Why Isn't My Team Using the AI Tools We Paid For?

A team not using AI tools the company paid for is responding to conditions at work. Use is more likely when the tool helps with their own tasks, is easy to learn, is expected by people they work with, and comes with access and help. Training teaches how the product works, and those conditions are set after it ends.

The licenses are paid for, the team went through training, and the pilot group's results were good. The usage report shows a few people opening the tool while most of the team does the work the way they did before. Your sponsor has asked you why.

Why Aren't Employees Using the AI Tools You Bought?

Viswanath Venkatesh, Michael Morris, Gordon Davis and Fred Davis compared eight earlier models of why people at work take up a new technology or leave it alone, and combined them into one (Venkatesh, Morris, Davis and Davis, MIS Quarterly, 2003). They tested it in four organizations, with measurements right after training, one month after the system was put in place, and three months after. Then they tested it again in two more organizations.

Their model has four conditions. Three of them predicted whether people intended to use a new system. One is whether a person believes the system will help them do their job better. Another is how easy the system is to use. The third is whether the people who matter to them at work think they should use it.

Intention to use the system predicted actual use. The fourth condition, whether the organization and its technical setup support using the system, also predicted use for older workers who had been using it for a while. How much each condition mattered depended on people's age, gender and experience, and on whether use was required.

Why Doesn't AI Training Change Whether People Use the Tool?

Training teaches how the product works. Whether people use it afterward depends on conditions in their workplace once training ends.

Researchers study this question as transfer of training. Timothy Baldwin and J. Kevin Ford described two parts to transfer (Baldwin and Ford, Personnel Psychology, 1988). People use what they learned in their work, and they keep using it over time. Baldwin and Ford reviewed research on three sets of factors that affect transfer. These are the trainees, the design of the training, and the work environment people go back to.

Brian Blume, J. Kevin Ford, Timothy Baldwin and Jason Huang later combined the results of many transfer studies (Blume, Ford, Baldwin and Huang, Journal of Management, 2010). Transfer was higher among people with more cognitive ability, conscientiousness and motivation, and among people whose work environment supported it. Most of these links were stronger when the training taught open skills, such as leadership, than when it taught closed skills, such as using computer software.

Learning the features of an AI tool is close to a closed skill. Deciding which of your own tasks to give it, and checking what it gives back, is closer to an open skill, the kind where Blume and colleagues found most of these links were stronger.

A training session can show people the features. The conditions that affect whether they use the tool afterward include time to try it on real work, a manager who asks about it, and someone to go to when the output is wrong.

Why Did the AI Pilot Go Well and the Rollout Stall?

A pilot tests whether the tool can do the work for the people and tasks in the pilot. The rest of the team uses it under their own conditions, and those can differ. Compare the pilot group with the wider team on these questions.

• Did the pilot group volunteer?

• Did they use the tool on the same tasks the wider team does?

• Did they have time set aside to learn it?

• Who did they go to with problems, and can the wider team go to the same person?

• Did their managers ask them about the tool?

Marcie Tyre and Wanda Orlikowski studied how people adapted new process technology in three manufacturing and service organizations (Tyre and Orlikowski, Organization Science, 1994). They found a fairly short period after a new technology was put in place when users explored it and changed it. After that, use became more routine with experience, and users changed the technology less. Problems that hadn't been solved by then often became part of everyday practice. Later changes came in episodes, triggered by unexpected events or by things users discovered.

If the pilot group had time to explore the tool and the wider team got it in a busy month, the team's routine may have formed around the old way of working before anyone tried the tool on real work. Tyre and Orlikowski's finding about later changes suggests a stalled rollout can start again around a specific event, such as a new piece of work or a problem the old process can't handle.

Does the AI Tool Help with the Work Your Team Does?

Start with this question. In the models Venkatesh and colleagues compared, belief that a system would help with the job was the strongest predictor of intention to use it. It predicted intention at every point they measured, whether use was voluntary or required. Fred Davis found a similar pattern in two earlier studies of people using computer programs (Davis, MIS Quarterly, 1989). How useful people rated a program was more strongly linked to their reported or expected use than how easy they rated it.

With AI tools, whether the tool helps depends on the task. Fabrizio Dell'Acqua and colleagues ran an experiment with consultants at Boston Consulting Group, some working with GPT-4 and some without it (Dell'Acqua et al., Organization Science, 2026). On consulting tasks within what the AI could do, consultants using it finished more tasks, worked faster and turned in higher-quality work. On a task the researchers chose because it fell outside what the AI could do, consultants using it were less likely to reach the correct answer. The authors wrote that the boundary can be hard to see, and that tasks that seem similar in difficulty can fall on either side of it.

Someone who tried the tool on a task it handles badly and got a worse result has a sound reason to stop using it. Ask the people who stopped which tasks they tried it on and what happened. If the tool doesn't help with the tasks your team does, low use is an accurate response. The next step is to find tasks where it does help, or to tell whoever chose the tool which tasks it doesn't fit.

Do Employees Trust the AI Tool's Output?

What people have seen a tool get wrong can change whether they keep using it. Berkeley Dietvorst, Joseph Simmons and Cade Massey ran five studies in which people chose whether their pay would depend on a statistical model's forecasts or on a person's, usually their own (Dietvorst, Simmons and Massey, Journal of Experimental Psychology: General, 2015). People who had watched the model make forecasts were less confident in it and less likely to choose it, even when they had seen it outperform a person. After seeing the model and a person make the same mistake, people lost confidence in the model faster. The participants were volunteers in a university lab and an online panel, forecasting MBA applicants' later success and ranking US states by airline passengers. The study didn't test AI tools at work.

In six experiments by Jennifer Logg, Julia Minson and Don Moore, people followed advice more when they were told it came from an algorithm than when they were told it came from a person (Logg, Minson and Moore, Organizational Behavior and Human Decision Processes, 2019). The effect was weaker when people chose between the algorithm's estimate and their own, and when they had forecasting expertise. Professionals who forecast regularly relied on the algorithm's advice less than other participants did, and were less accurate as a result. Both conditions fit someone using an AI tool on their own work, though Logg's studies used forecasting tasks.

In later studies, Dietvorst, Simmons and Massey found people were more likely to use an imperfect algorithm's forecasts when they could adjust them, even slightly (Dietvorst, Simmons and Massey, Management Science, 2018). People who could adjust the forecasts performed better, were more satisfied, and were more likely to use the algorithm again.

People can edit an AI tool's output before they use it. Ask the team to treat the output as a draft to check and change. Ask them to report wrong answers to a named person, who keeps a list of the task and what went wrong. The list shows which tasks need the most checking and which ones the tool doesn't handle well. Share it with the team and with whoever chose the tool.

Do People Have the Time and Help to Learn the AI Tool?

In the models Venkatesh and colleagues compared, how easy a system was to use predicted whether people intended to use it right after training. It stopped predicting intention once they had used the system for a while.

The fourth condition in their model covers the setup around the tool. The questions they used for it asked whether people had the resources and knowledge to use the system, whether it worked with the other systems they used, and whether a specific person or group was available to help with problems. They built this condition partly from an earlier measure of whether using a system fits the way a person works.

For each team, find out the answers to these questions.

• Have people had time set aside to try the tool on their own work?

• Who do they go to when it gives a wrong or unusable answer?

• Does it work with the systems they already use?

• Can they use it where they already do the task, or do they have to copy work into a separate window and back?

• Do they know what they're allowed to put into it, such as client information?

Do Managers Expect Their Teams to Use the AI Tool?

In the models Venkatesh and colleagues compared, what the people who mattered to someone at work expected affected that person's intention to use a system only when use was required. In their combined model, this pressure mattered most early in people's experience with the system, and more for women, older women in particular. Their questions on it included whether senior management had been helpful in using the system and whether the organization had supported it.

Ask each team's manager what they've told their people about the tool. Ask which tasks they expect it to be used for, and whether they bring it up in one-on-ones or team meetings. If nobody has told the team anything, ask the manager to tell them which tasks the tool is for.

Are Employees Hiding Their AI Use or Using Other AI Tools?

Jessica Reif, Richard Larrick and Jack Soll ran four online experiments on how people who use AI at work are judged (Reif, Larrick and Soll, Proceedings of the National Academy of Sciences, 2025). People who imagined using a generative AI tool to produce reports at work expected others to see them as lazier, less competent, less diligent and easier to replace than people who imagined using a dashboard tool for the same reports. They also said they'd be less likely to tell their manager and colleagues about it.

In a second experiment, observers rated employees described as getting help from AI as lazier, less competent and less diligent than employees who got similar help from another source, or no help at all. In two hiring studies, participants who used AI less often themselves judged AI users more harshly, and frequent AI users didn't. When AI was described as useful for the task being hired for, AI users weren't rated as a worse fit for it. The authors note that their participants were online volunteers responding to written scenarios, and that whether the same judgments happen in established organizations is still to be tested.

In thirteen experiments by Oliver Schilke and Martin Reimann, people who disclosed that they'd used AI for a piece of work were trusted less than people who didn't mention it (Schilke and Reimann, Organizational Behavior and Human Decision Processes, 2025). In one of them, a tax advisor whose AI use was revealed by someone else was trusted less than one who disclosed it himself. Most of their studies asked about trust in people the participants didn't know.

If people expect to be judged for using AI, asking who uses it in a team meeting may undercount the people who do. When you name the tasks the tool is for, say that you expect people to use it on them and to mention when they have. Ask the managers who review the team's work to try the tool on their own tasks. Reif and colleagues' finding about frequent AI users is a correlation, and in a supplementary experiment by Schilke and Reimann, writing a letter with AI first didn't shrink the drop in trust when a job applicant disclosed using AI.

Before you count someone as a non-user, check whether they're doing the same tasks with a different AI tool. If they are, ask what it does for them that the company's tool doesn't, and check it against your organization's rules on what information can go into AI tools.

How Do I Measure AI Adoption beyond Licenses and Log-Ins?

A license count shows who has the tool, and log-ins show who opened it. Neither shows which tasks it was used on or what it changed about the work.

Andrew Burton-Jones and Detmar Straub described use of a system as having three parts, the person using it, the system and the task (Burton-Jones and Straub, Information Systems Research, 2006). They argued that researchers should pick measures of use that fit what they're studying. In their test, which looked at how use relates to task performance, a poorly chosen measure of use could lead to the opposite conclusion.

Detmar Straub, Moez Limayem and Elena Karahanna-Evaristo compared what people reported about their use of a system with what the computer recorded, and found the two weren't strongly related (Straub, Limayem and Karahanna-Evaristo, Management Science, 1995). They concluded that self-reported use and recorded use should be treated as separate measures.

To find out whether your team uses the tool, measure use on the tasks it's meant for.

• List the tasks the tool is meant for on each team.

• Take a few recent pieces of work for each task. Ask the person who did each one whether they used the tool, and check the tool's records where your organization allows it.

• For each piece done without the tool, note how it was done, such as the old way, a workaround like a spreadsheet, or a different AI tool.

• For each piece done with the tool, ask what it changed, such as time saved or fewer corrections.

Where people's answers and the tool's records disagree, ask the team about the difference.

Why Won't Some Employees Use AI Tools at All?

When someone avoids the tool on every task, ask them why before you decide how to respond.

Hee-Woong Kim and Atreyi Kankanhalli studied employees' resistance to a new enterprise system before it was put in place (Kim and Kankanhalli, MIS Quarterly, 2009). The costs people expected from switching to the new system were central to their resistance. Colleagues' opinions, and how confident people felt about making the change, affected resistance through those expected costs. The value people saw in the new system, and the organization's support for the change, reduced resistance.

Sort the reason the person gives, and answer that reason.

• Fear about jobs. Tell them what's been decided about their role, what hasn't, who decides, and when they'll hear more. Promise only what you control, such as the scope and dates of your rollout.

• Workload and learning time. Ask how long the task takes the old way and how long it took with the tool. If the old way is still quicker, ask their manager which of their other work can wait while they practice with the tool.

• Training and rules. Show them how to start the tool on one of their own tasks, and tell them which tasks it's approved for and what they can and can't put into it.

• The tool. Find out which task it was, and try the tool with them on a few recent pieces of that work. If it does worse than they do, stop asking them to use it on that task, and tell whoever chose the tool.

• Work they value. Ask which parts of the task they want to keep doing themselves, and check whether they can review and change what the tool produces before it's used.

How Do I Get Employees to Use AI Tools?

Start with one team and one task the tool handles well. Pick a task the team does often, and try the tool on several real examples of it before you ask anyone to use it. Compare the results with the team's own work.

When you introduce the tool to that team, tell them which task it's for, what it got wrong on the examples you tried, and how to check its output on that task.

Amy Edmondson, Richard Bohmer and Gary Pisano studied hospitals putting in place a new technology for cardiac surgery that required substantial changes in how the operating team worked (Edmondson, Bohmer and Pisano, Administrative Science Quarterly, 2001). Every hospital had a highly regarded cardiac surgery department, and they differed sharply in how far they got with the new technology. The teams that succeeded went through a different team learning process. They held practice sessions and early trials designed to make it safe to try new ways of working, and they reviewed together how the work was going and improved the process.

For an AI tool, that can be a session where the team uses the tool on a real piece of their own work, a few weeks of using it on that one task, and a short meeting afterward about what worked and what didn't. To get people using a new tool on a team that doesn't report to you, ask their manager to set aside time for the practice session. During those weeks, check whether the task is still being done the old way. If it is, ask the people doing it what got in the way.

This week, choose the team and the task, and schedule the practice session with that team.

Frequently Asked Questions

Will more AI training get my team to use the tool?

Not by itself. Training teaches how the product works, and use depends on conditions in the workplace after training ends. Baldwin and Ford grouped what affects whether training gets used on the job into three sets of factors, the trainees, the training design and the work environment. Changing the training works on the second set. Time to practice on real tasks, a manager who expects use and someone to ask for help are part of the third.

Should we make using the AI tool mandatory?

If you require it, limit the requirement to tasks the tool handles well. In the models Venkatesh and colleagues compared, what the people who mattered to someone expected affected their intention to use a system only when use was required. Belief that the system would help with the job predicted intention at every point they measured, in required and voluntary settings.

Should my team use the AI tool for every task?

No. In the Dell'Acqua study, consultants using AI did better on tasks within its abilities and were less likely to reach the correct answer on a task outside them. The researchers wrote that the boundary between the two can be hard to see. Work out with the team which tasks the tool helps with, and track use on those tasks. Low use on tasks it handles badly is the result you want.

Who should own whether people use the AI tool after rollout?

Someone close enough to the daily work to see whether a task is still being done the old way. That can be the person running the rollout or a lead inside the team. IT can own access and setup. Whether people use the tool on their tasks needs someone who talks with those people regularly and can take problems to their manager or the sponsor.

How do I find out why one person stopped using the AI tool?

Ask them which tasks they tried it on and what happened. Then ask what would have to change for them to use it on one of those tasks next week. Their answers will point to the task, a wrong answer the tool gave them, the effort of learning it, what their manager expects, the help available, or something else, such as worry about their job. Respond to the reason they give.

If you're running a rollout and want a plan for each stage, The DANCE System gives you 29 tools across five phases, from defining the change to keeping it in place after launch.

Related Blog Posts:

How to Roll Out a Change at Work

How to Talk to Employees Who Fear AI Will Take Their Jobs

How to Introduce AI to Your Team

References

1. Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425-478. https://doi.org/10.2307/30036540

2. Baldwin, T. T., & Ford, J. K. (1988). Transfer of training: A review and directions for future research. Personnel Psychology, 41(1), 63-105. https://doi.org/10.1111/j.1744-6570.1988.tb00632.x

3. Blume, B. D., Ford, J. K., Baldwin, T. T., & Huang, J. L. (2010). Transfer of training: A meta-analytic review. Journal of Management, 36(4), 1065-1105. https://doi.org/10.1177/0149206309352880

4. Tyre, M. J., & Orlikowski, W. J. (1994). Windows of opportunity: Temporal patterns of technological adaptation in organizations. Organization Science, 5(1), 98-118. https://doi.org/10.1287/orsc.5.1.98

5. Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319-340. https://doi.org/10.2307/249008

6. Dell'Acqua, F., McFowland, E., III, Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science, 37(2), 403-423. https://doi.org/10.1287/orsc.2025.21838

7. Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114-126. https://doi.org/10.1037/xge0000033

8. Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90-103. https://doi.org/10.1016/j.obhdp.2018.12.005

9. Dietvorst, B. J., Simmons, J. P., & Massey, C. (2018). Overcoming algorithm aversion: People will use imperfect algorithms if they can (even slightly) modify them. Management Science, 64(3), 1155-1170. https://doi.org/10.1287/mnsc.2016.2643

10. Reif, J. A., Larrick, R. P., & Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences, 122(19), e2426766122. https://doi.org/10.1073/pnas.2426766122

11. Schilke, O., & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188, 104405. https://doi.org/10.1016/j.obhdp.2025.104405

12. Burton-Jones, A., & Straub, D. W. (2006). Reconceptualizing system usage: An approach and empirical test. Information Systems Research, 17(3), 228-246. https://doi.org/10.1287/isre.1060.0096

13. Straub, D., Limayem, M., & Karahanna-Evaristo, E. (1995). Measuring system usage: Implications for IS theory testing. Management Science, 41(8), 1328-1342. https://doi.org/10.1287/mnsc.41.8.1328

14. Kim, H.-W., & Kankanhalli, A. (2009). Investigating user resistance to information systems implementation: A status quo bias perspective. MIS Quarterly, 33(3), 567-582. https://doi.org/10.2307/20650309

15. Edmondson, A. C., Bohmer, R. M., & Pisano, G. P. (2001). Disrupted routines: Team learning and new technology implementation in hospitals. Administrative Science Quarterly, 46(4), 685-716. https://doi.org/10.2307/3094828