rw-book-cover

Metadata

Highlights

The actual analytics was always a side quest, a reward for those rare occurrences when everything on the tracking side was working correctly. One report there, a dashboard maybe, but not so much of deep-dives, exploratory analysis, regression or whatever an analyst is supposed to do. So today I boast the title of an Analytics Engineer humbly admitting the fact that I have never been a full-time analyst after all. Even though I had that word included in my title at three different companies, analytics was very rarely a thing I did for living. (View Highlight)

Digital analytics as a discipline was brought to life because of the popularity of an amazing tool Google Analytics once was (and still is, to some degree). (View Highlight)

That created a second stream, sort of data-samizdat for those who wanted some charts but didn’t have the capacity, knowledge and in fact, valid reason, to fight with OLAPs and Hadoops of the world. It developed separately from the rest of the data world, had its own focus points and buzzwords (with attribution being possibly most notable of both) (View Highlight)

Fast forward to now, this second stream still exists. It’s much more connected to the first one with the emergence of free and paid BigQuery exports, development of R and Python packages for pulling out GA’s data. But one has to wonder, are the reasons for its existence still valid? Is there still a hole big enough for digital analytics driven by out-of-the-box tools to fill? (View Highlight)

However, the most important change is the pace at which the “mainstream” has begun to reshape itself. It’s fast and we, digital analysts, have a lot to catch up on. (View Highlight)

Even with the dbt itself, it’s now possible to model data from generic clickstream tracking platforms easily. With the right amount of modelling, you can enrich those with the business logic that they usually lacked compared to the Google Analytics coherent model, being just a set of loose events. Sessionization, marketing channels extraction logic, user stitching — all of that can be done based on that loose set of events in plain SQL — plain SQL, which dbt made maintainable, organized and reusable. (View Highlight)

This data revolution is surprisingly quiet on the measurement side. Bar Snowplow, there’s not much of that movement there and data collection is still associated with ETLs and ingestion frameworks, leaving front-end data gathering (which is vital to so many businesses) aside. With free BigQuery export, the GA4 can still be of enormous value if used solemnly as a tracking platform for the BigQuery based warehouse. (View Highlight)

Integrations provided by Google are just not replicable regardless of how advanced your data stack is. (View Highlight)

And, most importantly, our “digital analytics” crop still has a unique expertise at the intersection of software engineering and understanding HTTP, marketing, conversion optimization and UX. There are not so many data engineers who are on top of the cookie game, understand the fragility of any data collection done in a browser environment, or know what the utms are and why those pesky things are so important. (View Highlight)

it’s always the IT who controls what lands in your Data Layer. Every single tag that you deploy is just code (even if it’s abstracted by a tag template). The GTM influences site performance and interacts with other components. It employs version control mechanisms coming directly from the software development world. Yet, correct me if I’m wrong, in many cases measurement is not integrated into development flow, creating this mini-IT outside the actual IT departments. That results in frustration, misunderstanding and broken tracking. (View Highlight)

Front-end data should eventually land in your warehouse (or “data mesh” if you’re unlucky) and most of the analysis should be done from there. Not in some out-of-the-box tool. Digital (web?) analysts should spend most of their time in Python/R/Excel/BI tools, just like the rest of folks with “analyst” in the title, not in Google Tag Manager (View Highlight)

Behavioral data from web and mobile applications is pretty complex. And usually it’s also pretty unique to every business, in a way that makes standard metrics just not enough. Analysts jumping between collecting the data, auditing it, maintaining marketing tools, probably won’t have time to invest in data exploration, applying models or just expanding their knowledge. So we’re stuck with funnels and bounce rates. (View Highlight)

I believe that digital analytics discipline should be conceptually decomposed. Or rather split into data collection as engineering task and web-related analytics as a purely analytical role. (View Highlight)

I see more and more former digital analysts moving into the data/analytics engineering side (yours truly included) or focusing solely on the usage of data. I see more and more setups where GA is a part of the data stack, not its competition. (View Highlight)