Thread by @ValentinUmbach

Analytics work can be roughly split in two buckets: (1) Building automated systems, from metrics to dashboards, to enable self-service use cases for business users. This is what we now typically call analytics engineering.

(2) Doing ad-hoc analyses, to answer some questions directly. The problem with this second type of work, as @bennstancil points out in another insightful post (benn.substack.com/p/big-whiff), is with storing and organizing the results and insights we produce here.

So instead of systematically building on top of previous work, we often forget about it and start all over again every time.

At @lovoo we try to solve this by maintaining an Insights Knowledge Base. Our approach is very simple technically, and they key for making it work is more about having a good process in place. So what do we do?

(1) We do an analysis. This can be done in different tools, and take many shapes and forms. But ultimately, every analysis is documented in a Jira ticket. Including all work artifacts that allow us to recreate it, as well as results and recommendations.

(2) Every week on Friday, we get together as a team and go over all analyses that were completed. We determine what information is worth storing for later retrieval.

It needs to be (a) relevant to our business, (b) focussed to a specific insight, and (c) general enough to be useful outside a one-time use case.

(3) We write down the insights that fulfill these criteria in a Google Sheet. There is typically only 1 sentence per insight, and every insight gets a new row. One analysis obviously can result in multiple insights.

We also use tags for categories, which makes it easier to find and connect existing insights. And we link to the original ticket.

(4) We feed this Google Sheet into a Tableau dashboard. This brings it into the same environment that we use for our self-service dashboards, and therefore closer to our business users.

It also allows for easy search and filtering, as well as some simple statistics (we visualize the accumulation of insights over time, by categories).

(5) Once a month, we automatically push the latest insights to the #general Slack channel, to share them with all team members. This is great for creating awareness, and encourages everybody to dig deeper into our knowledge base.

Now, every time we get a new ad-hoc request, we start out by checking if there are previous insights that can at least partially answer the question, or should inform any new analysis we do.

(Of course, we also get ad-hoc requests that can simply be solved by pointing the person to the right self-service dashboard.)

We have used this approach for 2.5 years now. It requires some conscious effort, but it has enabled us to look back at our accumulated insights as a growing treasure of knowledge, instead of a meaningless pile of ad-hoc results.