Artificial intelligence (AI) is one of the most prevalent buzzwords of our time. Everyone expects it to change the world, from driving our cars and making our burgers to policing our citizens – even writing surprisingly believable articles called "How AI Will Change Marketing." And what industry could be better suited to an AI revolution than e-commerce, with its constant flow of big data and the pressure to lower costs through automation?
Don't panic. While AI is going to have a big impact in some very specific applications, the robots aren't coming for your jobs in the foreseeable future. With AI being such a buzzword today, it's thrown around far too often and not always accurately. So let's try to demystify AI in digital marketing to understand how it really works for e-commerce agencies and brands, what it can do and, more importantly, what it can't do.
What is AI?
The foundation of all paid search is the second price auction, a system in which the winning bidder pays the amount of the second-place bid plus $0.01. For example, let's say an Amazon.com advertiser sets their max bid for a given term – i.e., the highest amount they're willing to pay for a click – at $1. If they win the ad and the runner-up bid was only $0.30, then the cost per click (CPC) is set at $0.31, a penny more than the closest bidder. The challenge for advertisers is in dialing in their bids to drive volume and/or improve campaign efficiency. Naturally, AI can help.
There are so many factors to consider when trying to optimize keyword bids on Amazon: keyword traffic and competition, bid price, CPC, impressions and clickthrough rate, and advertising cost of sale (ACOS) at a minimum. This can be a laborious task for a human being. With an AI tool, you input just a few parameters (usually budget, ACOS goal, and target keywords) and the program does the rest. But what's really happening behind the scenes?
In practice, what we call AI is actually a combination of several technologies. An AI tool for Amazon Advertising takes in all these different signals – advanced tools even factor in inventory, competitor pricing, market share, and other inputs – and uses the data to build complex optimization models. These models then allow AI to "decide" whether to raise or lower keyword bids, and by how much, in order to accomplish your predefined goals.
To make these decisions, the tool might run the inputs through dozens of different algorithmic models, including statistic models, heuristics, Bayesian algorithms and machine learning. The "intelligence" of any given AI relies on the quality of its models and their ability to decide the best possible action to take. But crucially, it's never 100 percent confident – and that's a good thing.
If you take out all of the buzzwords and jargon, all you have left is math calculations and statistics. Because of the exponential growth of processing power and the availability of big data, AI has become very powerful. However, AI decision making still is fundamentally statistical. It always has both positive and negative examples, and therefore there's rarely a conclusive "right" action to take.
This is why AI necessarily relies on heuristics – i.e., shortcuts that allow for quick judgments and problem solving based on probablilities, as opposed to guaranteeing that any given solution is correct. This is important because AI needs to have the flexibility to adapt to new and shifting trends, to make educated guesses about the best course of action, and to learn from past mistakes.
AI is a machine. No matter how smart it seems, it doesn’t understand your marketing goals at a high level. And the only action AI actually takes is to update your keyword bids a few times a day, so it's not going to take over your marketing department anytime soon.
AI is limited by the budget and ACOS parameters you set. It can be very effective at optimizing bids within those parameters, but it can't fix a broken marketing strategy. Furthermore, AI relies on pattern recognition, which makes it very effective at optimizing for highly trafficked keywords but ineffective at executing unconventional campaigns.
For example, when launching a new product or selling off old stock before an upcoming release, you may prioritize volume over ACOS. In this case, AI's ability to find efficiencies goes against your marketing goals. And during promotions and deal days when the primary goal is to improve visibility, it's probably a good idea to take a manual approach to your campaigns.
Where AI comes in is at the tactical level. AI is great at fine-tuning paid search performance of established products for commonly searched terms during regular market conditions by identifying which terms drive the most conversion and reducing inefficient spend. However, AI can't respond effectively to long-tail terms, which are especially valuable for targeting new customers with niche interests. AI is most effective in campaigns that have at least 15 conversions per day, so long-tail terms don't generate enough search data for AI to identify patterns and optimize bids.
AI is also limited by what Amazon provides. Unlike Google, Amazon doesn't return position for paid search nor the exact time a click or a conversion happens. It only releases campaign performance reports a few times per day. Both Google and Microsoft have gradually exposed more and more data over time so we may see a similar growth in transparency from Amazon, but in the meantime AI's ability to test and learn is limited.
That said, AI will only get more intelligent over time. AI learns from experience through accumulation of data and Bayesian reasoning, and therefore will theoretically improve its performance the longer it's in use (not to mention improvements in processing power).
Even allowing for significant advancements, AI won't replace humans in digital marketing within the foreseeable future. There are a lot of strategic decisions that only marketing professionals can make. The uniquely human aspects of marketing can't be understood by math and statistics alone. Today, AI is a key part of a well-rounded Amazon Advertising strategy, and we're likely to see some specific new applications like customer care bots that can reference specific product details and order history, but truly human-like AI isn't realistically on the horizon.
Melissa Burdick is the president of Pacvue, an e-commerce SaaS platform that empowers brands, sellers and agencies to programmatically manage and optimize their advertising on Amazon.