A governed semantic layer between BigQuery and Claude

600

dbt models

150

People using data

80–90%

Fewer tokens

After going AI-first without a semantic layer, Wunderflats is rolling out Steep so all 150 people get the same governed answer, every time.

Natasha Ellis-Knight

Content Marketing Lead, Steep

Wunderflats is Germany's No. 1 platform for temporary furnished housing. The Berlin-based company connects tenants and landlords for furnished stays of a month or more, across Germany and a growing number of European cities.

Industry: PropTech
Headquarters: Berlin
Visit: wunderflats.com

Background Going AI-first

Behind Wunderflats is a small data team, around 600 dbt models in BigQuery, and 150 people across the business who need answers from that data.

Earlier this year Wunderflats made a call a lot of companies are debating right now. They didn't renew Tableau and connected Claude straight to BigQuery instead, going AI-first without a semantic layer in between.

Challenge No shared definitions, no consistent answers

Without shared definitions, the same question could get different answers depending on how it was asked.

“We didn't have one definition of anything. Ask the same question twice and Claude could give you two different answers, because it was writing new SQL from scratch every time. And as we invest more in agentic development, one set of definitions goes from good to have to necessary.”

Manu Kumar, Senior Data Engineer at Wunderflats

Solution The governed layer between BigQuery and Claude

Wunderflats is now rolling out Steep as the governed layer between BigQuery and Claude. Every metric is defined once, in code, so all 150 people get the same governed answer.

“Much better than Claude Artifacts. Easier to share, easier to maintain.”

Behdad Nayebi, Business Intelligence Lead at Wunderflats