Crow Canyon SharePoint Blog

Predictive Analytics: Yea or Nay?

August 14, 2017

Office 365 Graph Delve Big DataThe latest buzzwords in IT revolve around the use of historical and current data to predict the future: Predictive Analytics, Big Data, and Data Science are all topics getting a lot of traction these days. The general idea is that massive volumes of data undergo cleansing (formatting, parsing, outliers removed) and are then fed into an algorithm-powered model designed to convey the most probable outcome of a given situation.

Data Science is practiced across multiple industries; for example, the insurance industry uses real-time driver data to calculate future premiums while the retail sector uses customer loyalty data to predict customer behavior and, subsequently, reduce churn.

All of these technologies are designed to accomplish what was once the province of shamans and wizards: predicting the future. Tea leaves and arcane spells are being replaced by datasets and algorithms… while the latest incantations are coded in R, Spark, and Python. Times are changing and the Predictive Analytics hype is ubiquitous.

Microsoft Gets On Board With Office 365 Delve and Graph

In early 2015, Microsoft jumped on the Predictive Analytics bandwagon with two additions to the Office 365 family: Office Graph and Office Delve. The former is the hardcore Data Science component: churning data, running models, and spitting out predictive data. The latter is the front-of-stage performer designed to convey intelligent content to Office 365 users. You may want to check out our previous blog articles where we introduce this duo and also discuss how Office Graph will play a more dominant role in powering the future of Office 365 content.

In addition, many analysts consider that access to a source of Big Data was a major reason Microsoft acquired LinkedIn in 2016.

It’s clear that Predictive Analytics is here to stay and, if the marketing is to be believed, will shape the analysis & presentation of content for years to come. But… is this a good thing?

The Other Side of the Coin

By its very definition, Predictive Analytics carries a lot of heavy baggage with it. After all, it’s designed to predict the future likelihood of events… but what happens if the prediction is wrong? If your entire marketing strategy is dependent on a Big Data-derived prediction, then you may be in for a surprise if the dice don’t land your way. And, really, the “dice” metaphor is applicable here — no matter how certain an outcome appears, it is still unknown… and trying to fathom the unknown is always a gamble.

Organizations that plan to test the Data Science waters should be aware of some unique considerations:

Data
Data lies at the backbone of accurate predictive analytics. If your data is flawed, then the most efficient model in the world won’t produce an accurate result. Do you have enough high-quality data? The accuracy of predictions is directly related to the quantity and variety of data analyzed. All relevant factors must be considered when formulating a data project, and each of those factors needs to contribute meaningful information… preferably in abundance. If the quality & quantity of raw data is lacking, then the results will be compromised.

Timing
One of the biggest hurdles of Data Science is that machine learning (i.e., artificial intelligence) is constantly behind the 8-Ball. For predictions to be accurate, they need to hit constantly moving targets. This is particularly true for real-time content delivery models, such as Office Graph and Office Delve. The goal is to serve up content that is relevant & meaningful to the user at a specific time — the problem, however, is that humans are unpredictable. Our interests, likes, and dislikes change rapidly, often without any indicative historical markers. For example, a Predictive Analytics-powered online marketing app may keep pushing ice cream advertisements… but what if we just started a diet? Or what if the user suddenly craves pie instead of ice cream?

If Predictive Analytics relies only on historical user data, then predictions may or may not hit the mark. The only way to mitigate this is to request user feedback to fine-tune the predictions (e.g., “Was this advertisement applicable?”), but one look at survey completion rates is more than enough to prove that this solution is problematic at best.

Black Swans
In 2007 the book Black Swan: The Impact of the Highly Improbable, by Nassim Nicholas Taleb, made its appearance. The book was groundbreaking in its two straightforward assertions: a) highly unlikely & rare events happen and b) humans tend to simplistically rationalize such events in hindsight. Insofar as Predictive Analytics is concerned, these rare outliers can totally demolish even the most well-modelled prediction. If backend algorithms do not factor-in the possibility of unlikely events, then predictions will not be slightly off… they’ll be completely wrong. This can have a major impact on organizations who rely on Predictive Analytics to help them make major decisions.

Does Predictive Analytics Have a Place in IT?

In short, yes. Despite the concerns mentioned above, Data Science, Artificial Intelligence, and Predictive Analytics is the future — there’s no stopping it. Issues arise when organizations misuse the technology. After all, predicting the future is like a double-edged sword… it can make a significant impact, both positively and negatively:

  • Selective Usage: The power to predict can easily go to people’s heads and the urge to start predicting everything may come into play. Remember that this technology can only work when the quality & quantity of meaningful data is available and, even then, the results may be skewed.
  • Decision Adviser, Not a Decision Maker: Predictive Analytics should be used to help decision makers formulate strategies and, of course, make decisions. It should not be solely relied upon to drive the decision making process.
  • Little Stuff, Not Big Stuff: Using Predictive Analytics to help formulate a marketing strategy or to deliver online content is one thing, but betting your company’s assets on the likelihood of a machine learning-based prediction is folly.
  • Sensitivity to Security: Data Science is being increasingly used to deliver content to a high volume of users, particularly internal users. Putting your content delivery on auto-drive without the proper security safeguards could truly wreak havoc in your organization. Be particularly careful of access rights, permissions, and the type of content available for dissemination.


Crow Canyon Software delivers the power of a connected, digital workplace to your organization, helping you be more successful and competitive. We provide the tools that allow you to gain the maximum benefit from SharePoint and Office 365, Microsoft’s premier collaboration platforms.

Want to learn about how our business productivity applications for SharePoint and Office 365 can power up your organization? Give us a call at 1-925-478-3110 or contact us by e-mail at sales@crowcanyon.com. We look forward to hearing from you!

Predictive Analytics: Yea or Nay? was last modified: by
Please follow and like us: