• by Judah Phillips , June 10, 2011

During the past few years I've had conversations with many analytics practitioners, consultants, and vendors about "real-time" data. I tell people that real-time data for human analysis is not useful unless automation is involved; then, I explain that I prefer timely data. I can understand the value of information and insights in real time, but I still do not see much, if any, utility of real-time data for human analysis.

Few business decisions are made or site optimization activities are made in real time. In fact, analysis and the resulting business-focused insights and actionable recommendations can sometimes take months to execute --whether on a Web site (within a controlled release schedule or more ad-hoc), as part of an inventory fulfillment process, or within a mathematical model that results in some downstream business change (such as a promotional mix or offer modification across time zones).

People have told me about very rare cases where real-time data was acted upon in real time. One case is online marketing or CRM emails, where looking at real-time data about whether the campaign is driving traffic back to a site might be useful for making sure the campaign worked immediately. However, acting on that data won't be done in real time. Thus, I wonder what business impact would occur if the data came in three hours later?

What is required is timely data to correct this issue. It's going take time to fix whatever problem occurred and re-traffic the campaign or re-email the list. And with today's campaign management technologies, if you've messed up a campaign, it's likely a process or people problem, not something real-time data should be used to identify.

The same can be said for real-time site data where people claim you can see if the site is working by watching the data move across time. Sure, it looks cool to see the data tally in real time. It's fancy. But any resulting recommendation based on "watching the data" will take time to implement. Timely data is necessary and required for creating analysis and insights that help to guide decisions generating revenue or reducing cost. But the timing that defines timely data depends on the situation. Timely data is data that you can use to make a business-critical decision delivered when you need it. Thus, timely data can come five minutes, five hours, five days, or five weeks, depending on your business needs and goals. And timely data will suffice, because the action that results from any decision is very rarely executed in real-time. Think of operations management and inventory replenishment. It's not "just in real-time." It's "just in time" because the process is timely to business needs and goals.

Where real-time data is useful, without a doubt, is for automation. Consider a highly advanced behavioral detection system where the data input is used to render some immediate result or experience.

Take a simple case from the financial services industry. A bank knows the average account balance for each customer. When an outlier deposit is made, say one million dollars, the bank detects that deposit, and, when the customer next logs into his online bank account, the real-time detection of that login -- in the context of the outlier deposit -- could be used to automate the delivery of a promotional offer for a mutual fund or other financial instrument. Then, if the customer does not act upon the offer, a discount on the product could be automatically sent, in real time, to the customer's email address, texted to his cell phone, or @customer'ed on Twitter. Another example is ad targeting, where cookies are evaluated in real time and a targeted ad is served in real time.

In these cases, real-time data enable competitive advantage, reduced cost, and maybe even increased profitable revenue. In these examples, real-time data is only useful to the business because of automation. At the end of the day, real-time data is rarely useful outside of automation. Business leaders need timely data and analytics, not real-time data.