How To Avoid Tableau Chaos: Dos And Don’ts

By Alex Hossner, BI Consultant at BI:PROCSI

?CHAOS

Noun

A state of utter confusion or disorder; a total lack of organisation or order.

The definition of chaos may sound a little on the extreme side but it’s safe to say that if your Tableau implementation is in disarray, it will feel like chaos.

Tableau, in its essence, is a great tool for quickly looking at your data and should be implemented and maintained in a way that allows streamlined navigation and fast, accurate reporting.

This article highlights: chaos in Tableau = frustrated users and inaccurate reporting leading to mistrust in the tool, its contents and underlying contributors’ abilities.

How easy it is to get into chaos with Tableau?

At BI:PROCSI we’ve helped numerous companies that have found themselves in varying states of confusion and it’s an all too easy path to head down. At first, the challenges won’t be glaringly obvious and slowly but surely, they will rear their ugly head.

What do the consequences of chaos look like in Tableau?

All of the consequences of chaos are intertwined with each other, but can be addressed in the 4 topics:

Reporting inaccuracies e.g. a KPI has changed and users are still relying on outdated reports

Performance issues e.g dashboards take too long to load or data isn’t available in a timely manner

Frustration e.g users don’t know where to find reports or what the reports mean

Lack of trust e.g. consumers of reports don’t believe in the integrity of the data

Reporting inaccuracies

For the sake of argument let’s say your senior team access a dashboard that contains some high-level numbers showing the business pulse on a monthly basis.

This dashboard relies on some complex aggregations that are performed in Tableau and presented on the dashboard. A month ago the business agreed that the underlying KPI calculations are now different but these haven’t been factored into the senior level dashboard’s charts.

As these KPIs have been changed elsewhere in other dashboards, the rest of the business are using the correct figures but the leadership team’s are inaccurate.

As KPIs on strategic dashboards are often a mix of many other aggregations it’s vital that any changes are captured correctly.

Dos:

1. Establish a data dictionary that evolves and is constantly updated when any new, or changes to, metrics are introduced

2. Introduce a robust QA process for changes to KPIs ensuring that any dashboards consuming these are validated

3. Push logic down to a central source in the data warehouse which is maintained by data stewards

Don’ts:

1. Make changes to custom SQL without checking business impact. Another team may be using different logic to calculate measures

2. Overcomplicate KPI logic making it incredibly hard to unpick

3. Rely on Tableau for complex data blending - can easily become out of date if a metric changes in the business

Performance issues

Undoubtedly, the most noise from users will be around dashboard performance. This will vary from dashboard load time to dashboards freezing when applying a filter.

There are many factors that affect performance and these 10 dos and don’ts will help avoid performance chaos:

Dos:

1. Limit custom SQL in live connections as they can be inefficient. Where possible, create a view on the database server to implement your custom SQL and connect Tableau to your view

2. Keep it simple. Break large workbooks up into separate files, and consider using fewer sheets on a dashboard. Limit the number of tiles on dashboard and use drills to further explore the data

3. Limit the number of filters on a dashboard and use guided analytics and filter actions when possible. Every filter is an addition to query logic

4. Schedule extracts to fall in line with reporting needs. If a monthly report is accessed once a month, schedule that extract for once a month and where possible, use incremental extracts and optimise extracts by hiding unused fields

5. Index the most queried tables in your relational database

Don’ts:

1. Use useless clauses, e.g. ORDER BY. Tableau is going to sort the data once it’s loaded anyway

2. Show everything on one dashboard

3. Create queries that involve multiple fact tables. These will likely end up with lots of duplicate records resulting in weird results and slow performance or both

4. Allow your Tableau version to become outdated. Often there are performance enhancements in each release

5. Release dashboards into the wild without suitable QA checks. Have clear owners of dashboards/data sources and regular check-ins with users. It may be that a problem has been noticed and not been reported

Frustration

Performance issues and reporting anomalies will ultimately lead to user frustration but it won’t be limited to just these factors. Users crave a streamlined UX and if reports aren’t at their fingertips frustration will ensue.

Additionally, Tableau may be used to replace legacy reporting, which may have had functionality that isn’t available.

Without suitable guidance and education these users may try and source their insights elsewhere.

Dos:

1. Educate users on the importance of governance and create a data culture. Ensure the benefits of using a centralised BI tool are known throughout the business

2. Release and retire dashboards using well documented schedules and QA processes

3. Assign owners to reporting buckets who are responsible for of the tidiness of Tableau Projects

4. Have a dashboard catalogue or title page which describes what the purpose of the dashboard is, contains useful links and escalation paths

Don’ts:

1. Fire and forget. Every dashboard should be signed off by its owner or reporting authority

2. Allow users’ frustrations to go unheard. Create a channel for continuous feedback and improvements

3. Assume users know everything about the tool. Ensure regular training takes place with nominated power users

4. Promote overdesign. Teach the user base how to build slick and well performing dashboards

Lack of trust; a conclusion

A lack of trust will be the by-product of the primary 3 underlying factors that can cause chaos in Tableau.

If your users don’t trust the integrity of the data this will cast a huge black cloud over the whole deployment and nature of how reporting is perceived within your business. Who’s at fault? Do we trust the engineers building the models? Why are the dashboards slow and the data isn’t available when I need it?

Terrible feedback spreads significantly more quickly in the BI world than its counterpart: amazing feedback.

Strive for your users to be the best, and to use best Tableau practices in order for them to succeed by following these main steps:

Push as much logic down to your data warehouse as possible and document all calculations in a centralised data dictionary. Transparency is key here and many questions will be self-answered using this approach.

Create a data culture and ensure users recognise the importance of clean data, clear communication and good governance practices. Enable your users!

Train your power users to be the first port-of-call for issues surrounding dashboards. Use these guys as a channel for improvement suggestions, give them a voice.

Introduce a (non-restrictive) QA process that ensures dashboards have the same validation before release and also stick to a set of design principles. Dashboards and underlying data that fall outside of these are a lot easier to spot and rectify.

Keep it simple: dashboards, logic, models, instructions and life will be simpler, your users will be happy and reporting will be accurate.