Machine Learning - Marketing Hype or Something Useful?

Ok, so the elections are over in the UK, the dust has settled and everyone resigned. Seems like a good time to get back to normal and write a blog.

Something I’ve seen a lot of in social media and marketing communications in the last few weeks has been the heady topic of "machine learning". Is this some terrifying view of the future, marketing snake oil, or something which might add value to our lives? I believe the latter, so let’s make sure we all understand what we’re talking about.

What does it actually mean?

I’ve heard a lot of people talk about machine learning (ML), but it’s pretty clear to me many of them don’t actually have the first clue what it is. My general view on this is that if you can’t explain something to your Mum (sorry Mum), you probably don’t know what it is. So, keeping it simple, I’d describe ML as using clever maths to draw predictions from vast data sets which are provided as inputs. The commonly cited example of this, when done properly, is spam filtering.

Ok, but spam filtering has been around for some time, what’s new? Well of course it’s our good friend "public cloud computing" making this available to you and I as a service. What might once have required legions of engineers, plus a huge upfront investment in skills and tooling, now simply requires a cloud account with one of the major providers.

Recent launches

When the big boys are all in the game, you know it’s being taken seriously. Amazon, Microsoft and Google all now have offerings in this space. Google, in their common role as innovators, introduced the Prediction API some years back: see this announcement from way back in the mists of time (alright, 2011). It was a cool idea, but I’ve not seen/noted huge take up, maybe they were a bit early, maybe it was too developer focused. I would expect some pretty hefty investment in Google’s offering given the buzz in this space at the moment (and given Google’s own revenue generating interest in this area with ad prediction).

Microsoft stepped up publicly a month or so back with the general availability of Azure Machine Learning - and then with an acquisition of a well respected business in this space, Revolution Analytics (if you don’t know the company, you will likely have heard of "R", which they contributed to heavily). From my initial investigations, this looks like a slick offering which Microsoft say uses principles that they drive Xbox and Bing services with.

Amazon waded in even more recently with AWS ML and knowing AWS, they will iterate on this first release quickly and aggressively based on market and customer feedback. Obviously in their retail arm must be using something in this sphere around stock prediction and recommendation engines.

There’s plenty of commentary out there on which might be the "best", so I won’t duplicate it at length here. One such interesting piece suggests Azure has the edge thus far in terms of ML feature set. If so, well done to Microsoft for leading the way - something they’ve been criticised in the media for not doing in the cloud space previously. It’s worth noting that Microsoft do appear to have the most accessible platform and interface if you’re new to the concept of ML. This is highlighted here amongst a few other salient points.

Ok, what might I actually do with this ML of which you speak?

Case studies from the earlier adopters are out there already. If you’ve got any kind of field service management (typically hairy engineers maintaining stuff in the field) capability, there are some great references around efficiencies in predicting stock levels to drive tighter margins, or predicting faults *before* they occur so they can be fixed and ensure greater uptime/reliability.

Thinking beyond that, there are definitely applications for businesses in marketing and also for anyone who could benefit from more reliable automated fraud detection.

Not interested in business applications? Well, there are very likely social benefits as well linked to the analysis of large scale data sets around diseases like cancer and (for example) the prediction of which treatment plan is more likely to work for a given individual.

Any watchpoints?

Well, as with anything in the world of analytics, you either need some very smart people or you need to buy/consult with some smart people to make the most of the offerings. Large data sets are non-trivial, no matter how good the underlying technical platform.

You should also consider your commitment to a given vendor, as the work done with one will likely be largely 'stuck' there as the ability to import/export models between the platforms is not going to happen any time soon.

I’d also note that each vendor will force you to use their own platform as the source of the data - which does of course make sense, but just be aware you are buying in to their ecosystem.

Want to learn more?

Of course you do. There are lots of resources on the web, including some strong looking Uni led courses. Have a hunt around and you’ll find some interesting options for you.

Expect ML to be the next "big data" buzzword in the world of analytics. But it’s not just marketing, it’s here to stay and it really should drive value in your business - whatever industry that may be.