Paralegals used an AI tool 58% more when managers backed a better-job pitch with time and training

In the first division of the law firm, the manager gave paralegals LawBot with a speed brief: draft documents faster and shorten turnaround times ^[7]. The paralegals used it for tasks they already had. They said they felt like "just another cog in the machine" ^[7].

The speed brief is common, in the view of Arvind Karunakaran, the Stanford assistant professor who studied the deployments with Roshni Raveendhran of the University of Virginia and Tami Kim of Dartmouth College ^[2]. "When AI tools are introduced, managers are often obsessed with productivity gains," he said. "But if you aren't also working to answer how people's jobs are going to improve or be enriched, the initiative will most likely backfire." ^[16] Most organizations already use the tools. The share reporting AI use rose from 55% in 2023 to 88% two years later, the Stanford account says ^[1].

"I'm hesitant to learn new skills related to my job," one interviewee said, "because it feels like whatever I learn today will be outdated by next month." ^[5]

The second division got a different question and a budget. Its manager asked paralegals which tasks they found most boring and what they had always wanted to do but never had time for ^[8]. Managers then added formal training, weekly knowledge-sharing lunches, Q&A sessions with the vendor, informal mentorship and time set aside each week to explore ^[11]. In the Stanford account, all of that support sits in the second division. So the 58% usage gap between the divisions ^[9] measures the question and the budget together, and the study as summarized cannot split it between them. The researchers attach the condition themselves: enrichment framing motivated workers "assuming they were given the time and opportunity to explore" ^[3].

In our view, the usage count is the weakest number in the result. More use says the tool was opened more often. The summary does not report document quality, error rates or hours saved. The stronger evidence is the kind of work that moved: paralegals took a drafting tool into legal research and case analysis, traded tips, and lobbied to attend legal strategy meetings with the attorneys ^[15]. A change in task type is usage depth, and we would track it before any login count.

That new work runs into the review cycle. "But that was not the right metric for the new type of work they were doing with AI," Karunakaran said of case volume ^[14]. A paralegal who spends the protected weekly hours on research closes fewer cases in those hours, and a volume measure will score that as a decline.

A rollout can be placed on a grid with two axes: the pitch, speed or better work, and whether the pitch is funded with protected time, training and a changed metric. The law firm showed two cells. A speed pitch with no described support got routine use ^[7]. A better-work pitch with funding got more use and more experimentation ^[9][10]. Karunakaran predicts the third cell, a better-work pitch without funding. "You need to put this scaffolding in place to make the commitment feel credible," he said. "Without that, employees recognize it as little more than cheap talk." ^[12] The fourth cell, a funded speed pitch, is open. A funded rollout plan includes one line naming the weekly hours protected and the metric retired. A plan that leaves that line blank sits in the cheap-talk cell regardless of what the launch email says.