The situation
The client had been running Tableau Server on-premise for seven years. The environment had grown organically — 340 workbooks, over 800 active users, and 60+ data sources feeding into extracts that had never been optimised. Average dashboard load time was 38 seconds. Several business-critical dashboards regularly exceeded 90 seconds.
The IT team wanted to decommission the on-premise infrastructure as part of a broader cloud migration. Tableau Cloud was the clear destination. But previous internal attempts to scope the migration had stalled because of the environment's complexity and the organisation's strict requirements around patient data handling and access controls.
The project came to us with a 12-week timeline, a non-negotiable requirement for zero data loss, and a secondary objective to fix the performance problems during migration rather than carrying them over to Cloud.
What the environment looked like
Before we touched anything, we ran a full environment assessment using the Tableau REST API and Metadata API. What we found:
- →
44% of workbooks had not been accessed in over 90 days. A large portion of the migration load was content that could be archived rather than migrated.
- →
Extract architecture was layered: many workbooks were pulling from extracts of extracts, creating refresh dependency chains that nobody had documented.
- →
Row-level security was implemented inconsistently — some workbooks used user filters, others used entitlement tables, some had no access control at all.
- →
18 of the 60+ data sources were unused by any workbook but were still being refreshed on schedule, consuming extract compute.
- →
The slowest dashboards were querying live connections to source databases without any query optimisation. These were the primary performance problem.
How we approached the migration
We scoped a parallel environment migration — the Tableau Cloud site ran alongside the existing Server environment for the final 6 weeks of the project, allowing us to verify parity before cutting over.
The first three weeks were spent on environment cleanup and data model remediation before moving a single workbook. We archived 148 workbooks that had not been accessed in 90+ days (with a 30-day restoration window). We removed 18 unused data sources. We rebuilt the data models for the 12 highest-traffic dashboards, converting live database queries to well-indexed extracts with proper aggregation. This alone eliminated 80% of the performance problems before migration started.
Access controls were standardised to Tableau's entitlement table approach with a single row-level security model applied consistently across all clinical and patient-adjacent content. This also resolved a compliance concern the IT team had raised but never actioned — some patient-adjacent dashboards had been accessible to users who should not have had access.
Migration was executed using the Tableau Migration SDK with custom scripts for access control verification, data source connection remapping, and post-migration validation. Every migrated workbook was tested against a defined checklist before user access was enabled in the Cloud environment.
The outcome
Migration completed at week 12. Zero data loss. Zero access control regressions. The Tableau Server environment was decommissioned two weeks after cutover once the validation period confirmed stability.
Average dashboard load time went from 38 seconds to 3.8 seconds — a 90% reduction. The 12 highest-traffic dashboards load in under 3 seconds. The dashboards that previously exceeded 90 seconds load in under 8.
The compliance access control issue was resolved as a byproduct of the migration work. The IT team decommissioned three on-premise servers, reducing infrastructure overhead. The analytics team reported that user adoption increased post-migration — slower dashboards had been a known adoption barrier that nobody had solved.