Machine-augmented idea triage

Estimating the importance and dynamics of a topic

Idea triage plays an important role in business strategy, particularly in resource allocation. For instance, imagine that there are 10 technology fields or business areas that all sound promising, but you only have the resources to actually get into 3 of these topics. Which 3 out of your 10 technology fields should you select, and why?

In order to address this question, you could for instance try to estimate the importance and the dynamics of all 10 fields, and then select the top 3. You could map “importance” e.g. to investment volumes, numbers of publications, patents or news; “dynamics” could be mapped to the growth of these volumes and numbers.

The growth-share matrix implements this idea: it maps importance, or share, against the dynamics, or growth, of a technology field, business area, or other topic. So far however, such matrices have to be generated manually because the underlying data is usually more qualitative than quantitative (I am talking about technology fields or business areas here, not e.g. portfolios of investment instruments, where one usually does have quantitative data). As everybody knows who has ever done it, building such a matrix manually is a lot of work. And because it is so much work, when you are done, you are not exactly keen on “just quickly” revising 8 of your 10 data points again (e.g. because something new and important has happened since you made your matrix).

We wanted to make the generation of such matrices more agile. This is why we built an automated version of the growth-share matrix. This new tool, the Topics Matrix, is available in our analytics platform. The Topics Matrix implements some observations that we have made over the recent past. When we were searching for various topics in our platform, we noticed all four combinations of big vs. small and low- vs. high-growth topics.

Big, low-growth topics

We noticed that there are big but low-growth technology fields such as “graphene”. Graphene has quite a large share of patents, scientific publications and news, for instance (so far it lags behind in venture investments though). While the share of graphene is large (except venture investments), it has not really grown a lot over the past few years. You can see this from the sparkline charts below, which are taken from our platform:

Graphene in scientific publications

Graphene in patents

(patent publications tend to lag behind a bit, so the drop in 2016 is an artefact)

Graphene in technology blogs and news

Small, high-growth topics

On the other hand, we noticed that there are small but high-growth technology fields such as “edge computing” (there is a very good presentation on edge computing by Peter Levine from a16z, and we also published a blog post on edge computing previously). So far, edge computing does not have a very big share across our information categories yet. But it has been growing a lot more, compared to graphene, for instance:

Edge computing in scientific publications

Edge computing in technology blogs and news

And we saw…

Big, high-growth topics

…such as “internet of things”, where all volumes are big and growing a lot:

IoT in scientific publications

IoT in technology blogs and news

And, well, of course we also saw small, low-growth topics. But there just is not much to see there, so no charts for these topics here.

Putting it all together

Now we wanted to have a more systematic approach, rather than a collection of observations. So after quite a few experiments, we decided to build an automated version of the share-growth matrix. And here is what the first version looks like:

(did I mention that the Topics Matrix is available in our analytics platform?)

The x-axis represents share (watch out — log scale!), and the y-axis represents growth (calculated as a simple compound growth rate of shares), all over the past 36 months. Each data point represents a stored search in Mergeflow. So, in order to add another data point to your Topic Matrix, simply run a search (iteratively fine-tune it if necessary)…

…store it…

…and add it to your Topics Matrix:

In the current version, you can add up to 10 stored searches to your Topics Matrix. By the way, of course, topics can be anything that you can express via a Mergeflow search: technology fields, companies, investors, people, etc..

Now, back to our Topics Matrix from above, and to some of our observations. Above, we compared “graphene” (big, low-growth) to “edge computing” (small, high-growth). Notice how you can see this difference in the Topics Matrix:

For each search, the Topics Matrix calculates an average growth and share score across VC Investments, Scientific Publications, Patents, and Technology Blogs and News. Of course, it is possible that a topic grows or is bigger in one of these categories but not all of them. In such a case, this category (e.g. VC investments) is the main reason why a data point sits where it sits in the matrix. We made the matrix interactive, so you can see whether this is the case. You can right-click on any data point, and then zoom in on the sub-categories. For instance, here is what this looks like for the IIoT (industrial internet of things) datapoint in the matrix above:

The IIoT tooltip covers the “VC investments” data point in the screenshot, so I added a boxed version of it (cf. grey boxes and arrow) so that you can see it better. The distribution in the matrix above shows that Industry News and VC Investments are the main drivers of IIoT share. Industry News and VC Investments also drive IIoT growth, along with Scientific Publications.

If now you want to dig even deeper, you can do this too. Left-clicking on a data point opens a new tab that shows you the underlying data. This way you can check, for instance, whether there are outliers or what exactly is behind the data in your matrix.

What’s next?

“Interesting! I can do this for any of my topics on-the-fly, and the data will always be real-time. But can I also see how the data points have evolved over time?” — “Well, not yet…”

The conversation went something like this whenever we showed someone an early prototype of the Topic Matrix. So, yes, this will be next. We are working on an update that shows how each data point developed over the recent past. As soon as this is available, we will post it in our blog.


Also published on Medium.