Byline
Barbara Galiza is a freelance growth & analytics lead, who's worked with brands like Microsoft and Her (YCombinator). Read more about Barbara at www.barbaragaliza.com.
Version 2 - September 29, 2020
Investing in data documentation, governance, literacy and consolidation is usually postponed for too long. There are a few reasons for that.
The most glaring one is that there's no direct ROI from investing in a data stack. Therefore, it's usually deprioritised against projects that impact revenue: building the features your customers request or spending on sales and acquisition. That's understandable.
The other reason is that there's no "one size fits all" when it comes to data. How much you should invest (time and $) in a data stack will depend on your stage and your product. A B2C company needs to mature in data a lot faster than enterprise software: they're building for the many so customer feedback is less useful and UI event tracking gets expensive very fast (think Mixpanel).
Over the years, I've seen signs that indicate that your level of data maturity is holding your revenue growth back. In these cases, it's worth taking a step back to reevaluate how you're capturing, measuring, passing and documenting your data.
You're forwarding spreadsheets around and you're still referring to the same out of date growth model from two Qs ago.
It's natural that ad-hoc research ages, but still holds value. But, unless you've pushed no product or marketing changes in the period, you shouldn't assume your conversion rates have stayed unchanged.
That's why data models that impact business decisions should be maintained and/or be easily updated by your data analysts.
The last thing you want is for the process of updating data to take as much as it did to build the model in the first place. In fact, I'd say updating data sets shouldn't take over 10% of the time it took for the original research. If it takes longer, you're playing with the risk of human errors.
You cannot postpone documentation if you rely on complex models for business decisions. There are new tools focused on bringing consistency to data models (e.g. dbt) which can help you fix this issue. To start off, keeping a Notion, Github or Slite document up to date with how models are built could already be of big help.
Looks like an innocent enough question, but usually points towards two foundation problems: