AI micro-lessons are indeed effective, but will employees really "do" what they have learned from them?
Generative AI has significantly reduced the cost for enterprises to develop courses and tests, but for capabilities such as sales communication and customer service that ultimately require front-line employees to demonstrate through behavioral changes, what enterprises find difficult to provide in the long term is the continuous practice and feedback process after the course ends. In the past, this was largely a cost issue. Now, generative AI is changing the cost structure, making long-term large-scale practice possible.
After AI Makes "Teaching" Easier
Another Problem Has Been Exposed
In October 2025, the author, as the head of human resources, led the launch of a micro-course on conflict resolution for the sales team at an industrial product enterprise. After the course ended, employees showed significant improvements in pre- and post-tests of knowledge mastery and attitude, indicating that they did acquire knowledge through this AI-developed course. According to common training evaluation methods, the person in charge could already write a decent project summary at this point: the course is completed, employees have mastered the skills, and the data proves that the training is effective.
However, compared with the improvement of training data, what enterprises actually need more is that sales personnel can adopt a learned and more effective way to understand customer emotions and guide the conversation to an ideal result when facing customer rejection, price questioning, product complaints, or even emotionally agitated scenarios in real work situations.
In other words, the real problem lies in: Is it the same thing that employees have learned the knowledge and that they can actually perform well in practice?
Sales personnel can easily understand that "do not rush to defend when facing customer objections", and supervisors can also easily select the correct answer of "listen to employees' opinions first before providing feedback" in training. But the most strenuous process between "receiving training" and "performing better" — trying, finding it does not work, getting feedback, and trying again — is left for employees to solve on their own.
Enterprises also understand the importance of practice, but the problem is that practice has always been expensive. Assuming that 100 sales personnel need to practice customer conversations 5 times per person per week, that adds up to 500 practice sessions per week. Trainers can hardly provide enough practice opportunities; even if supervisors are willing, it is difficult for them to provide long-term assistance; it is even more impossible for enterprises to take real customers as practice partners for employees to make repeated mistakes. Therefore, the competence training of many enterprises will eventually choose a more achievable path: taking classes — taking exams — returning to posts, and the training ends here.
AI Makes Practice Possible
Since the problem occurs in the behavior transformation stage after learning, there is no need to arrange more training. Instead, practice opportunities should be provided. Therefore, after the micro-course ended, the author carried out a follow-up experiment and equipped employees with an AI situational coach. It will first imitate real customers to give negative feedback and hinder communication, let employees organize their language to respond, and then play the role of a coach to make comments based on the methods taught in the course, so that employees can accumulate experience after quick review.
Different from practicing through general chatbots, the AI situational coach will let employees go through the complete process of "situational response — feedback — adjustment" based on post scenarios, training objectives and evaluation criteria, rather than just giving simple answers.
During the experimental phase of more than one month that followed, 117 sales personnel in this company completed a total of 918 practice sessions, with an average of about 8 sessions per person. At the same time, we also asked supervisors to pay attention to employees' practice status and their performance in real visits. Of course, the subsequent behavioral changes of employees cannot be simply and entirely attributed to AI, but one thing is clear: compared with simply completing the course, this targeted practice support can indeed make the trained sales personnel perform better in terms of psychology and skills.
What Has Really Changed Is Not Just the Entire "Competence"
To further determine the actual effect of practice, we tracked the subsequent work performance of employees, set up another group of sales personnel who did not receive intervention during the same period as the control group, and identified two target behaviors to be directly trained in the project design phase: First, continue to try to solve problems when facing potential conflicts or disagreements; Second, better handle the other party's emotions when facing customer complaints or deliberately difficult situations.
Among the employees who had complete supervisor scores in both the two evaluations in June and December 2025, the two intervention target behaviors in the intervention group increased by an average of about 0.40 points, while the control group increased by about 0.05 points. The net change between the two groups was about 0.36 points, which reached statistical significance on a 5-point scale. However, if the other two behaviors that were not directly trained under the "conflict resolution" dimension are averaged in at the same time, the difference between the two groups will be significantly reduced and no longer reach statistical significance.
This result is more worthy of discussion than "how many percentage points the overall conflict resolution competence has improved" — in this project, employees did not suddenly gain better "conflict resolution competence" just by finishing a course. What first showed clear changes were the specific behaviors that had been practiced repeatedly and finally needed to be demonstrated at work.
When reviewing the practice data, the study also found that high-activity employees in practice had significantly higher improvements in both two intervention target behaviors than low-activity employees: the first behavior was about 0.54 points versus 0, and the second was about 0.51 points versus 0. Part of the reason is that people who are willing to take the initiative to practice have stronger internal drive, and supervisor support, customer opportunities and personal experience will also have an impact. But this set of data reminds us that between "acquiring knowledge" and "performing at work", enterprises should also pay attention to the issue of employees' enthusiasm.
The Story at the Performance Level Is Not That Satisfying Yet
After that, we compared the performance levels of employees in the intervention group and the control group before and after. The intervention group improved slightly, and the control group decreased slightly, but the net change between the two groups did not reach statistical significance.
At this stage, enterprises can easily draw conclusions in two directions prematurely: when the post-test scores improve, they announce that "the training is successful"; when the short-term performance does not improve significantly, they judge that "the training is useless". With AI complementing the key process of "practice" and related data, making it an independent analysis item, managers can also try to rethink the training evaluation method from the following four aspects.
- Acquisition: Have employees learned the knowledge?
- Practice: Have employees practiced?
- Behavior: Have employees performed differently after returning to work?
- Performance: How will these behavioral changes ultimately affect business results?
From these perspectives, behavioral changes brought by practice can indeed improve employee performance, but the long-term impact of behavioral changes on business results still requires longer observation and verification.
After AI Makes Training Cheaper
Where Should the Next Dollar Be Spent
Many enterprises are increasing their investment in AI skills, leadership, sales capabilities and employee retraining. No matter what measures are taken, the management ultimately needs to answer a question: whether people in the organization have really performed better because of these measures?
In the past, content was expensive, so enterprises spent a large amount of budget on the courses themselves. But as AI makes large-scale targeted practice support possible, the next budget for competence building should not only be used to continue adding content, but also to increase investment in employee practice.
AI cannot replace real work, nor can it replace supervisors' judgment and coaching in complex situations, but it can allow employees to turn their expressions into "muscle memory" in a psychologically safe environment before they really enter high-pressure scenarios, so as to get fully prepared before actual combat.
For enterprises, if they are promoting a sales, service or operational competence training, they may wish to make a small attempt first, select a work scenario that occurs most frequently and where they most want to see employees' behavioral changes to provide training. After the course ends, do not rush to move on to the next course, but provide employees with opportunities to try repeatedly, get feedback, and start over.
If this step can be achieved, the value of AI in improving competence and performance will be brought into play at a faster speed.
This article is from the WeChat Official Account "Harvard Business Review" (ID: hbrchinese), Author: Geng Yi, Editor: Sun Yan, published with authorization from 36Kr.