Recently, it feels like there has been another big push for Machine Learning (‘ML’) aimed at the SME market. Several leading cloud providers are offering more convenient ways to perform ML, simplifying much of the setup. Plus, the marketing machine seems to be more aggressively targeting SMEs – my feeds and email are jam-packed with ML webinar invites and free e-book download offerings!
After discussing ML with dozens of contacts within this space, I can feel a general sense of confusion as to what ML is, and how it can help small-mid sized enterprises. So, I thought I’d write this short piece to help ‘point things in the right direction’.
Timeout! What is ML anyway? It’s the same as AI…right?
Not quite, and there’s quite a bit of confusion here. The term Artificial Intelligence (‘AI’) is a high-level term which addresses how we can get machines to ‘think’ like humans and behave in intelligent ways. It’s a broad goal and encompasses many disciplines. The term was coined in the 1950’s and gained a huge marketing hype…before nothing much happened, and people started to roll their eyes at the term ‘AI’.
Fast-forward 40 years and there’s a new kid on the block ‘Machine Learning’ (or ‘ML’ for short). ML is a subset of AI, it’s just a part of the broader AI picture. Simply put, ML is about training a machine to recognise patterns in data, so that it can make accurate future predictions with new data that you give it.
Interesting, but how does that help me?
Good question. Many examples that I’ve found are all very similar, suggesting that this hasn’t quite found its place in the world yet (in my humble opinion). That said, here are a few of the more interesting examples I found:
Healthcare: Training a machine to detect anomalies in x-rays. Putting the ethics of this aside, this is very interesting. Healthcare providers have been training machines to recognise things like tumours in x-rays. They feed the machine lots of examples of both healthy and non-healthy x-rays, and the machine starts to ‘learn’ what to look out for. Other health providers are leveraging predictive insights from patient profile data, with the aim of more accurately identifying patients that may be at future risk of an illness.
Marketing: Imagine if you could accurately predict the lifetime value of a customer from their attributes. What smart decisions could you make to boost your ROI? By analysing customer profile data vs their lifetime value, companies are more accurately predicting the lifetime value of a new customer. They’ll then make their propositions more attractive to those who they believe to be of more value – perhaps initial discounts to get these customers onboard?
Demand prediction: No-one wants to hold more stock than they must, or even worse run out of a popular product during its peak. B2C companies are combining sales trends with data such as weather patterns, social media sentiments, currency strength, interest rates etc. Personally, I think this is very exciting, however it’s clear there are macro-economic forces at play, lots of variables and thus I suspect making accurate predictions is more difficult (and probably requires huge datasets).
So, what’s the problem?
The biggest problem that I see is that ML relies on one thing more than anything else…data and lots of it! For any reliable pattern to be ascertained, the dataset provided to the machine must be of a decent size. Hundreds of records won’t cut it, thousands of records might give you something, but I suspect that tens of thousands of records and above would be required to discover ‘accurate’ patterns.
Many SMEs simply don’t have that much data. And for those that do, it’s sometimes very messy data which would need significant work to get it into a state ready for analysis. The ‘big guys’ have huge amounts of data that has been cleansed and is ready to feed to the machine, I suspect most SMEs do not.
As an SME, should I ignore ML?
If you are one of the few that has ample clean data, then ML could present a great opportunity for you. For those who are not in this boat, there are still some ways in which you can engage ML:
- Use external datasets: No-one said that the data had to be your own! External data can be purchased for a huge range of subject areas. This includes market specific data as well as more general economic, weather and sentiment data. Take a look at the AWS Data Exchange to foster a few ideas. But be careful, data is not always cheap, and it may be wise to get some expert advice before making a large investment.
- Make use of pre-packaged services: Many companies are providing ‘turnkey’ AI and ML solutions for generic tasks such as image and text recognition, document extraction and threat prevention. There is an opportunity to use these solutions to automate labour intensive tasks. For example, a client of mine recently started to use one of these solutions to automatically extract data from invoices they received, saving 2 days a week of manual data entry.
A final thought
When researching this article, I came across something that annoyed me (quite a bit actually). Several well-respected websites were talking about AI and ML solutions that were anything but.
For example, one was talking about a garage sending automatic reminders to people whose MOT was due. Or redirecting support requests to the right person based on a keyword in a question. Call me crazy, but I do not see this as either ‘intelligent’ or having any element of ‘learning’. These are just examples of simple logic, which we use every day.
One final, final thought. It’s tempting to use shiny new toys just for the sake of it. In essence, becoming technology-led, rather than business-led. The best advice I have heard is to identify business problems and see how ML (or any other technology) can solve them – rather than inventing a business case to use technology. The former is far more likely to deliver real value.
