Machine learning, a part of the broader field of artificial intelligence, uses computers to spot patterns from data. Larger companies like Amazon, Google, Walmart, and eBay use machine learning in multiple areas of their businesses to great effect. But what about small businesses? Can they use machine learning as well?
What is Machine Learning Good For?
There are numerous blog posts out on the internet listing many different use cases. You have the more advanced examples like chat bots learning from customer support responses so you can improve your customer support quality without having to hire and train more staff, or image recognition to automate categorization of products (or spot categorization errors – such as “short dress”) based on your personal catalog. But there are many more mundane examples such as product recommendations “my customers who bought X typically also buy Y, so I should offer that a recommendation”. (Note the emphasis on *your* customers – you want to learn from your own demographics.)
I am not going to go into details of use cases in this blog post, but rather provide some considerations to think about when deciding if you should look at applying machine learning to your store.
Access to Technology
Access to the technology required for a machine learning project is definitely more accessible than just a year ago. Hadoop and Spark clusters can be spun up with a few clicks on cloud hosting providers; Google, Amazon, and Azure all have specialized machine learning offerings. Access to cloud hosted technology is easily accessible. Getting technology running is not the main problem.
The challenge is to learn how to use these technologies. Data scientists are in demand. What is your return on investment going to be on building up your own data scientist team? (I personally am wary of getting only one data scientist – single points of failure always scare me.) Are you going to get a return on such a staffing investment? The smaller the business, the less likely this is going to be true.
So smaller businesses are more likely to get benefit from one of the increasing number of vendors that use machine learning technologies within their product offerings. This can be a much more serious option for small businesses as it avoids the need to have in-house expertise. The vendor makes it cost effective by building up the expertise for you.
Volume of Data
So what area should you tackle first? Should you take on inventory forecasting, specific customer predictive personalization (offer a discount if the customer behavior indicates they may be about to leave your site without making a purchase) – there are many areas you could tackle.
One key point to remember is machine learning is primarily about learning patterns that emerge from data. If you don’t have much data, then the machine is not going to learn very well. For example, if you are trying to make personalized predictions based on an individual user’s behavior, but most customers do not return to your site often, the project is probably doomed for failure.
So don’t only look for use cases where you think you can get good ROI (which is clearly also important), think also about how much data you can collect in that area to feed into a machine learning algorithm. Do you have lots of anonymous customers or do they log in? Do you have a large catalog? Volume of data is one of the real challenges for smaller businesses that I don’t see talked about as often. If you don’t have volume, machine
Another consideration is site performance. If you have a more personalized experience you need to be careful of the performance impacts on your site. Personalized content caches less well. Is the improved experience you offer better than the performance hit customers observe?
Performance is more of a consideration of where it fits into your overall business. If you are using machine learning to optimize shipping expenses, then that will not affect the on-site experience of a user. If your site only has low traffic, again, caching may not be such a significant issue.
And if you do decide to go ahead with a project, think about how you are going to test your new solution. If an on-site customer experience, are you going to use an A/B testing framework to make sure the offering is improving? This is more important with an external vendor as you won’t necessarily have the same access to data as you would with an in-house team. Shipping cost optimization however you can automatically compare by computing shipping using two strategies and comparing the results (there is no need to split users into different groups to compare results).
This blog post is not a “all machine learning is hype” bash. On the contrary, I am a big believer in machine learning. But it is important to understand its strengths and weaknesses. Large can organizations get real benefits from machine learning. Machine learning is becoming more affordable for smaller businesses. But if you don’t have data volume, don’t expect machine learning techniques to do better than what you can observe yourself manually. Machine learning comes into its own when you have volume of data, more than humans can handle.
My general advice for smaller businesses is to find a vendor that helps you solve needs in your business. The fact that they use machine learning is actually secondary to whether they can deliver you a good ROI on your investment.
Building your own machine learning expertise however becomes more useful as your business grows and you have the data volume and potential benefit to make it a worthwhile investment.