Change BI tools. Keep every number.
Migrations stall because nobody wants to rebuild 200 dashboards by hand, and nobody trusts the new ones until the numbers match. We use agents to do the rebuilding and people to sign off on every number.
- 220+dashboards in a Tableau-to-Looker migration, drafted by an agentic pipeline and checked against the originals.
- 46→9Looker models being consolidated in phases, with a check after each one that no rendered number moved.
- 0dashboard regressions recorded in the consolidation phases shipped so far.
- ~1,000dashboards in the Looker migration that started one of our biggest client relationships.
The dashboards are the easy part.
The hard part is everything under them. Calculations nobody documented. Fifty models that grew one request at a time. Refreshes that fail quietly until someone in finance notices. Move that as-is and you've paid to keep the same mess in a new tool.
So we fix the model layer as we go, and we don't call a dashboard done until it matches the original. If a calculation can't be translated one-to-one, an agent flags it and a person decides. Nothing gets quietly rounded off.
Production deploys stay manual. On purpose.
Five steps, from inventory to cutover.
Export and screenshot everything
A tool pulls every workbook out of the old platform, screenshots every view and writes a migration brief for each one.
Fix the foundation first
We consolidate and clean the models underneath, grouped by subject area, so the new tool isn't built on the old sprawl.
Agents draft, people decide
An agent proposes each rebuilt dashboard and validates it. Any calculation that needs a judgment call gets flagged for a person. After three failed correction passes, it stops and asks.
Compare against the original
A visual check compares the rendered result with the source, and we reconcile the numbers. A dashboard isn't done until they match.
Deploy by hand, then switch
We deploy to production manually. Once everything matches, your team switches over and the old tool can go.
From whatever you have to whatever you picked.
Dashboards
- Tableau → Looker
- Google Sheets → Omni
- Tableau → Power BI
- Looker model consolidation
Databases
- Access → Azure SQL
- App database → warehouse
- Snowflake, BigQuery, Redshift, ClickHouse
Transformations
- Scheduled SQL scripts → dbt
- Tests and lineage added on the way
After cutover
- Freshness and failure alerts
- Documentation your team can use
- Ongoing support, if you want it
Before you ask.
How long does a migration take?
It depends on how many dashboards you have and how tangled the model layer is. The inventory step answers both, so you get a real timeline after the first couple of weeks instead of a guess on the first call.
Do the agents deploy anything?
No. Agents draft and validate. Production deploys are manual, by design.
What if a calculation doesn't translate cleanly?
It gets flagged, and a person decides how to handle it with your team. We'd rather ask than guess.
Can you keep running things after the migration?
Yes. One of our biggest client relationships started as a migration.
Tell us what you're moving. We'll send back a plan.
Fill this in and we'll come back with how we'd approach your migration: what to fix underneath first, what to leave behind, and a rough timeline.
- How we'd sequence the migration
- Where the model layer needs work first
- What we'd need from your team
- A rough timeline and team size
We reply within one business day. No cost, no sales sequence. If we're not the right fit, we'll say so in the reply.
Rather just talk?
Thirty minutes. Bring the old tool, the new one and a rough count.
questions@kelihi.com