Coaching works when it drives change. Human coaches use their intelligence to learn enough about a team to teach them how to change themselves. CoachBot shares that aim and needs a similar level of intelligence to succeed. Artificial intelligence (behaviour and features that are learned or evolved by algorithms) can help, but many features work well using only intelligent ideas that are programmed with simple logic.

How CoachBot grows in intelligence

Most of CoachBot's current intelligence is non-artificial intelligence. The Saberr team's research into the fundamentals of teamwork means CoachBot asks clever questions that make teams think about themselves. And CoachBot guides teams themselves to think about and discuss smart ways of improving themselves. 

A lot of what teams write to CoachBot is open text, and learning from that text will gradually allow us to teach CoachBot to engage more directly with what people say. 

Some things are already possible with natural language processing: we already categorise and automatically illustrate "agreed behaviours" that teams write for themselves. And when a team has entered retrospective notes they want to collectively reflect on, we can sort them so that the discussion is more fluid from one to the next. These features are only possible by using trained natural language processing models. We usually hide the fact we're even doing this, since these ideas just subtly improve the user's experience. 

We're experimenting with more immediately-visible ideas, like turning a goal ("Produce 300 widgets by May") into a progress tracking slider for teams to easily update. Our main challenge with natural language processing is that teamwork is quite a niche topic. There aren't any available datasets of "how teams talk about themselves". So we rely on generic language models to understand user's text (trained on things like Wikipedia and news articles), with a sprinkling of expert knowledge about teamwork topics. As CoachBot grows, this niche knowledge will become richer and surpass our expert knowledge. That's the future of CoachBot's intelligence: learning from working with thousands of teams. Not just what they write about themselves; but also tracing the actions and journeys teams take and the resulting changes in their productivity and happiness. Combining those observations with the team's context, we'll be able to recommend the best next step for any team.

Did this answer your question?