Natalia Fedorova, Regional Coordinator

19 November 2020

3 Ways Machine Learning Is Used in Everyday Life

Machine Learning is a fancy IT term (also often a buzzword), and it also seems something distant and/or expensive from the outside. The truth is quite the opposite: we interact with the fruit of that tireless machine learning labor practically every day. Here a few examples.

Machine Learning is a fancy IT term (also often a buzzword), and it also seems something distant and/or expensive from the outside. The truth is quite the opposite: we interact with the fruit of that tireless machine learning labor practically every day. Here a few examples. 

Spam Filters

Do you know how now and then a website would ask you to check the spam folder for their registration confirmation? Well, this is largely on such websites. It is generally these days to have a genuine personal email or a consensual and beneficial business email to be flagged by a spam filter. This is a somewhat recent achievement that was not truly possible without machine learning algorithms.

Let’s consider good old RayBan 90% discount scams. It does not take a genius (or a machine) to deduce words that scream fraud, be it the discount percentage or a 1-day shipping policy in the early 2010s. Great, so we can just add “90% RayBan” and “same-day shipping” to the spam filter and call it a day? Well, yes, but not in 2020—or, for that matter, 2015. 

The truth is, a blanket ban would lead to false positives or even a lot of false positives if you’re actually subscribed to multiple clothing stores. It would be frustrating for everyone involved to have an email that talks about both a 90% discount on basic clothes and a RayBan re-stock to end up in the spam folder. The lack of action, of course, would lead to spam emails landing into your inbox.

Email services have long been using machine learning algorithms that take into consideration and cross-reference multiple features of an email before placing it. A basic example would be that 99.5% of users do not mark ASOS emails as spam, so the spam filter should not be triggered even if ASOS offers a RayBan at a 90% discount and ships them the same day. Some features are less obvious than the rest: for example, all email domains have a basic multiplier that indicates how likely it is to be used by spammers. Examples of false-positive protections include links with buttons (both spammers and online stores love them), big header images, and a phone number at the end of the message.


Machine Learning powers the anti-fraud effort on a large scale. While some transfers are reviewed on an individual basis (e.g. international transfers in non-SWIFT countries), most transactions are approved seamlessly. That seamlessness, of course, comes from Machine Learning solutions.

Hundreds of banks already prescreen all payments and transactions to detect unusual behavior. Some criteria are quite simple, such as comparing the country of offline merchants with the owner’s residence. Others are much more complex, as banks try to discern patterns of genuine clients and compare them against patterns of fraudulent individuals, identity thieves, carders. 

Once again, hardcoded (as in predefined) anti-fraud measures have been used by banks for a while, but false positives were rampant. It wasn’t uncommon to get your card locked for innocent recurrent payment: my colleague personally had that happen after topping up a PlayStation Network multiple times. Depending on the bank policy, such errors could force you to reissue the card—or lock you out of funds while abroad. Open-sourced solutions easily have 90% success in both tagging fraudulent transactions and approving genuine payments, and proprietary models go further than that.

You could, however, argue that machine learning solutions in banking & finance do not necessarily benefit the customer. The infamous banking score has been getting more aggressive as the software and both process more data and extracts deeper insights from it. Given how liberal free services tend to handle user data, it’s totally possible to have a co-signed mortgage application declined if one of the partners has been spending too much time on Tinder lately. The same applies to frequenting bars too much or spending on particular products or services disproportionately to your income. 

Customer Support

In the truly pre-automation era, customer support is what corporate execs often perceive as bloat. It is especially true for the first line of communication, where most of the inquiries come from the lack and/or inconvenience of looking up information elsewhere, not the fault or a temporary issue with the service. The challenge here is bridging the gap by providing on-demand information and fulfilling requests automatically while maintaining the traditional and/or more convenient contact options.

Nowadays, engineers use Neural Language Processing that engineers use to group user queries into options suggested on the chat and also expand the dictionary. Whenever a client request had to be processed manually from step 1, the algorithm analyzes the initial request (one that the machine couldn’t understand) and stacks it against similar requests that could partially or fully be addressed without human input. Next time around, “I would like to withdraw my investment” (and something along these lines) would automatically bump the client to the second line of customer support for a Kickstarter chargeback request. 

NLP also comes in handy for voice calls. The “robot” can identify you by your phone number and further verify your identity with card digits or the secret answer. From there, the system can either change that forgotten PIN or get a customer support agent on the line for something less trivial. This is exactly how you apply the latest technology to yestercentury way of customer interaction.

These days, Machine Learning starts with Python. We still have a couple of spots left for our introductory Python course starting on November 30. We will both set you up for some programming-centric jobs and give a foundation for mastering Machine Learning tools.