I don’t know if you have a robot lawn mower at home. Honestly, I do not have one, I always liked to cut the grass on my own, but who knows, I may buy one in the future. But… hold on a minute, what a robot lawn mower is doing? You start it, and it cuts the grass automatically. Meanwhile, you can read...listen to your favorite music or even work. I mean... it is not as precise as you, but in return, you get more time.
But I have never heard of a hairdressing robot! Apparently, the problem is similar - I have something to cut, the only difference is that this time it's not my grass but my hair. Why cannot we buy a robot hairdresser if there is a robot lawn mower? What is the difference?
Robotic lawnmowers have proliferated because the task is relatively generalized: the lawnmower can mow anything in its path, and all it needs is a (relatively) flat ground and a fence to turn back from.
In contrast, the situation for a robotic hairdresser is a little more complicated. For a haircut, a lot of information has to be transmitted: how long the hair should be after the haircut, which way the hair should be combed, etc. Moreover, unlike with a master hairdresser, the "ground" is not fixed - as many people, as many head shapes, which of course the robot-hairdresser must take into account. And let's not forget that there is less room for error. If the lawnmower cuts a flower, it's a sad but correctable "accident", whereas if the robot-barber cuts your ear, it's a very sad and probably uncorrectable accident!
But why am I writing about robot lawnmowers and robot hairdressers? Recently I was looking for software that detects insurance fraud. I found several of them - and I was curious, professionally, if there is a universal "fraud" model that works for all insurance companies. According to the promotional material, of course, there is one... Or is there?
In the detailed description of the software, which uses the slogan "Advanced Predictive Analytics Solution for Fraud Detection and Prevention", I found the following diagram:
Yes - this is exactly the same as the steps of a general data science project. We would do the same thing if we didn't have a model (software) at all, but we would undertake to build a fraud detection model.
And much the same is true of most "out of the box" AI software. In fact, these softwares provide a framework that supports the business processes of a given problem in software, but the predictive model itself always has to be built on the data of the company. And this is perfectly normal.
Remember why there are no robot hairdressers!
· Each person has different needs for haircuts -> likewise, each insurer has different expectations from a fraud-detecting model
· Each person’s head has a different shape-> insurers' data are also in different database structures and formats
· You can't make a big mistake -> if a fraud model is inaccurate, insurance fraudsters will continue to cause huge damage to the insurance companies
Lesson?
When a company knocks on your door saying it has a ready-made AI solution, it's worth being skeptical. A "one-size-fits-all solution" probably means a framework (which is undeniably valuable), but it will require a separate project to build the AI model. That is why if a company wants a good AI model, it is worth looking for a good "hairdresser" - i.e. a good AI consultancy.