Bringing A.I. Tools to the Workplace Requires a Delicate Balance

By midyear, all of Morgan Stanley’s thousands of wealth advisers are expected to have access to a new artificial-intelligence-powered chat tool.

The tool, which is already in use by about 600 staff members, gives advisers answers to questions such as “Can you compare the investment cases for Apple, IBM and Microsoft?” and follow-ups such as “What are the risks of each of them?” An adviser can ask what to do if a client has a potentially valuable painting — and the knowledge tool might provide a list of steps to follow, along with the name of an internal expert who can help.

“What we’re trying to do is make every client or every financial adviser as smart as the most knowledgeable expert on any given topic in real time,” said Jeff McMillan, the head of analytics, data and innovation for Morgan Stanley Wealth Management.

Experts disagree about whether A.I. will wind up destroying more jobs than it creates over time. But it is clear that A.I. will alter work for most knowledge workers, shifting the skills they need and changing the staffing needs of most companies. Now it’s up to business leaders to figure out how to take advantage of the technologies today, while preparing workers for the disruption that the tools present over the medium term.

Moving too slowly may mean losing out on gains in productivity, customer service and — ultimately — competitiveness, similar to what happened to businesses that didn’t embrace the internet fully or fast enough. But at the same time, leaders must guard against the mistakes and biases A.I. often perpetuates and be thoughtful about what it means for employees.

“Almost no matter which sector you are in, you need to be thinking about your company as becoming an A.I.-first company,” said Alexandra Mousavizadeh, chief executive at Evident, a start-up that analyzes finance companies’ A.I. capabilities.

The type of A.I. underlying Morgan Stanley’s tool for advisers is called generative A.I. It can create content — including text, images, audio and video — from information it has analyzed. In addition to answering questions, it can be used in countless other ways, such as drafting memos and emails, creating presentation slides and summarizing long documents. Early research suggests that tools built using generative A.I. could speed up many tasks and increase employee productivity.

Massachusetts Institute of Technology and Stanford researchers, for example, found that customer support staffs equipped with an A.I. tool that suggested responses resolved 14 percent more customer issues each hour on average.

But the gains were not evenly spread. Less-experienced workers made greater productivity jumps, because the tools effectively “captured and disseminated” the practices of their higher-skilled colleagues. Other recent M.I.T. research similarly noted that workers who weren’t initially as good at tasks managed to narrow the gap with those who were more skilled, performing better and taking less time when aided by A.I.

One possible conclusion from these findings is “that the advantage that someone had from tenure in terms of their performance has now diminished because a youngster with ChatGPT can perform as well as somebody who’s had a few years’ experience,” said Azeem Azhar, chairman of Exponential View, a research group. If the research plays out in broader practice, that could potentially lead some companies to invest more in junior staff members, while going lighter on more expensive workers who have been around longer.

Some companies are already starting to make staffing decisions based on the anticipated impact of A.I. tools. IBM recently said it was slowing or stopping hiring for some back-office roles, such as human-resources functions, that could be replaced by A.I. over the next several years.

The speed and productivity gains from A.I. will raise customer expectations, said Bivek Sharma, the chief technology officer for PwC Global Tax and Legal Services. “It’s then about making sure we can re-skill the work force quickly enough and A.I.-enable them quickly enough to meet the obvious demand that’s going to come on the back of it,” he said.

PwC is working with Harvey, an A.I. start-up creating tools for lawyers, to roll out a chat A.I. tool to its entire legal advisory practice over the next few months. It plans to extend such technology to its tax and human resources experts as well.

Beyond quickly providing staff members with answers that draw on the firm’s expertise, PwC’s goal is to generate new insights, including eventually by analyzing its clients’ data as well, Mr. Sharma said. The A.I. could potentially be fed all of the contracts of two companies contemplating a merger, for example, and allow PwC experts to query for specific types of provisions and risks.

“Think of this as really an augmentation play rather than a timesaving play for us,” Mr. Sharma said. “This is almost like a senior associate that is attached to every one of our legal and tax advisers augmenting what they can do day to day for their clients.”

Larger companies generally need to invest in A.I.-savvy technical staff members, who can adapt the technology for their business. Already, “there are companies that can’t adopt ChatGPT because they simply don’t have the sort of basic rails upon which to run it on, which is content management and the data in order,” Ms. Mousavizadeh said.

They also need to hire or train new specialists, for roles that don’t necessarily require technical expertise. Morgan Stanley’s Mr. McMillan and other corporate executives say the A.I. platforms require continuous “tuning,” with humans adjusting parameters and information sources to get the best results for users. This tuning has created a need for a new group of workers known as “prompt engineers” or “knowledge engineers.”

Morgan Stanley and PwC are among those building their own versions of A.I. chat tools that draw from internal materials.

Concerns about security, confidentiality, accuracy and intellectual property rights have led many companies to restrict their staffs’ access to public ChatGPT and other generative A.I. tools. They want to avoid what reportedly happened at Samsung, where employees working in its semiconductor division are said to have shared confidential computer code and meeting notes while using ChatGPT. Executives are also concerned about the frequent errors and built-in biases with some A.I. tools.

But part of the opportunity with tools that use generative A.I., which allow users to type questions or commands in normal language, is to include a broader group of nontechnical staff members in figuring out how it can change a company’s business. “Your people should be using these tools really, really regularly so they can start to build up their competencies and your own internal firm competencies,” Mr. Azhar said.

He suggests that A.I. public tools can be used in ways that don’t endanger confidentiality or security. For example, an employee could ask ChatGPT about the best ways to combine types of sales data to tell a compelling story without actually entering the data itself. The opportunity, he says, comes from “front-line employees of whatever seniority deciding to improve their work through generative tools.”

Kevin J. Delaney is co-founder and editor in chief of Charter, a media and research company focused on the future of work.