The AI-Human Collaboration at Work: Q&A With Albert Technologies CTO Tomer Naveh
There’s a lot of talk about artificial intelligence (AI) in marketing and advertising, but often what people are referring too simply automation. But the past couple of years have brought what can really be considered AI into the marketing and advertising realm—namely, technology, platforms, and services that use machine learning, new sources of data, and increased and more ubiquitous computing power to learn, improve, and execute. One such platform is Albert Technologies, which essentially sets up, runs, and optimizes digital marketing campaigns across multiple channels including Facebook and Google, freeing up marketing professionals to do more strategic and creative work. Far from taking jobs, it’s actually created them, as evidenced by Houston’s Gallery Furniture, which put more and more people to respond to social media leads (all while seeing a 22 percent increase in revenues and a 394 percent increase in their Facebook ad spend ROI after just two months of using Albert).
Business improvements aside, Albert is a fascinating early implementation of the ways that marketing and advertising professionals and machines will work together in the future. To find out more about how the platform works and what marketing professionals can expect from an increased partnership with AI in the future, we spoke with the company’s Chief Technology Officer, Tomer Naveh.
We’re in a digital marketing world that’s jam-packed with tools that help marketers optimize ads. How is Albert different?
If you look at one of those LUMAscapes for the marketing ecosystem, there are hundreds of hundreds of companies solving each and every problem of digital marketing, but most of the work of actually running the marketing campaigns – optimizing them, analyzing the data, and continuously improving them — is still done by human marketers. They have all kinds of power tools to help them manage campaigns with 500,000 keywords or targeting a million websites, but it’s still all power tools and absolutely all the decisions are being made by people. But people can’t really manage more than a few hundred keywords at a time, not to mention having to do it across multiple, ever-growing and -changing platforms. We view and approach this process a bit differently. Albert connects via an API to all the different execution platforms – email, Twitter, Facebook, SMS, Google AdWords, display – and then basically sets up the campaigns and manages them automatically, across platforms, in real time, and at scale. It takes over a lot of the tedious, nitty-gritty work that humans are still doing and amplifies it.
What makes it AI?
It learns and gets better over time. The way we work is we have an initial integration process where we basically connect Albert to your AdWords account, display account, and analytics (Google Analytics, for example), and sometimes your CRM. Albert then basically sucks in all the data and learns about your business. That’s the initial stage. Then we basically start taking over campaign management. Most organizations already have ongoing activities and we take those over and we optimize, getting you more ROI or more revenue, depending on your business goals. Others may want to run dedicated campaigns, which is really quick and easy– just open up the platform, tell us your budget, upload some creative materials, choose the channel or channels you want to target (social, social display, search, etc.), and then just press play. Albert then tests and optimizes in real time, spawning hundreds of different campaigns on different platforms, shifting budgets, bidding, segmenting audiences, not to mention coming up with new keywords or interests on Facebook or people to follow on Twitter – basically all the tasks you would do as a human campaign manager – but on a much larger and more precise scale and 24/7.
Where does this leave humans?
As with all AI, humans are an essential component. Business and strategy considerations — your overall budget, for example, and your business objectives – are done by people. And the creative and creative strategy –what brand message do you want to convey – also falls on people to generate and provide assets. Those are the sorts of decisions and executions that we believe humans are better at. But which website should I advertise on, which keywords should I target, what should be the specific budgets between Facebook and Google, how often should I send emails to my customers – these are executive decisions that make less sense for humans to take on. It’s all fairly repetitive manual work that can be done much better by machines, which can analyze and process millions of interactions across multiple platforms and audiences at a time. Even very experienced campaign managers will tell you that they have intuition as to what’s going to work, but when you actually try to understand what’s behind this intuition, you realize it can all be built into statistical models that will basically take on the same steps, and do it much faster on a larger scale and continuously.
How does Albert pull off all of this optimization?
There are different algorithms for different problems and they all have to work together. For example, there’s one that decides how much to focus on a specific search term in Google. And the bidding for that keyword might be affected by the audience that the person running the search belongs to. So, the algorithm needs to take that into consideration, and most of the processing is done in real time. Bidding can be very different in just two hours because a competitor launches a campaign and they’re bidding for the same people or keywords or audiences. We also do a lot of language analysis to understand the relationship between keywords and phrases, as some word choices and combinations work differently depending on the audience. And we have algorithms that can come up with 50 or 500,0000 keywords that are relevant to a product and test them out in a smart quick way without spending a lot of budget. That’s something that would take a campaign manager a day or two or sometimes more for just 2,000 or so keywords. And again, we can do that across many platforms all at once because we’ve developed a machine that understands what keywords need to go into the campaign.
It sounds like creatives will still be very busy. What do they have to provide Albert on a day-to-day basis?
Well, literally you’d upload images and videos and text for your business and ads – like your slogan or other copy – and you’d upload several variations of a promotion or ad that are fairly interchangeable. Albert then takes those base materials and mixes and matches the images and different types of text, creating many variations on your creative and basically testing them, optimizing the combinations that work better, matching the creative towards audiences that are in different stages of brand or product awareness, consideration, or about to buy very soon. Albert can make suggestions as to which keywords and phrases are working better, and what types of images with what word and phrase combinations and pairings – all of which can then be acted on by human creatives. Creatives increasingly will need to understand how to work with these kinds of systems in the future, to respond to insights such as “this creative really worked well, can you come up with three or four more variations,” or “creatives that have the word X in them underperform, so please don’t upload any more creatives of this nature.” I think a lot of the human role in the future will be in planning creative business aspects of marketing and less in the technical media buying aspects.
So, what does this mean for media buyers or other marketing roles?
Well, I think we’ll see less and less of just plain media buyers, but these roles will evolve over time, and mainly it’ll be in the way that people operate the systems. The technology is still in a position where it needs to get feedback from people on its decisions. I’ll give you an example. Say I’m currently at 300 percent ROI, and I need Albert to get me to 600. Albert goes ahead and optimizes and within a couple of months, give or take, we get to 600. That’s great, but suddenly that put a lot more emphasis on the high-priced products, which means I’m selling less of the lower-priced products, and that is actually not good for my business because of logistics, supply chain, and so on. I need to be selling more from the lower-priced products, too, even if my ROI isn’t as high. So that’s a different goal, which is to increase my business. So ultimately you need to understand that you have a very strong weapon in your hands. If you tell it to get to some point, it will get you there, but you need to be very clear and consistently and interactively correct your guidance, because now you really can get too many business objectives and you have to choose wisely and understand how it affects your business.
So basically, humans will be just as busy and making just as many decisions, but they’re at a higher level.
Yes, the pace and complexity that you can think about your business gets to be much higher when your campaigns optimize faster, and you don’t need to spend time doing that, so you’re free to really understand the strategic effects of your marketing on your company. It won’t happen overnight; people need to learn to understand how to work with an AI system, but it will happen eventually.
Illustrations by Leandro Castelao