About the group
The group sells software to schools, and was formed when two education software companies merged. Between them they ran two warehouses: a BigQuery estate of 1,382 dbt models across six cloud projects, and a Snowflake warehouse.
Three Salesforce connections, HubSpot, Stripe, support data across Zendesk and Intercom, and training data all fed in, through about 60 Fivetran connections. Tableau and Power BI sat on top.
Two companies’ stacks, and almost nothing modelled for one of them
Each company had kept its own stack: two warehouses, separate definitions and two BI tools. Nearly every model joined Salesforce just to work out which school a record belonged to, and most of the people who had built the BigQuery side had moved on.
For the acquired company, the curated layer of the warehouse held two production models, both of them customer tables. Everything else was raw data. Meanwhile the Fivetran bill rose every month, from $2,548 in September 2025 to $18,901 in September 2026.
The group chose to rebuild only what people use
The data team decided against copying the old estate into Snowflake. Raw data lands in Snowflake first, the models people use are remodelled there, and everything else is dropped. The team kept dbt through the whole stack, with every change going through Git and its existing CI/CD path.
Find out what people use, overnight
Before anything moved, Rig audited both estates. Overnight it traced every Tableau workbook to the tables behind it, and found that 108 of 322 were still in use. The other 214 had never been opened.
Of 1,702 undescribed raw tables, 52% were empty, 18% were stale, and only 49 were queried five or more times a month. The rebuild came down to 309 of the 1,382 dbt models, plus a list of 46 unused sources to switch off.
“Overnight, Claude, Rig, and Tableau worked together to: find all 108 of the active Tableau reports; recreate in Rig as a data app from Snowflake; where they couldn’t, make a PR… That’s led to about 60% being rebuilt.”
Head of Systems and Data
Two days on site with the data team
Rig’s engineers sat with the group’s data team for two days. The data team raised its own pull requests through Rig. By the next morning, 1,220 models were building across the nine areas of the business the migration is split into, from product usage and AI to commercial, marketing and support.
In those two days, the acquired company’s curated layer went from two customer tables to tiers, subscriptions, free trials, content usage, active days, AI-assistant usage and training. The first three areas were about 85 to 90% done by the end of the second day.
In the first week, the Head of Systems and Data made more requests to Rig than anyone else at any Rig customer bar one, and one of the group’s analysts rebuilt the commission dashboard remotely before anyone at Rig knew about it.
“There is nothing it cannot do.”
Data lead
Fewer reports, and ones Tableau never had
Each Tableau report is rebuilt as an app on Snowflake and checked against the original before it replaces it, and the validated apps match Tableau to the decimal. Five per-year reports became one app with a tab for each year, product usage became one app in place of six, and the apps roll over to each new academic year on their own. 36 replacement apps are live.
With both companies’ data in one place, the group has built reports it never had before, including product usage for each of about 22,500 school accounts and an explorer for annual recurring revenue on Salesforce data.
Customer and training numbers were previously misleading
Those side-by-side checks also showed where some legacy figures had been misleading. Per-school training figures in the legacy reports had been inflated about 131 times by a join that multiplied rows. They have been rebuilt correctly.
A question about how many customers the group had came back as 43,245 from the raw data. The real number was 12,411. 98% of Salesforce accounts had no account type set, so the raw table counted every account ever created.
The team now keeps a list of every metric that gives a different result in Tableau and in the new reporting, so leaders can review each one and agree the true answer.
Governed for schools’ data
The group’s legal team required all processing to stay in the EU, so the AI runs on Claude through AWS Bedrock in an EU region. Sign-in is through the group’s own single sign-on for named users only, and student, staff and guardian data is kept out of what Rig can read.
The old commission dashboard showed each sales rep only their own pay. Rig shipped row-level filtering for the rebuilt version within a day.
Replace the pipelines in checked waves
The ten most expensive Fivetran connections were all software tools, from support and product analytics to code and CRM. Rig already covers 45 of the 60 connections the group reviewed, and Ingest will replace Fivetran in five planned waves. Each source is checked for parity, including deleted records, before its connection is switched off.