Get a first-hand look at innovation as it happens at Docebo.
Because we are an innovation-driven learning technology company, we wanted to pull back the curtain and give a first-hand look at how new ideas come to life within the Docebo learning platform. The Docebo Discovery Lab explores the technology (and the people) behind the projects guiding us into the next generation of learning technology.
Building a Smarter Platform with Artificial Intelligence
In 2018, Docebo introduced the first generation of our Artificial Intelligence-powered learning platform, equipped with a set of new and exciting functionalities designed to initiate a seismic shift in the way learning is delivered by, consumed and produces value to business outcomes. This groundbreaking project was the result of a prolonged research and development effort, powered by a dedicated team of Artificial Intelligence data scientists, machine learning specialists and adjacent field experts.
Their mission: build an AI-powered learning platform that will not only enhance and personalize the learning experience, but also make the lives of learning administrators easier by automating routine and menial tasks.
And we’re off to a great start.
The Data Advantage
The AI team works in two ways to formulate solutions and make them a reality within the Docebo platform.
First, CEO Claudio Erba consults with the AI team whenever he has a potential idea and encourages them to develop and test it. The AI team is also constantly on the hunt for the latest developments in the AI space, researching and investigating the potential of these by running experiments to understand if they would deliver value to the platform.
Developing an AI algorithm always starts with a dataset. As a learning platform that has existed for over a decade, Docebo has developed a massive bank of data streams that are based on a number of variables, such as learner behavioral activities (answers, likes, dislikes) as well as information and metadata on learning content. All sensitive data is anonymized and the artificial intelligence used in the platform is GDPR compliant.
Using this data, the AI team has been busy testing hypotheses and developing machine learning algorithms to evaluate potential solutions. The algorithms identify patterns within the datasets to understand learner and admin behavior and then generate a solution to the original problem tackled by the AI team.
Algorithms are then tested to see if they perform well or not – a process known as the Train, Test and Validation procedure. If it isn’t working, adjustments are made and the entire process is repeated to eventually uncover an algorithm that delivers the most effective model to solve the given problem.
AI in Action
For example, when a learner uploads a piece of content in Coach & Share, Docebo’s Invite-to-Watch feature automatically produces a list of learners within the organization who might find it interesting, based on similar content they have interacted with in the past. While the feature ensures user-generated content is shared across the organization in the most effective and valuable way, the AI-powered tool is also able to refine the results it produces over time as it continues to understand different learner preferences and how users interact with different pieces of content.
As another example, when admins upload a new learning asset (e.g. a video), the Auto-Tagging feature uses AI to “listen” to the entire video, understand the most important keywords, and automatically create up to 10 tags. This streamlines categorization for admins and also makes it easier for learners to find content that is relevant to their interests and professional development. Admins can manually edit the generated tags, and AI tracks this feedback to produce more relevant tags the next time an asset is uploaded.
The continuous nature of AI-powered functionalities is an exciting one because it has the potential to completely change how learning can provide value to the learner. No longer are they given a solution and that’s the end of the conversation. As AI solutions refine themselves in the background as they’re fed more and more data, the quality of the results increases and the value of these tools grows.
“The base of every single block we are building is machine learning,” says Calogero Zarbo, who works as a Machine Learning Specialist on Docebo’s AI team.
“Every day we are improving bit by bit the features that already exist; we’re uncovering new things and implementing many new features.”
Innovating the Core Platform
In addition to new functionalities, the AI team is also leveraging its data sets to discover how admins are already using the platform – and how we can build solutions that continue to simplify their activities.
Within the Docebo platform, there are two types of admins – Superadmins, who have control over every aspect of the platform and Power Users, which are created by Superadmins, who are assigned a specific set of admin permissions for certain users, courses, catalogs, locations, and more.
Superadmins manually assign permissions to Power Users and the AI team has collected data on this to investigate the possibility of pre-creating Power User profiles in the platform, reducing the time taken for Superadmins to carry out this task.
“We have a lot of permissions for Power Users that are adjusted for different profiles. One problem we have focused on is understanding if there are patterns showing us what permission settings people prefer to use so that we can possibly offer those as bundled presets and simplify people’s choices.”
Calogero applied machine learning algorithms to the Power User permissions data to cluster it into groups of the most-used profiles – and then visualize these findings.
He discovered that there were at least 3 very different profiles. This finding on its own presented a useful insight into how the platform is being used, which allows us to produce better and more user-friendly solutions.
“This type of innovation comes only after discussions with many people around the company. Everyone brings their little bit to the table – they might not be research scientists, but they have incredible domain knowledge about the platform. Our goal as data scientists is to empower that knowledge by using machine learning and AI techniques.”