At my mid-sized US tech company we are adopting more and more of the Tableau ecosystem. Not just the notebooks or server, but there is a cloud service and I was surprised to learn some form of orchestration and data pipeline management system that everyone is learning as well.
Let me just openly say I dislike Tableau for reasons I’ll list at the bottom but I have an emotional reaction too and I’m writing this post to ask if I’m missing some great virtue of the Tableau way of providing data artifacts to data consumers. The business people and PMs who consume these tableau products seem to be happy with it.
Things I hate about Tableau:
- GUI driven
- Source code for transformations and calculated fields is inaccessible (locked up in GUI)
- Highly customized DSL which requires expertise in a specific vendor’s ecosystem.
- Again, a lot of clicking around the GUI all day.
Put it this way: Using Tableau means putting aside skills in software development and data engineering to become a “Tableau Engineer”.
I believe the buy-ins are:
- You don’t need to be a developer to use it (it’s better if you are not)
- It’s a paid, commercial product, and many companies see that as additional value
I don’t agree with neither of those, but it’s not easy to convince C_O to change their decisions…
Sympathize with @merlin’s sentiments and will add the following to @gvdr’s list
- Provide a robust platform to distribute and regulate access/security to reports according to user role in the enterprise.
For an analytics manager, that point is crucial. Products like Tableau, SAP, and PowerBI provide all three of the above. To some extent, Tableau and PowerBI allow integration of other languages, such as R, as ‘plugins.’ A data scientist at my organization is using an R package to provide run-charts to a user in PowerBI. It may be worth exploring what it would take to have Julia integrated via ‘plugin.’
How easy is the access policy regulation in something like Tableau?
Something like that can be achieved in Julia (e.g. with genie) but would require some “low-level” messing with user authentication, role management, … But, also, I don’t see how that can be done without that kind of low-level decisions. Do the peeps at Tableu or PowerBI have found a clever solution?
I have had the same questions from observations around Tableau adoption at my workplace, so keen to read how others navigate this cultural divide between encouraging use of open tools like Julia with a higher barrier to entry, compared to “locked in” data analysis tools.
Tableau is a good example of the long trend of the “democratization” of specialist skills. I’m sure FORTRAN programmers complained about Excel just as illustrators complained about PowerPoint. We specialists are not going to stop this trend. The trick is to identify where Julia is useful and Tableau is not. Tableau is very strong at sharing/communicating results interactively. It is not strong at analysis, modeling, etc. Those are things that Julia is very good at. If you accept that, you don’t have to read the rest of the post; it is just boring background.
I recently (well, in 2020) retired after 30 years in a science-y corner of a big industrial company. Tableau was used widely for sharing analyses and monitoring operational results, but we couched it as a communication tool more than an analysis tool. I think we landed on a scheme where most technical specialists still used their discipline’s “traditional” computing tools for really understanding their data and then, if that work had a broad audience, it might be cast into some Tableau form. Simply put, we tried to resist the one-size-fits-all mentality and tried to use the best tool for the job. If your work requires analysis capabilities that Julia has and Tableau doesn’t (likely), then you have a good argument for Julia. To be honest, I wouldn’t try to argue that Julia is better at sharing/communicating.
Thing is, there is a lot of technical effort spent developing these tableau/managed extacts that combine many tables and generate 100+ columns, plus dozens or hundreds of calculated field definitions.
Its a lot of technical mind power to create these dashboards, and its onerous to examine the lineage and definitions for all the components. Then our team has to maintain all these dashboards and change management is impossible.
I’m really confused why analytics managers are proponents.
Limited data maturity in the company. Most of those columns should be just available through a data catalogue and an internal API, with transparent coding. And plotted in Tableau if that’s their thing.
Helping a company develop data maturity Is possible, but not easy, because technology is only one component (and the easiest to change). The hard part organisational and cultural (and that’s hard to change).
I think this is fair.
I’ll add the data engineers are appalled by the burying of business logic in a dashboard and are pushing users to consume only DE team datasets to use tableau as a presentation layer only, and not as part of ETL pipelines. They seem to be losing that battle tho.