Driving efficiencies in training with algorithms

Arm your workforce with precisely the skills to perform their roles more efficiently and your business will truly prosper, Chris Littlewood says

The need to train staff to ensure the workforce is able effectively to drive the business forward is a well-recognised business priority. But despite this, there is a shortfall being seen in terms of implementing training in practice. Research conducted by Filtered.com with Opinium of 2,000 UK employees indicated that only 25 per cent had received training, despite 60 per cent claiming they needed support in developing key workplace skills to perform successfully.

A disconnect exists then in the training requirements of staff vs what’s being delivered.  Further digging shows that time (40 per cent) and cost (31 per cent) are the main barriers to training. In looking to find a solution to this, Filtered established that finding ways to personalise training for staff, to ensure each student is taught just what they need to learn, is one way of reducing the drain on resources. Creating a smart algorithm for digital learning modules, that intelligently strips out material that a student already knows, or that isn’t valuable in their work, is one way of doing this.

Algorithms in general

The use of algorithms is widespread and mature in many online industries – from Shazam helping you to identify the tune playing in a bar, to your online bank encrypting information so it can be communicated securely to the Cloud, or Google correctly identifying that you wanted to read this article from a few search terms. Many of these companies are doing almost exactly what needs to be done with training – they are selecting content for an individual based on what’s known about them.

So surely we can just copy this process and apply these algorithmic techniques to L&D? Well, yes and no. Undoubtedly there’s a lot to learn from these giants of prioritisation, like Google, but there are some specifics of training that mean we have to adapt their approach. An example relates to the importance of getting recommendations right. When Amazon recommends that I buy something I don’t want, sure they might have missed a sale, but I’ve wasted no time. Generally with products, the buyer can make a well-informed decision as to whether they want the product instantly. With training content, the student will generally be taking a leap-of-faith in committing time to the material that’s being recommended, so the onus on making all recommendations useful is much higher.

We’ve crystallised our experience of developing algorithms into five principles. Some of these apply generally to the use of algorithms for personalisation; some are particular to training.

Principle 1: Evidence basis

Personalise content according to evidence from users.

This almost goes without saying, but your algorithm needs to have a foundation in real user data to stand a chance of returning reliable results. Often usage data will be at a premium, even in online businesses with global users, because training content is continually revised and extended. So treat evidence like this like gold.

Principle 2: User focus

Inferences need to be useful for real users, so you need a good understanding of them.

Having said that data are paramount in Principle 1, we need to add a caveat that the training that we present to students is coherent as a whole. There is a danger, when employing algorithms that the micro-decisions being made do not add up to a sensible whole. For example, Filtered users study courses, so we need to think not only about which course modules are most relevant to each student, but also how those selected modules fit together to form a complete course.

Principle 3: Supplementarity

Add to rather than replace a user’s own insight.

Students of training materials will often have a strong, and accurate, idea of what they need to learn, and how they want to learn it. So it is important that any insight an algorithm can add in personalising their training enhances rather than replaces this prior knowledge.

Principle 4: Transparency

Explain clearly why the decisions of an algorithm are what they are.

This principle is about gaining the trust of your students. If you are asking someone to commit time to training, they are more likely to be engaged if they trust your advice. If it is clear why a training recommendation is being made, students are more likely to trust the advice and to make more robust decisions as to whether to accept the recommendation. This principle lay behind our decision to develop our own user-clustering and Bayes’ theorem-based algorithm, rather than use an off-the-shelf but black-box machine learning algorithm. Our Bayes-based approach means we can isolate the effect of each piece of information on a student’s tailored course, so the algorithm’s decisions can be unpicked step-by-step.

Principle 5: Acknowledge uncertainty

Give information on the confidence that can be ascribed to an algorithm’s decisions.

This principle is about empowering your students to make good decisions about their own training.  If you can (in a clear and simple way) let students know how certain you are that some training material is relevant, they can supplement your advice with their own judgment more intelligently.

Working smarter, by incorporating ways to personalise learning materials to reduce time and cost is becoming a key enabler of robust and effective training. But beyond time and cost, the opportunity to deliver what is truly valuable to the student’s role, significantly increases the impact of the training on performance.  Arm your workforce with precisely the skills to perform their roles more efficiently and your business will truly prosper.


Training Journal

Learn More →

Leave a Reply

Your email address will not be published. Required fields are marked *