The quality of the members that comprise an organisation is among one of the most important factors that can determine the output and survival of the organisation. To improve the quality of existing members, especially if replacement is undesirable, leaders of organisations often use various forms of training, such as directly organising workshops, seminars, online training, or perhaps encourage employers to enrol in job-related 3rd-party training by including it in performance review or providing subsidy. However, research in education has shown that such formal, top-down, one-to-many, short-term transfer of skills is often not a very efficient way of training. An alternative or complementary approach is to foster horizontal transfer of skills between current members. This approach ensures that most skills being learned are job-related and can be applied immediately, allowing reinforcement learning through positive feedback, and the learner often takes a more active role. Indeed, mentorship, journal club, and shadowing can be seen as systems that attempt to harness the power of such peer learning. It is however unclear how training compares to peer learning. It is possible that their comparative effectiveness depends on the type of skills being learned. Using agent-based models and data from a technology company, we compared the improvement of professional and teamwork skills through training and peer learning. Here, a professional skill is defined as something that allows a worker to directly contribute to job output, such as problem solving skills, technical knowledge, work efficiency, etc. Teamwork skills on the other hand are skills that do not contribute directly to work, but instead allow members of a team to cooperate more efficiently so that the whole is greater than the sum of its parts, with members of better professional skills contributing more to the final output of the team. We hypothesise that professional skill can diffuse from more skilled employees to less skilled employees at a constant rate over time as they interact, while teamwork skills directly improve as employees interact with each other on a social network. The effect of this peer learning is compared to the effect of employee training organised by the company. In contrast, participation in training events on various topics is assumed to provide an increase in the corresponding skills at a constant rate. We first used an agent-based model to simulate the whole process and generate pseudo-data to test the robustness of our statistical model, ensuring that the parameter values can be recovered, and then used the real data to calculate the efficiency of each learning. We found that training is more efficient for improving professional skills while peer learning is more efficient for improving teamwork skills.