Google Sheets as a BI tool

People are already using Google Sheets as a limited BI tool. Generative AI will only make it more powerful! Brace for massive disruption here.

I probably first got this idea when I was working for Delhivery, in 2022 or so. My team had the job of putting together, manipulating and disseminating some data on a regular basis to a bunch of people in the company. The question was how we will do the last bit.

We were an AWS shop, and one option was QuickSight. But building a dashboard there was insanely painful (all drag-drop; no way to code it; etc. I have no clue how much easier it is now with Q). The other problem was that all our users then had to be onboarded on to QuickSight, which was easier said than done.

The standard way of disseminating data in our company was through scheduled reports (using Redash, I think - I don’t remember for sure). A lot of reports were sent this way. However, most of them were simply (though not simple) SQL queries against our Athena backend whose results would then get mailed to a bunch of people at a particular frequency.

I remember one ops leader cribbing to me that he would get 500 such emails a day (all of which would go to some folder in his GMail, of course), and it was impossible to figure out what to pay attention to (that was one of several triggers that led to what ultimately became Babbage Insight). Anyway what we were doing went beyond just SQL queries - by pulling data into Python / R, we were doing “gymnastics” before disseminating it. Also, we didn’t want to send a 501st email that might get lost in the clutter.

And so the answer was Google Sheets - we could populate and manipulate them programmatically, everyone used them anyway, and people could simply bookmark the page and see it every single day. There was only one fly in the ointment - you couldn’t put filters on them. Rather, if one user put filters, those would apply to everyone else as well, and that would mean that they would be ineffective as a dashboard. At least what we meant by “dashboard” in 2022.

Google Sheets in the era of LLMs

Last month, I had written about how LLMs will transform BI. One prediction I made (obviously it’s too early to evaluate it) was that dashboards will become personalised, and the concept of “dropdown” will simply disappear.

Now think about it from the point of view of Google Sheets - if the Google Sheets are hyper-personalised, and they can anyway be produced algorithmically, then they perfectly fit being a BI tool! Even if you want to have dropdowns, they allow for it - and won’t mess with anyone else’s experience.

You will end up with one separate Google Sheet per user (all of which will be populated based on numbers populated into one “master” data mart; so that numbers across people’s “dashboards” agree - this is one objection made to the post linked above). And each user’s view of the data will be put there.

Customisations!

It gets even better! With each user having their own Google Sheet, they can make their own personalisations, if necessary. Maybe someone will add a bunch of pivots. Someone else will add a bunch of graphs. Someone else might choose to highlight or format (“decorate”) the sheet in some manner.

The beauty with generative AI is that all of this can be learnt (periodically), and persisted with. So if someone has a pivot, then whenever their sheet is populated, the same pivot gets updated (and data added in a way that the pivot doesn’t break).

The other thing is that pivots, graphs, etc. are things that business users are familiar with. It is part of their workflow. They know how to make and manipulate them. This is very unlike traditional dashboarding tools (such as PowerBI or Tableau) where users don’t know how to get things done, and only particular specialists can do so.

Combine that with the fact that these traditional dashboards are “one size fits all” (one dashboard used by multiple users), and you know why dashboards bloat like crazy, and the BI team is always under stress from users asking for additional feature requests. Because users can’t implement these features themselves, for themselves!

It’s already being done!

I know of several companies that, one way or another, have implemented Google Sheets as a dashboarding tool already. They are typically read / comment only (so no custom pivots or charts, or custom data for each user), but all data in one place that people can chat about and have meetings based on.

They also have more traditional dashboards, but leadership usually prefers Sheets. This is a format they are used to, there is no VPN nonsense (one reason I’m told for low adoption of Tableau / PowerBI), they don’t need one extra login (Google SSO is good enough) and it is readily available on mobile (remember that business leaders seldom use laptops).

And after a lifetime in data visualisation, and advocating good data visualisation practices, my interaction with business leaders tells me that for the most part, they don’t really care - as long as the data is accurate and well-presented, visualisation is overblown.

And with full benefit of hindsight, maybe we could’ve simply eschewed all the front end and authorisation development, and built Babbage Insight with a Google Sheets front end!

Reminder: I’m in the job market, looking for leadership roles in AI / Data. Ideal would be something at the intersection of AI, structured data and business. If you might find someone like this useful, please let me know!

PostScripts

1. Having got this far, I realise there is one contradiction. One downside of everyone having their own sheet is that the social element of a shared sheet (comments, and replies, and tagging) goes away. I don’t know how to square this. Yet.

2. I hope there is no smart Looker Product Manager looking at this post and noting the threat that Google Sheets presents to Looker, and lobotomising Google Sheets (Google as a company is well known for making products worse or even killing them). Microsoft’s online office stuff sucks big time, and is not a viable alternative.

3. Thanks to Vikram Nayak for a brief discussion on Tuesday that led to this post.