next step – scaling. The partnership between human and artificial intelligence always leads to the fact that high customer cost and low number of customers begin to change places. That is, thanks to the joint work of humans and artificial intelligence, we can optimize our marketing strategy in such a way as to reduce the cost of customer acquisition and increase the number of customers received.
