Pretty. Unstable product + poor support = operations nightmare.

The reporting is pretty smooth. Management/administration is bad. Doesn't suit or use-case well.

"What support?"
Support barely exists at the "standard" (5 figure) price point. The SLA's are terrible. They won't help you expediently in an outage scenario. Their approach to any server problem appears to be "ask them for logs until they solve the problem themselves". For functionality questions they're somewhat useful, but it'll take you a few rounds (over 2 weeks) to get the answer you need.

Prices for better support are around the same as a full-time professional.

As a person responsible for administering their self-hosted product, I would never recommend this to a collogue. Fragile product + ineffective support = operations nightmare. It breaks frequently (upgrades, extracts, backups/restores, sometimes a service just stops during normal usage), and basic support (costing 5 figures/yr) with a 3 bizday follow-up SLA on downtime events is essentially useless for anything more than functionality questions (which take weeks to resolve). Stack restarts take over 15 minutes and are required for just about every configuration change. [I believe] the product was a poor fit for our use-case (add-on service for multitenant customer SAAS applications) for several reasons: - There's no supported method for changing datasource connections, which is a problem if you plan on publishing data sources from dev to prod environment without Tableau Desktop and manual effort. You'll have to make your own tooling to modify the XML directly (unsupported) - Trusted ticket implementation is secured via IP whitelist. If you want to use a modern SAAS product and not have to manage this IP whitelist (every change requires 15min restart), you'll have to write your own ticket-granting service over top of this API. - Security architecture offers poor segmentation between client databases. You can very easily provide customer A access to customer B's database. This danger is aggravated by the lack of a supported method for modifying data source connections in an automated fashion. - All users exist at a server level. This adds to the "poor segmentation between sites" argument, and also complicates things when you have 2 clients bound to different sites, trying to add the same username "jsmith". But hey, it's really pretty. Our product team bought it, infra team has paid for it since.