The buzzword these days seems to be Artificial Intelligence (AI) and how it will solve for what humans seemingly cannot. We see it advertised in all sorts of technology suites, and to people like Elon Musk, it petrifies them as to the harm it can do if used incorrectly. I am going to focus on retail and stay away from the broader argument.
People far smarter than me are working on neural network type things to inject into technology and enable computers to begin to think for themselves, but can a computer really outsmart a human? After all, we have been told that the computer is only as smart as the person who programs it. However, when I think back to my Six Sigma days, we were taught that humans can only be accurate 92% of the time over a period of time, which is why training can never fix a process problem. You can’t train a human to be more accurate.
What do I think of AI in retail? I think it has its place, but I do not think of it in the same way that some of our competitors view it, or market it. I’m a strong believer that there must be a human in the loop, and that the idea of AI replacing the job of someone like a site merchandiser is crazy. If you look at Netflix, they paid a $1 million prize for a recommendation engine, only to admit a couple of years later that they needed human curators.
Where do I see Searchspring using AI in its products? We are constantly injecting more and more computer decision-making, and employing Data Scientists and Machine Learning Specialists to comb through the mountain of data we have, to determine how best to use it to increase conversion rates, as that is the ultimate goal. You will not see us touting that you can turn on some algorithm and the system will magically increase your sales. However, you will see us push recommendations on ways you can increase conversions, and then allow you to A/B test those ideas. We will also show you what we are doing. We like to call it “No Black Box AI”.
We want to give users and merchandisers superpowers. We do this by employing Machine Learning models, as well as operational tools that allow users to merchandise their site in the way they choose with the assistance of the computer – not vice versa. Only the merchandisers know the MAP policies, the margin goals or the co-op at a given time. While these are inputs that you could feed the system, how accurate would it be? It is best to leave the merchandising to the professionals, and just give them more power.
