Data tooling is not the problem, processes and people are

Summary — build tools that focus on driving behavioral, process change with Data teams.

Data tools have had their “Cambrian explosion” in the past decade. Each data tool is solving a new problem at a different layer — but IMO, the eventual problem of “data driven decision making” is solved less by a BI tool, but more by people and processes. We have more or less a similar set of core features at each layer of the wide variety of tools -

Creation > Ingestion > Warehouse > DE / Modelling > Consumption
where Consumption includes BI / Analytics, Experimentation and ML.

Every tool has some form of criticism, because usually teams do not end up using the way the creator built, envisioned and even validated them to be used. This is contrary to good product principles, but IMO, it is true! Data tools are complex concoctions and unlike Consumer apps, getting the interface and process “delightful” is really hard. It might be even unachievable if the “user” layer (aka DS, DE, DA maturity level) is the problem or where the users are not able to derive enough value.

Some hypothesis of what leads to success with data tooling (i.e. success = teams feel / believe / measure that they are a data driven or informed organization):

1. Forcing function for education for behavioural change as part of tool adoption and onboarding (e.g. looker modeling language, dbt for workflows).
2. High investment in “consulting” — e.g. sales engineering, contractors, onboarding to drive behavioral change. Not scalable — but works for some data products.
3. Filter for organisations/companies that are already data mature (e.g. airbnb — which has a principled data organization) but in this case they are likely doing well with their existing tooling.
4. Niches which have clearly defined outcomes e.g. funnel analytics with amplitude, crash analytics with crashlytics, sequence analytics with Interana.

#1, #2, #3 — are all pretty hard to achieve, which leaves us with #4 as the likely viable option for driving success for a data tool.

In summary — there are 2 broad categories of data tools — General purpose data tools and Behavior change data tools.

• General purpose data tools are massively adopted (as they are usually at the infra/platform layer — though not necessarily) and their success depends heavily on the organization’s data maturity. The tools have to be exceptionally good at what they do e.g. fivetran or airbyte need to have high coverage of supported sources and sinks or e.g. bigquery needs to be crazy fast with petabytes of data.
• Behavior change tools examples would include dbt (with their tagline being — “transforming data, transforming teams”), or any BI tool that enables a specific analytics need e.g. amplitude funnel analytics.

Note: Thanks to Cedric Chin (Eli James) for some seed thoughts and a bunch of other people who I discussed this thesis with.