This article was originally posted at http://mse238blog.stanford.edu/ as an assignment for course MS&E 238 at Stanford University 2017.
I’m not an engineer, I usually say, but I speak engineering.
As a business person in a tech startup, I had to learn the basics of how programming works, what it can do and not do. It is a great asset to be able to model and specify what a product or organization should do without having to always be told: “that’s not possible”. It lets you design products and organizational support with all the cool technology your team can build in a lean way.
If you’re a businessperson like me, an engineer or from any other major and you want to start your own company, lead or manage a team in any company, there are some ideas I want to give you about how to leverage machine learning, both in the product space (where I’m sure, if you’re an engineer, you know even better than me) but also in the business space. Because the truth is, you don’t need technology to be your core product to have technology be the core of different aspects of your business. Technology can simplify the life not only for your customers but also for your employees and your operations.
Pinterest is an example of a company – though with a tech product – who’ve integrated machine learning into multiple aspects of their business. They are among other things using machine learning to detect spam, increase the performance of ads and in their search to provide the best results for the user. There are plenty of other applications, however. I’ll list a few of them below.
Any product could technically build machine learning into its features. Two popular areas of use in product are recommendation engines and perfecting search. Recommendation engines guess what content users should be shown based on data from other users or that user’s data in particular. An example of the former is Amazon’s “People who bought this, also bought” and an example of the latter is how Facebook is personalizing your newsfeed depending on what you like and who you interact with.
A big part of product is also the user behavior in it. We can learn a lot from analyzing data, what features people use most and who does what, how and at what time. However, we might not always correlate all relevant data points that might have been interesting. This type of analysis is easy to automatize with machine learning. Why not let the algorithms predict what feature to build next for your product?
For those of you who are product nerds like me, you know that testing the product for bugs after it’s been built is a huge hassle. Actually, machine learning algorithms could learn what is correct and what is not and make testing autonomous. Prolifics is an IT solutions provider that uses machine learning to automate test design. Sarat Addanki, Head of Testing and Business Assurance Practice, talks about how they are doing that in an article from QA Financial.
“A machine-learning algorithm that generates test design can be much more thorough than any human tester. If a functionality has 100 different conditions it is impossible to prove that a manual tester is hitting every single one. The system can also verify that you are not repeating tests that look different, but are actually testing the same conditions. This maximises test coverage and minimises redundancies.” 
Revenue in sales is usually funneled down to how many leads we can generate for the company, giving us prospects, giving us customers when they are converted. But it’s not only the quantity of the leads that matter but the quality. With the help of machine learning, we can generate better leads by making predictions about each lead we get into our funnel, and labeling them with a probability of converting, for example.
With the data we gather about our users, we can keep retention up and prevent churn through marketing efforts such as email, text message or other types of push notifications to the user with machine learning. By analyzing user behavior, the algorithm can predict their future actions and upsell, cross-sell as well as converting users who are not paying or retain users that we’re afraid are leaving.
Customer support teams could to a large extent be replaced with algorithms that do close to the same work, except for some work that can’t be automatized yet. Nowadays, customer support bots are close to passing the Turing test; a test that measures the extent of which a computer could be “mistaken” for a human being . The likelihood is very high that you have experienced half-automated conversations in customer support without you knowing it.
How would you use machine learning to simplify the lives of your customers, employees or your operations? Would love to hear it!