Invest Employee Time Before Expecting ROI

Businesses that will benefit heavily from artificial intelligence must first make a capital investment–in the form of employee time. That may seem strange given that AI is supposed to be a time saver. But like all capital investments, money or time must be spent first before the gains are enjoyed. One companys AI strategy experience highlights the need to explicitly tell employees to spend some time exploring rather than producing.

With AI, employees must first find specific tools and then learn how to use them effectively. Then they will become much more productive, more than paying back their time investment. That is, if all goes well. Just as with a physical investment, a time investment may fail to produce useful results. But occasional failure of specific employees learning efforts is not failure of the program. A zero-failure standard would mean hardly anything is every tried. And AI being such a new technology, failure is pretty common according to several surveys. Huge success will also be common, justifying the inevitable failures.

Workday Inc.s experience offers a compelling lesson about how to invest employee time to develop AI-driven productivity. An internal survey found that 43% of its employees did not have time to learn about AI, with many also uncertain about how to use the tools.

Chatbots (such as ChatGPT, Clause, and Perplexity) are just the beginning of the learning process. They can be very useful in some cases—I use them regularly. But the largest business AI successes will come from highly specialized applications. The users may not know anything about AI. They may not even know that their application is using AI. All they will need to know is how to feed input into the system and how to use the output.

Work tasks vary widely from employee to employee. Even in a small company, there will be people doing widely varying jobs. And each employee likely performs different tasks through the workday. Telling workers to “ask ChatGPT” will not produce great results in most cases.

For some tasks, a chatbot may be very helpful—after some practice and learning. Often better results come from adjusting the “prompt” (the question or request posed to the chatbot). Some people will learn by trial and error, but others will be helped with some simple guidelines.

Workday found that peer-to-peer help was more useful than management exhortations. (And that is a generality that may well apply well beyond AI.) In some work environments, though, the most knowledgeable people may not be encouraged to spend time coaching other employees. First-level supervisors need guidance on how to balance the value of peer-to-peer coaching against the lost productivity of the coaches.

Beyond chatbots, though, are specialized applications that use AI to perform very specific tasks. As an example, Steve Browns Synthetic Newsletter this week highlights four AI tools for businesses. The tools support recruiting, receipt management, sales prospecting and assistants for websites, plus a general interest news app.

For example, someone in the human resources department may write position descriptions. That person may not know there are apps specifically for that task. And the HR manager may not know that, or may not know which ones work best in the companys specific situation. The employee should be provided with some background information about AI guidelines (more on that below) and then given rein to go browsing and trying different apps. It will seem that the employee has spent half a day without producing any position descriptions. But if a good tool is found, that half day of searching and testing will pay large dividends in the future.

Workdays effort included emphasis on some guidelines involving testing, risk assessment and documentation. An employee using an AI app should review some results to verify that output is correct. This may require some tedious double-checking.

Then high-impact results need human eyes before action is taken. If the AI seems good at writing up position descriptions, then maybe they go out without much review. But if the AI says that a particular customer is owed a million-dollar refund, the company probably wants a human to double-check before transferring funds. This is the risk assessment guideline.

And procedures should be documented, so everyone can verify exactly what is happening on the input and output of the app. Unfortunately, it will be difficult to know how the AI reasons. This area, called interpretability, is one of the most challenging fields within artificial intelligence.

The Workday experience provides good lessons for businesses trying to increase productivity through an AI strategy. The most fundamental lesson is that employees need to spend time before they can save time.