The Semiosis of Visualizing Data

What is semiosis?

Any sort of dictionary will give you a meaning for semiosis, but here I mean the process of using data visualizations to give meaning to data. It’s one thing to visualize data. It’s another thing to visualize it in a way that’s useful to the person who sees it. (We wrote recently about data visualization for blind people, such as myself. We’ll continue to use the verb see.)

Data visualization can be thought of as having several axes, with two of the most common being slickness (or beauty or elegance…) and utility. LinkedIn and other places are full of very slick, very useless data visualizations. Business Intelligence (BI) dashboards can be composed from numerous non-slick (downright boring) but useful visualizations.

Many people in the field of data visualization try to improve both of these axes at once, which is harder than improving only one at a time.

One size does not fit all data

Not all data needs a slick presentation or much of an explanation. “Annual Revenue for Fiscal Year 2024, in thousands of US Dollars”. Once you’ve written the title of the data, you’ve given all of the essential characteristics except the number itself. One could, I suppose, put it in a bold Gothic font or write it in multiple colors, but really, the thing explains itself.

This is in contrast with the same class of data, from the same source, but with a different time grain (a fancy way of saying granularity and detail). In many BI tools, users can “drill down” from the annual figure to get a finger grain of data. (This requires some setup.) And here, merely changing how many times we are simultaneously presented with the same type of data changes how we interact with it. There are people who can “read” a list of numbers and understand certain aspects of them without external tooling or summarization. There are also people who can free climb mountain faces. And then there are the rest of us.

Sense-making > Style-making

It should be uncontroversial to assert that it’s more important to make sense of data than to present data stylishly. And, in theory, nearly everyone agrees. But years and years of being in the field have taught me that this principle needs repeating, and that each new generation of business users (including business analysts) needs this repeated.

Slickness is nice if utility is already handled. But slickness is not a substitute for utility, and in most companies, once a dataset has had its accuracy and all of the other aspects that make it useful validated and has been made into a data product (or dataset or model in a BI tool or however this is done in the company), it’s time to move onto the next data.

This is also why, as of this writing, generative AI is (once again) not usually the right tool for generating data presentations to be used internally. We’ll write more about this later, but a fundamental flaw in, let’s call it vibe analysis, is that the generative aspect of AI is going to, by its nature, focus on the statistically popular methods and presentations. Maybe that’s fine, but it also requires a lot of checking to be sure the presentation is

• built on accurate data,

• has the same semiosis as other data in the company, and

• can be modified.

However, if you’re just trying to put lipstick on a pig to get new investors, you’re probably too busy talking with your LLM to read this. Semiosis serves a different purpose in that case.