28 Apr 2016 Unlocking the potential of the strategy machine: an experiment with NLP

What does it mean for technology to enhance or inhibit strategy? To explore this distinction, let’s consider how two software tools using the same underlying technology—natural language processing (NLP)—can affect human thinking and business outcomes in opposite ways.

We believe that unstructured data—text, in particular—could transform the way we think about business strategy. According to one estimate, 80% of the data that businesses hold is text data, which conceals rich and unexploited insights. That’s why we have been investigating NLP technology.

The first tool, which we developed with our collaborator Cortical.io, is a semantic engine whose role is to provide analytic horsepower and flexibility in order to stimulate human thinking. It’s able to detect semantic patterns in text data and to do so from the perspective of a particular user, like a consulting firm or a pharmaceutical company.

But to produce sharp insights, the technology needs to interact with people, who are able to interpret patterns and anomalies, connect findings with outside knowledge, and form new questions and hypotheses. The program serves primarily as a partner in human thought and thus can play a role within an integrated strategy machine.

We used this tool to investigate the relationship between corporate speech and an organization’s orientation toward either exploration (the search for new opportunities and business models) or exploitation (the refinement of existing products and business models). We started by using BCG strategy texts to train the algorithm to look at the problem in a strategy-specific context. We then developed hypotheses using both human-driven and machine-driven approaches: using human intuition and experience to posit testable relationships and using the machine to extract commonalities across known explorers like Amazon and Google. We analyzed the messy raw output for interesting correlations and repeatedly reframed the analyses to investigate the most promising patterns. Eventually, we found a way of accurately detecting and predicting the exploratory capability of corporations, which can be used to assess their strategic sustainability.

The second NLP tool we examined is a well-established one that is often used for “visual storytelling.” The software uses keywords to search the web and presents its search output in a visually compelling and accessible way. It gives us a plausible picture of reality, onto which people can impose their own interpretations. However, we found that it doesn’t point us to new concepts or questions. In fact, the outputs of this software are either questionable or obvious, because they don’t allow us to easily explore and test new hypotheses. We can either accept the output, or it remains closed to further investigation.

As a result, this software gave us a very superficial view into the same question that we investigated with the first tool. The software gave us interesting visuals but no real insights about the relationship between speech and exploratory behavior.

The fundamental difference between these software tools is the extent to which they are open to human interaction. Software that can produce complete outputs by itself is closed to human reframing. But an integrated strategy machine necessarily involves both people and machines. The kind of software that can stimulate human thinking and play a role in the integrated strategy machine is, paradoxically, the one that is incomplete by itself.

This column first appeared in bcg perspectives, as part of an article entitled “The Integrated Strategy Machine: Using AI to Create Advantage”.

Author: Martin Reeves, Senior Partner and Managing Director, BCG
back to opinions overview