How it Works

Pricing

Blog

Sign In

Try For Free

How it Works

Pricing

BLOG

Sign In

Try For Free

Businesses are collecting massive amounts of data as a part of their analytics pipeline. Most of the time, this data is filtered by a computer and presented in a way that a human interprets, through the analytics dashboard. That's a fantastic resource, and has no doubt been of great value to you in business decisions. However, restricting the interpretation of all that data that you've mined to humans leaves a lot of potential insights on the table. One method of recruiting a computer to look over your data and provide you with additional actionable insights is called anomaly detection. In this post, we'll discuss what anomaly detection is and how it can help your business.

What is Anomaly Detection?

One of the ways that computers can help parse the data for you is to use that data to learn what the normal operating metrics are for your company. The computer will know, for example, when your busy seasons are. Do you usually earn more in July than in January? On its own, this isn't very useful. You now that information as well. Let's say one year you earn less in July than usual. That's an anomaly that the computer can alert you to. Again, this simple example isn't particularly useful. You would know that without the computer's help. But it gives you, in simple terms, an example of what the computer can do.

The real strength of anomaly detection comes when the computer is constantly crunching the data on the hundreds or thousands of analytics variables that you feed to it. This gives it the possibility to detect much finer-grained anomalies than a slower than usual month. If you have a slower July than usual, then you know to look for potential causes and address them. With computerized anomaly detection, you will be alerted to many more such opportunities to adapt your business. Opportunities that, unlike simple monthly sales figures, a human is unlikely to catch.

What Types of Anomalies Can Be Detected

We've already given a simple example of the type of anomaly that might get detected by an anomaly detection algorithm. Let's take a closer look at the three major types of anomalies that detection algorithms are trained to recognize and alert you to. The anomalies can be tied to any of your KPIs, or any other data that you are able to feed the algorithm in order to give you a more complete picture of how your marketing effort or overall business is performing.

  • Point Anomalies - Also known as global anomalies, this type of anomaly exists well outside of the norm for the data being examined. This type of anomaly is often used to detect fraud.
  • Contextual Anomalies - These anomalies are only outside of the data for a given circumstance. Our sales example is a contextual anomaly. If sales are lower than normal in July, they might be perfectly normal for January.
  • Collective Anomalies - Collective anomalies are anomalies that might not be out of the norm when taken individually, but combine to form a pattern that is outside of the norm. Running low on inventory is a common occurrence, but if an unusual number of items are low then it may indicate a problem.

Anomaly Detection Methods

The earliest anomaly detection was done by hand. It's the same thing as your human employees do as they look at your analytics and your sales figures. As the amount of data to be interpreted grew, so too did the methods used to interpret it. The simplest form of computerized anomaly detection uses simple statistical methods to look for outlier data. The computer will calculate average values for your metrics and alert you when something goes outside of those values.

Our previous example of seasonal sales figures shows one area where simple statistical analysis is weak. What is considered to be a normal metric might change depending on the circumstances. In this example, it is the season, but it could be any number of complex circumstances that would result in a false positive using basic statistical methods. For this reason, machine learning algorithms have been developed that do a much better job of looking at your data and examining all of the variables. This allows them to more precisely identify when something is actually an anomaly in regards to normal operating metrics of your specific business.

The Benefits of Anomaly Detection

Anomaly detection is not limited in its use to just sales applications. Any data that operates outside of its normal bounds is something that a business is likely to want to be made aware of. You can probably think of a number of uses specific to the needs of your business. Here is a list of some of the things businesses use anomaly detection for.

  • Monitor the performance of a network or data centre to detect problems that may impede performance or customer experience.
  • Monitor sales funnels for anomalies that may indicate a problem such as a pricing error or potential opportunities.
  • Get early detection of potential security issues that could put your or your customers at risk.
  • Be aware of website errors or other malfunctions that can degrade or impede the user experience of your site as soon as they happen.
  • Take a more detailed approach to load balancing on gaming servers, or even balancing of gameplay elements themselves.
  • Detect potential security issues, such as employees accessing data that they usually don't access.

This list doesn't come close to covering all of the use cases for anomaly detection, but hopefully it has given you a better idea of what the technology is used for so that you can better imagine how it can be adapted to your use cases.

Conclusion

Depending on your industry, anomaly detection can have benefits throughout every aspect of your operations. Regardless of your industry, the technology can help revolutionize your marketing efforts and allow you to take control of your data like never before. It is the next generation of analytics processing and is quickly becoming the baseline of how successful businesses run their campaigns.

Related Articles

See all articles