One of the generally unheralded trends in enterprise IT is the proliferation of automation tools. Dozens of vendors offer systems to automate various tasks and business processes, and countless companies have developed their own homegrown tools. Many of these tools have begun overlapping, or sprawling, over time. I’d like to have invented the term “automation sprawl,” but I have to credit Vince Dimascio, CIO and CTO at BAL, an immigration law firm based in San Francisco. He was one of the panelists at a session I facilitated recently on the future of process automation at the 2019 MIT Sloan CIO Symposium. It’s clear that one aspect of that future is more sprawl.

Before I get into how best to respond, it’s helpful to have a little background on how we got here. Of course, “automation” is an old term that has been used to describe virtually every type of computer system, from general ledgers to manufacturing robots. More recently, it’s been focused on workflow-related capabilities involving business rules. Some automation offerings are oriented to specific organizational functions, some purport to be generalized platforms, and some are in the middle. The earliest broad application of automation tools was in the IT function, where they are now well-established for tasks like network management, provisioning, and configuration management. Marketing automation was another relatively early use.

Some of the companies that did IT automation, like ServiceNow, have branched into products for other functions — like HR onboarding. IPsoft, initially an IT automation company, now has a product called Amelia to automate conversational interactions with customers and employees. IPsoft also has a new tool called 1Desk that “automates the automation” — it keeps track of automation programs from multiple vendors and monitors their performance.

The focus of automation tools was at one point to automate structured, predictable workflows typically within a specific domain, for instance in IT or marketing. But one automation tool, so-called robotic process automation (RPA), has become a generalized tool for executing structured workflows, particularly for processes that involve data from multiple information systems.

One indication of the expanding sprawl is that RPA and several other automation tools are now going after data-intensive decision tasks that are made within those structured workflows. They are either adding machine learning capabilities themselves or making it easy for customers to use other vendors for them. The RPA firm UIPath, for example, has a partnership with the automated machine learning vendor DataRobot. ServiceNow has built its own machine learning programs for making recommendations for IT users and for onboarding employees. It’s clear that everybody is moving into each other’s territory and rapidly adding new functionality — the classic definition of sprawl.

OK — now what should you do about it? One answer is to look for generalized automation tools that can perform a variety of task types. For many companies, that increasingly leads them toward RPA, often in combination with machine learning. Eventually, I expect, the common attributes of RPA — things like an easy-to-use graphical interface and a built-in rules engine — will characterize other types of automation as well.

At ServiceNow, an automation software vendor itself, the company’s chief information officer, Chris Bedi, is trying to put a small number of automation platforms in place to cover different use cases. For employee-facing automation, Bedi employs his own company’s technology. He’s also seeking a broad customer-facing automation platform, but hasn’t found a suitable one yet. He also uses a platform for process mining to understand and monitor process flows, and implements RPA primarily to easily move data across platforms without having to fully integrate them.

Another approach to managing sprawl is to create a classification system for different automation types and how you will use them. One global consumer products company I work with, for example, created an automation classification system that includes content recognition, decision making, task automation, and process monitoring and a set of technologies including RPA, image recognition, and AI to support each automation category.

That same company, and another large aerospace manufacturing firm I researched, took a further step to managing automation sprawl: creating an organizational unit to do so. The consumer products firm has formed a “smart automation” center of excellence, with a small group of staff members. The aerospace firm has an “operations transformation” group, which began by pursuing IT automation and is now moving into functions such as human resources. The medical technology company BD also has an automation center of excellence. While each of these is structured somewhat differently, what they have in common — what any company setting up such a unit should consider — is they help business units and functions within their organizations figure out what type of automation tool best fits their needs.

Finally, in managing sprawl it’s important to let employees know what the plan is for the technology. Smart automation raises obvious concerns about the future of human employment. None of the firms I’ve interviewed thus far have eliminated substantial numbers of employees, and none say they plan to. However, in a 2018 Deloitte survey I helped to design and analyze, 63% of the U.S. executives surveyed said they wanted to “cut costs by automating as many jobs as possible.” Many in that same survey indicated that they preferred to hire new workers rather than retraining existing ones. There are some exceptions — Amazon, for example, recently said it would spend $700 million to retrain 100,000 employees to perform new jobs made possible by AI and robots. On a smaller scale, Allstate announced recently that it would devote $40 million of its corporate tax reduction in 2018 to training workers to work with automation-related technologies.

But to head off the increasing replacement of jobs by smart automation, more companies will need to follow Amazon’s and Allstate’s lead. That’s not just about being a good employer — firms need employees’ help to implement automation technologies. In most cases, automation tools will perform discrete tasks, not entire jobs or business processes. Human workers will need to assist in adopting, monitoring, and improving automation technology. If they suspect that they are being targeted for automation-driven layoffs, they are unlikely to cooperate.

Automation technologies will surely continue to sprawl. They offer exciting possibilities for organizations to eliminate drudgery for human workers and improve the productivity of the enterprise. It is time for firms to establish the capabilities, the coordination bodies, and the policies needed to deal with these technologies, and to harness automation sprawl to their advantage.

*This article was written by Tom Davenport and published in Harvard Business Review on July 19, 2019.