Our experience using NLP to identify internal company trends in Slack over the last 6 years
A long time ago, our CEO used to have one-to-one discussions with every single Futuricean over the span of a calendar year. He did it to keep his finger on the pulse of what’s happening in the company and what people are thinking about. It was a great way to maintain an understanding of whether our autonomous organization was really sailing in the same direction and aligned behind a shared strategy. As organizations grow, we need to look for ways to scale the appropriate level of control and understanding. Our old tools are no longer effective.
That brings us to the question of following how organisational strategy progresses, which usually means pre-set KPIs and initiative reporting. It works, but we lose some of the value face-to-face discussions provide. They have a hard time answering questions like: are people talking about our strategy? Are they excited about it? Are they aligned on the topics in everyday life?
Together with Aalto University, we tackled this challenge with the help of data & AI and asked, “How might we use the digital footprint of employees’ daily online discussions to understand how our strategy is diffusing across the company?”.
Three (not very easy) steps:
- Understand the data and the culture that generates it
- Analyze topics that are representative of your organization
- Create an action plan by first understanding what’s hindering your strategy progression and alignment
Let’s look at this through a case example — our very own Futurice Group!
Since 2017, building up capabilities in data & AI has been an important focal point in our strategy. We wanted to understand the role of this aspect of our strategy in the daily lives of our employees. Is it a topic for data scientists alone? Or are we as a company engaged with the idea that data & AI will fundamentally change the way companies operate and create value in the future — regardless of background or position?
Understand your data and the culture that generates it
Transparency is one of our key values. We are a company that discusses topics openly. This is a blessing, as it means we generate a lot of data through these discussions. As data sources, we used Flowdock & Slack — two discussion platforms we’ve used consecutively, starting with Flowdock, since 2014. The key is to identify byproduct data instead of directly asking people how they think our strategy is progressing. For the sake of privacy, we don’t use 1-to-1 chats or private conversations. We also proactively filtered out any other possible discussion threads regarding e.g. personal topics (salaries, etc.) and NDA stuff. For companies that don’t openly discuss topics in company chats, other potential data sources could include publications — in the case of Universities — or sales proposals — in the case of service companies.
Analyze topics that are representative of your organization
We use machine learning to automate the extraction of topics from unstructured textual data. Topics are extracted automatically, based on the distribution of words across documents. The method is designed to process a very large quantity of text materials. Topics emerge from the data, so there’s no need to define topics ex-ante. Topic extraction is “objective”, i.e. based on actual data rather than a survey or human-defined lexicon. Data is cleaned & harmonized by e.g. identifying the language, taking out filler words, identifying N-grams (artificial_intelligence vs artificial and intelligence), or removing personal information.
Create an action plan by first understanding what’s hindering your strategy progression and alignment.
So did we learn anything? Yes! We’ve actively discussed data & AI-related topics over a number of years, so it’s clearly identifiable as a topic among discussions. The figure below presents the Network of topics in futurice Slack discussions. Topics are related based on the similarities of their lexicon. We can see a huge variety of topics in this figure, related to the organization of the work, collaboration, technical topics, or even life around the workplace. We zoom in on a part that contains the topics related to artificial intelligence. Many other topics could also be highlighted: at the other end of the figure (the grey-white circles in the yellow area down left), we find topics related to organization and strategy. Those topics relate to a very different set of vocabulary than the AI topic, in current and past discussions at Futurice.
Network of topics in Futurice discussion
We can also clearly identify an increase in the amount of discussion around this topic in recent years — as well as some evolution in how it’s talked about. As the discussion evolves, the set of vocabulary becomes more diverse and you can see more detailed topic areas arise from the same high-level theme. For example, one topic around data & AI can become three separate but related discussions: data management & privacy, artificial intelligence and machine learning & algorithm and data training.
Very few people participate in both AI and strategy discussions. In the figure above, each dot represents a person at Futurice connected (i.e. taking part in discussions related) to the AI or the strategy topic, or both.
One of the most interesting insights we gained was the fact that there’s a lack of connection between AI and strategy as topics of discussion, as we can see below. A deeper look shows that most people discussing “strategy” are not participating in discussions about data & AI — and vice versa. In fact, based on our data only a handful of our management is fluent in both topics. To truly drive change, we need to get the management on board, too — not just our domain experts.
So what does all this change for our future?
As a company grows, the easy thing to do is to create control mechanisms based on hierarchy, structures and processes to ensure the company is sailing in the same direction. But what if we created increased visibility into our thinking, even at a scale of +1500 people, could we avoid employing control mechanisms and instead ride the wave of trust and transparency?
That’s one of the things we aim to find out.
We are on a journey to identify how we can evolve and transform the way we lead companies with the help of data & AI. We believe the future belongs to connected companies.
Stay tuned if you want to learn more!
If you’re interested in our thinking and work in this area, we are running a series of webinars around this subject, Amplify strategy execution with data & AI episode coming up already 3rd of June, and you can take a look at some of the concrete work we do here.
Originally published at https://futurice.com, written by Anni Harju, Mickaël Buffart, Mia Leppälä.