The marketing world is just about coming to reality with the term “Big Data”, people have realised that A) They don’t have any “big data” and B) 90% of people don’t know what it is or how to use it anyway.
So luckily the industry has come up with another new buzzword for us to throw about… and like “big data” this is something that has always existed, but it’s been given a cool new name.
“Predictive analytics” is what people in the programming world would call “machine learning” and people in the economics world would call “economics” - woah!
So what even is predictive analytics?
Wikipedia defines it as…
So in essence, we look at all our historical data, and try to find trends or patterns that are reproducible… nothing new here…. we’ve all been doing it forever.
- If (A) then what is B?
- When I change A what will happen to B?
- Now we know what happens to B… should we change A?
Predictive analytics may be ‘the new black’ but is it actually useful to small businesses?
I’m going out on a limb and saying… Maybe… and I should probably justify that.
The type of predictive analytics that people are hype-ing about is the crazy big type™, that requires a lot of data, dare I say “big data”… people like Google and Amazon and Facebook know enough about how people interact with their product to be able to create the complex data models that make this type of analysis possible. Most companies don’t have the level of data to make statistically significant assumptions, at least not any that will have a real impact on their business. That doesn’t mean that small companies should sit back and do nothing, there are still interesting things you can do.
You can do “small predictive analytics”
“Economics for dummies” would tell you that price affects demand, if you increase the price, demand drops, and vice versa. Knowing that is a type of predictive analytics right? Maybe it’s not a machine learned, Hadoop clustered data crunched insight… but it is “predictive”.
Now i’m guessing that most companies have the ability to test this type of behaviour for their customers, and have probably done so in the past. What they need to do now is test in a controlled way. Measure the results, and boom! Predictive analytics for everyone.
Experiment with some split testing… this is in essence a simple form of “predictive analytics” we test the control vs the variation, and we build up a teeny little data model that compares just 1 variable, and we can say that “changing A affects B (our outcome/revenue)” in some way… we can then decide what route to take.
When performing any type of data analysis, segmentation of your data is usually the key concept that analysts (or machine learning algorithms) are trying to uncover. When these discoveries are made, you can then target that segment differently, or show different information. There is nothing new going on with “predictive analytics” in this sense, it is just doing it more extensively and in an automated way.
Amazon’s product recommendations.
It’s a well worn example, but every time you go to Amazon you see these recommendations, this list of suggested products isn’t based upon the actual suitability of these items, it’s probably more closely based on “people that have a similar purchase history to YOU also bought these items” - Amazon are suggesting products it thinks you will need, not products that play nicely with the one you are looking at. This is built up from a vast dataset of purchases from many many people. This allows Amazon to make predictions about what you are going to buy, just from what products you are looking at.
Uber and AirBnB
…use “predictive analytics” to set prices on their products. Check out this video from their developers to see the extent at which they are using these techniques. (Take note that Uber talk about one man frantically changing prices and hoping it doesn’t break the whole system)
Just because predictive analytics seems all scientific and data crunchy… it usually boils down to a person sitting at a computer changing numbers and seeing what happens.