The IT productivity paradox apparently does not show any signs of going away. Some of the scientists who originally coined the term have just come up with more research. This new research is now linked to AI and shows that the deceleration of productivity growth has become even worse that it was twenty-five years ago when the discussion started. The average annual productivity growth in the US was at 2,8 % in 1995 – 2004 and it dropped to just 1,3 % in 2005 – 2016.

Interestingly, most of the presented explanations are still very much the same as they were in 1993 when the same author published his first famous paper on this subject.

The core of the paradox twenty-five years ago was “mismeasurement”, which meant that the productivity actually grows but we somehow do not see it. All right, we are now in 2018, so you would think twenty-five years were enough to figure out how to measure that invisible growth? Apparently the numbers are even worse now than then, so there is no invisible growth.  At that time, twenty-five years back, the second possible explanation was seen in “implementation lags”. Now, as the “mismeasurement” theory does not hold anymore, this implementation lag becomes “the biggest contributor to the paradox”.

But this also does not stand. Twenty-five years should have been enough to start seeing at least minimal growth even given all the lags, and the only trend we see is continuing deceleration.

So, I would have a much simpler explanation for the paradox. If productivity keeps decelerating during the twenty-five years of IT implementation across all trades, shall we conclude that it is this exact factor, information technology, that causes of the deceleration?

Here are two specific examples why and how this happens.

Let us look at AI and its flagship application, which is customer service through chatbots. There is analysis​ of cost per call going down from $35 - $50 on the phone to $8 - $10 in a web chat. Looks good in the first place, but note a very revealing phrase in the same paragraph about the increased handle time of the web chat. This is not just the agent’s time; this is also customer’s time. As it increases, the customer loses more time with the service and cannot dedicate this time to own business. Among the customers of Hypersoft, we have seen organizations where employees spend on average 15 or even 30 minutes in electronic interactions with automated service and support applications. This is as much as six percent of working time lost. In fact, some vendors of chatbot software even consult their customers​ how to increase the average session length! The balance is obvious: whatever is saved on the call center agent, is lost by the customer having to spend their expensive time dealing with the bots and automated response systems.

Another example is the incredible easiness of pulling people into virtual meetings with the modern conferencing software. I know one customer where the average count of meeting participants went from 3 – 5 to 12 – 14 after they made a virtual conferencing system available to all their employees. How does it help that those 14 persons did not incur travel expenses, and is it possible to believe that an average meeting in a large company needs 14 participants? I guess the answer is obvious here.

There are many more similar situations, which we encounter daily with our Productivity Analytics. This is not proof of digital technology being always counterproductive by its nature. However, this is proof that the technology does generate substantial loss of time by keeping people busy with it, instead of doing their work. And it is not just training people how to use it right, it is the very nature of many modern productivity tools to replace work by “work-related communication”.

We plan to continue this research. There is a lot of discussion going on at business about “technology adoption”, which is somehow presumed to always be a good thing. But we need to identify all those cases when implementation of digital technology results in an increase of the total time that all involved persons need to produce the same output. We can do this quite well already with Omnicontext, and we are improving these capabilities every day. Stay tuned, there will be more insight coming from Hypersoft on this topic.