Steven Forth is a Co-Founder of TeamFit. See his Skill Profile.


A well-respected book on competency management defines competency models as “a descriptive tool that identifies the skills, knowledge, personal characteristics and behaviours needed to perform a role effectively in the organization and help the business meet its strategic objectives.” (from The Art and Science of Competency Models by Anntoinette D. Lucia and Richard Lepsinger).

This book was published in 1999, and, reading this today, one sees two things that seem out of synch with current needs. Competency models are seen as ‘descriptive,’ and the focus is very much on the organization and not the individual. Let’s consider each point. In today’s more fluid and team-based organizations, the focus on ‘role’ also seems a bit anachronistic.

From description to prediction

Given the rapid advances in machine learning, we are moving from description to prediction. The title of a well-received new book on machine learning is Prediction Machines: The Simple Economics of Artificial Intelligence. The authors are Ajay Agrawal, Joshua Gans and Avi Goldfarb, from the University of Toronto’s Rotman School of Business and Creative Destruction Lab (an accelerator). Their key thesis is that machine learning makes prediction cheap, and that there are many problems that we do not generally think of as prediction problems that can be solved with prediction.

One of their examples is machine translation and speech recognition. In the past, teams of linguists and computer scientists worked together to build complex models of the connections between languages, with large syntactic and lexical mappings front and centre. Today’s systems are built by deep learning methods and are much more accurate. They take advantage of a large number of translations already done and available online to predict from one language to another.

What do we want competency models to predict? The obvious answer is performance. This needs to be unpacked. Performance can mean many different things. Conventional competency models assume we are talking about ‘performance in a role.’ Is this what is most important to business? No. Business is focussed on results, so a different way to think about this is ‘performance in achieving a goal or objective.’ Given the importance of teamwork, one may also want to be able to predict ‘success in working with these other people.’

Deep learning will change how we think about competency models in two ways:

  1. What we are able to predict
  2. How we build the models

What do we want to predict?

We need to look at this from two perspectives, the individual and the organization. In the past, too many competency models have been organizational centric and have not been of much help to the individual. See Lee Iverson‘s post on Why competency models work for companies not employees. This will change as people take more control over their own data and responsibility for their own careers.

As an individual I want to be able to predict a few things:

  • What skills will I need in the future?
  • What skills can I develop? (What are my potential skills?)
  • What roles will I be successful in?
  • Who should I work with?
  • What problems should I work on?

The next generation of the competency model, predictive competency models, have to help individuals answer these questions.

Organizations have related needs.

  • What skills are driving performance today?
  • Where do we have skill gaps?
  • What skills will drive performance in the future?
  • Where will we have skill gaps?
  • What resources will help us to close these skill gaps?

How will we build the models?

Given the need to build predictive models, how will we go about this?

The older top-down model of having a group of experts define the skills needed for a role will not be enough to support the transition from description to prediction. The next generation of models will need to gather and structure the data needed for prediction engines. The current top-down approaches do not do this as they assume what needs to be predicted. They need to be replaced, but with what?

The bottom-up approach to skill models developed at TeamFit using innovations from approaches like and Gleanr is more dynamic and gathers a richer data set. It is a step in the right direction. But our experience has shown that it is too chaotic for many companies and relies too much on employee engagement.

A new approach is needed, one that blends the top down and bottom-up approaches.

Top down and bottom up

(This is a visual representation of a concept blending is based on the work of Mark Turner.)


This is a necessary step towards the development of predictive skill models. Other things to consider are the range of data needed to power these models, which goes well beyond what is gathered in most conventional systems, and how to collect the outcome data on performance that is needed to train the system.

The role of judgement

One point made by Agrawal, Gans and Goldfarb in Prediction Machines is that improved predictions increase the need for human judgement. On page 75 they present the following model.

For this approach to work, the system needs to be able to collect Input and Training Data, a Feedback mechanism is critical (this is one of the things missing in existing competency models and talent management systems), and people have to exercise judgement.

TeamFit’s goal is to make it easier for both individuals and organizations to make predictions about the skills needed for performance, so that they can exercise judgement about what skills to develop, how to apply them and who to apply them with.