Unleashing the Power of Employee Learning
Unbelievably, some companies still doubt the benefits of investing in employee learning. Indeed, as an employer, it is not an easy task to set up a good learning program for employees. Employers look at the Return on Investment. It might be very unclear how to monitor the financial impact of a learning strategy. Those ROI-driven companies might get the feeling that it is a waste of time and money simply because they cannot fill the “credit” column in their financial reports. One of the worst examples I witnessed was an email from management telling me that I should learn to develop my skill set, with a link to a (super expensive) e-learning platform attached. They did not include any direction regarding which topic to pick and how I should manage the balance between learning time and other tasks. In addition, even if the material provided covered a large range of topics, none of them went in sufficient depth. I felt like I had received partial instructions, with no clear “why”, “how”, and “when”.
Since then, I was particularly interested to find the appropriate way for data science teams to learn during their work hours. To that end, I tried many different learning frameworks and in this article, I would like to share the most efficient one I found to date. I use the terms “learning” here rather than training to highlight the fact that the objective is not to create new capabilities for employees. Acquiring new skills is a side effect and not the main goal. The main objective of learning is to increase the employees’ knowledge, and it doesn’t matter if it is directly linked to a business problem. I have seen many examples where a good learning strategy solved many pain points, beyond the team’s productivity. I am writing this blog to share the learning strategy I put in place and its outcomes. Please share your feedback! If you agree or disagree, let me know…
Why should we learn?
Learning is an important part of our daily work. It is essential for many reasons. Among them:
- Keeping our skillset up to date. Reusing constantly the same tech, in the same way, is the best way to be outpaced by competitors.
- As a company, we constantly need to have access to the latest and best technology. The alternative to learning is hiring new talents, which can be up to 6 times more expensive for a company compared to paying employees to learn. source
- Attracting and retaining talent. Learning is the second most important factor in workplace happiness (source).
- Learning is the most effective way to improve productivity. (source)
- Learning is fun!
In my opinion, learning is not a luxury. As data scientists, it is part of our job and should be taken as seriously as any other task. However, since learning is never a priority, people tend to postpone it… and never spend enough time to do it properly. This blog post attempts to define a framework for a research/applied AI team to learn efficiently.
What should we learn?
I had some team members come to me and propose working on music generation… Well… it might be a good idea but the person should first learn basic concepts of Deep Learning. It happened to me many times that employees wanted to aim for the moon without first grasping basic concepts in the field. But, this is great! It demonstrates enthusiasm and this should be rewarded.
Learn a topic
I make a distinction between field and topic. A field is broader, and cannot fully be comprehended in all its complexities. EX: Deep learning, Natural Language Processing (NLP), databases etc… A topic is a subsection of a field, for which one can define a learning success criterion. EX: transfer learning using music embedding, logging in python etc… It can be hard to determine which topic to learn; that is why I help my team members with their choice.
Who should pick topics?
The primary choice of topics should come from employees. Learning a new skill, a new tech or an open problem can only be done if the employee has a personal interest in it. However, propositions can come from anyone to maximize the diversity of topics and inspire everyone. The idea here is to collect topics based on interests. Once this list of topics (which is an ever-growing list) is compiled, the employee and the manager can decide to pick one of them together. This should be at the employee’s initiative. I do not believe that pushing employees to learn something would lead to anything positive. If the employee manifests a strong interest in a specific topic, chances are he will invest himself to a greater extent to succeed in his task. My role is to help him!
Two different employees can learn the same topic. In this case, the two employees can either team up or share resources. This is specifically relevant to core topics (EX: PyTorch, business vertical).
How should we learn?
Learning is a 3-step process:
- Acquire knowledge
- Do
- Share
Each step is important to complete a learning task.
Acquire knowledge: The learning method can vary according to personal preferences. Some people prefer to read a book at their own pace, others prefer to attend a course etc… Please do not forget that anything can be an opportunity to learn. EX: You attended a Town Hall meeting from the marketing team, this is the opportunity to learn about marketing, how to present information etc… The next step for you is to use this inspiration to do a better presentation.
Do: Even if the knowledge seems easy to capture, by doing, the employee assimilates it. “Do” means fail, fail, and finally… succeed. In my opinion, failing is an even better source of learning than succeeding. Keep in mind that this process might cause frustration, the “peak-end rule” might prevent you to reuse this knowledge and bias your judgment. When an employee retains a negative emotion associated with a task, they may form a negative bias towards that task, which can result in a reluctance to perform it again in the future.
The peak–end rule is a cognitive bias that impacts how people remember past events. Intense positive or negative moments (the “peaks”) and the final moments of an experience (the “end”) are heavily weighted in our mental calculus. Daniel Kahneman
Share: The last step is to share the knowledge acquired. The main reasons are:
- It will help the learner formulate clearly what he learned and force him to organize his thoughts and focus on the essentials.
- It will help other collaborators get the gist of the topic.
- In order for the presenter to get feedback.
Sharing can be done in many ways. Through a notebook (not ideal), in a “lunch and learn” session, a blog post etc… Since presentations are open to anyone in my company, it is exciting to present complex things to a broad audience. The presenter must narrow it down to the essentials and make it understandable to anyone. Some people are naturally gifted for that, others are not. I help people with this step if they do not feel comfortable doing it.
The process
The most crucial aspect of this article is the process. It is not necessarily about adhering to each specific step. What is important is to grasp the essential ideas behind the process. In this section, I will describe the following:
- How to define a learning task,
- When is the appropriate time for learning,
- Negotiating to select the appropriate learning task,
- The “learning board”, which I use as a tool to keep track.
Define a learning task
As a manager, I own the process of making sure my team members can learn. During one-on-ones, I encourage my team members to voice their interests, based on which I help them formulate a topic. The elements of a learning task are:
- The title
- The description of the topic
- An end date. This is particularly important because the person will have to commit on a certain date. There can be some flexibility, of course, but I ask for good arguments.
- A definition of done. EX: “a presentation of how to serve a model with MLFlow”
- The name of the person taking on this task
Once selected, a learning task should take a maximum of 3–4 weeks (not full-time). During this period, the person can get organized but the priority of the learning task is as high as any other. It means that during the daily stand-up, I should hear about it every day. If not, I raise a flag.
When is the ideal period to learn?
When working as a data scientist, there is a lot of time we can exploit for learning. Ideally, a learning task can take place between projects. Data Science projects take about 3 months; I allow 1 week between projects to learn full-time.
As a manager, learning is always good to fill all the small gaps that could occur during a project. It is not worth spending 10 minutes to learn something. When working on the learning task, the person should spend at least half a day on it. If a team member scheduled half a day for a learning task but received an email with answers for a project during that time, they do not need to halt the learning and move on to the project immediately. It can be deferred until the next day.
The negotiation
The goal of the negotiation is to prioritize learning tasks. For each team member, I review the proposed learning task and I try to prioritize based on both the personal interest and the needs of the team. This negotiation also includes an external budget, for instance, when attending a conference or to access dedicated computing resources (training a custom LLM is not free 😊 ). I do not discard any learning task during the negotiation, but I give feedback. For instance, if the learning task is music generation, it might make more sense to learn first how to train VAEs or GANs, and then in a second learning task, how to adapt it to music generation.
Learning Board
To get organized, I set up a learning board. The goal of this board is to centralize and monitor progress on the different learning initiatives. In addition, if we see a name attached to a specific topic, others can ping the person since he just became our in-house expert (congrats 🙂 ).
The columns of the board are:
- Backlog for later: This is a list of learning tasks that seem exciting but need to be refined and negotiated.
- Propositions of learning: A list of tasks which individuals have committed to working on at some point. Those tasks are not started yet and are ordered in terms of priority. The end date remains empty for now.
- Learning: the list of tasks team members are working on. The end date is filled.
- Done: a list of completed tasks. This includes a link to the token of completion, such as the presentation prepared on the topic.
The outcomes
Here is a short list of outcomes I witnessed in the past few years:
- Fun. Talented people love to learn. It gets frustrating for them when learning conflicts with their daily tasks. Allowing individuals to take a step back and do something they want to do is fun.
- Retention. An immediate consequence of the first outcome is employee retention. In companies where I had the opportunity to put in place a good learning structure, no employee ever left my team.
- Productivity. Productivity can be enhanced by learning in two ways. Firstly, creating a GAN for the first time in the context of a project can be challenging. Dedicating time to learning about GANs and their challenges in a learning task helps reduce significantly the development time, even if the code isn’t reused. Secondly, taking a break and looking at a project from a different perspective, by “zooming in, zooming out”, can also improve productivity.
- Cross-pollination of ideas: Ideas can be reused across different projects. For instance, spending time learning recommender systems can be helpful to frame a new project that would match this paradigm.
As an example, in 2021, the research team acquired a lot of knowledge on many different topics. Even if those topics were not directly linked to future projects, it has been hugely beneficial to the research team. We managed to create state-of-the-art algorithms for some tasks in a very short period by reusing techniques or knowledge seen in related topics.
Conclusion
After reading this short article, I hope you are convinced that learning is a good investment and achievable in a corporate context. After a few years managing Data Science teams, I spent time reflecting on what is the ideal profile for a data scientist. I concluded that hiring employees with great learning potential rather than highly skilled ones is a better investment. For sure, if a candidate is highly qualified AND willing to learn, it is even better. This completely changed my vision of management and learning is now at the core of my team management strategy.
my Linkedin