Looker Data Catalog Data App: See The Definition Behind Every Answer
If you ask an AI assistant why repeat purchase rate fell last quarter, you’ll usually get an answer and often a chart to go with it, but not the detail behind it: which metric it used, how that metric is calculated and which business rules were applied. Without that context, people either take the number on trust or go looking for someone who can explain it.
I wanted to see how far we could close that gap with Looker Data Apps, so I built a prototype data catalog for our own Looker instance at Rittman Analytics. It’s a working example of the kind of tool we can now build inside a client’s Looker, and in this post I’ll explain why I built it, what it does and how I put it together.
Why it was built
The idea came from client work. On one project, for example, a stakeholder wanted to see every metric, how it was calculated and which dashboards used it. My first answer was a long Google Doc, which did the job, but few people could see it and parts of it went out of date as soon as the model changed.
Around the same time, another client was thinking about buying a separate data catalog tool. I suggested building one inside Looker with its new Data Apps feature instead, which would mean no extra software to pay for, metadata kept in the client’s own Looker and BigQuery, and everything in a tool their team already used. The catalog in this post is the internal version I built for Rittman Analytics, and Data Apps made it straightforward to build.
A data catalog inside Looker
The catalog runs as a Looker Data App. Google released Data Apps into preview in October 2026: they’re custom web pages that run inside Looker on top of its semantic layer, and we covered how to build them in Data Apps and Custom Dashboards Come to Looker (and Wire). Because it runs inside Looker, there’s no separate platform to manage, and people sign in with the Looker login and permissions they already have.
The prototype covers four areas:
- Metrics and definitions: what each metric means and how it’s calculated.
- Explores: the tables, dimensions and calculated metrics in each explore.
- Warehouse schema: every table and column in the warehouse.
- Data lineage: how data flows from source tables to the metrics in an explore.
Why definitions matter more with AI
Looker sits between the data warehouse and the people who use the data. Analytics engineers define each metric once in LookML, so every dashboard, explore and report works from the same definition, and that matters even more once AI is involved: Google’s own testing found that LookML-governed queries reduced generative AI data errors by as much as two-thirds compared with querying warehouse tables directly.
In larger organisations, three groups usually work with this layer: analytics engineers, who build and maintain the warehouse and the LookML models; data analysts, who explore the data and build reports; and business analysts, who turn business questions into data requirements and insight. They all depend on the same definitions, but only the analytics engineers can easily read them, because business users see little more than the field labels and descriptions in Looker’s explores.
AI tools are making it easier for business analysts to answer their own questions. Conversational Analytics in Looker, which we’ve covered before, lets people ask questions in plain English and get a chart back in seconds, and we’ve also shown how to connect Claude to Looker using Google’s MCP Toolbox and query Looker through LangChain agents. What these tools don’t always show is which tables a metric comes from or how it was calculated, and when two metrics have similar names, like net_revenue and net_sales, it isn’t obvious which one the assistant picked or whether it was the right one for the question.
The catalog is my attempt at closing that gap. It gives business users the same definitions our analytics engineers work from, without anyone needing to open LookML, so they can check where an AI-generated number came from and whether it’s the right metric for what they’re asking.
How it was built
The catalog has two parts: a Python build script that gathers the metadata, and the Data App itself, a JavaScript app that Looker hosts and serves to users.
The build script reads the LookML repository, the dbt project and BigQuery’s own table metadata (columns, sizes and when each table was last updated), then maps every table, field, metric and dashboard and how they connect. It adds the glossary, business rules and verified questions, which we write by hand in YAML and check against the code, along with usage figures from Looker’s System Activity. A full build is a single command:
python build_metadata.py \
--config catalog.yml \
--parse \
--refresh-warehouse \
--source apps/data-catalog/src/app.src.js \
--inject apps/data-catalog/app.js
The --refresh-warehouse option runs read-only metadata queries against BigQuery, and --inject writes the finished catalog into the app file, which I keep under 900 KB. A daily job then loads each build into BigQuery, and the app reads the latest one through Looker, falling back to its own built-in copy if it can’t reach BigQuery.
Everything specific to one organisation, such as the logo, colours, glossary and business rules, lives in a config file and a few YAML files, while the app code stays the same. I built and published it with Wire, our Claude Code plugin for delivery work, using Wire’s integration with Google’s Data Apps toolkit to push and check each version.
- Claude Code:
/plugin marketplace add rittmananalytics/Looker-data-catalog, then/plugin install looker-data-catalog@looker-data-catalog - Claude.ai: download the zip from the v1.0.0 release and upload it as a custom skill
- Manually: follow
skills/looker-data-catalog/SKILL.md
What it changes for business users and analytics engineers
Business analysts don’t usually write LookML, but the catalog shows them enough of the model to make specific requests. Where a request might once have been as vague as “Can we track repeat purchases?”, an analyst can now look up the orders table, see which columns it has and ask for something much clearer:
“I need a repeat purchase rate metric. The
orderstable already hascustomer_idandorder_date. Could we add an order sequence number column?”
That gives the analytics engineer a clear starting point and cuts out most of the back-and-forth. The description they then write on each new LookML field appears in the catalog, and Conversational Analytics uses it when it picks a field to answer a question, so the new metric arrives already explained.
Business rules work in a similar way. Questions such as what counts as a sale, when a customer is “active” or how refunds are handled are usually answered in Confluence pages, Google Docs or someone’s head, and those answers don’t always keep up with the model. In the catalog each rule sits next to the metric it applies to, so someone wondering whether a revenue figure excludes refunds can look it up instead of asking. Here are some examples of the rules it can hold:
| Business rule | Governs | Example definition |
|---|---|---|
| Sale | net_revenue |
An order with status complete, excluding test orders |
| Active customer | active_customers |
A customer with at least one order in the last 90 days |
| Refund treatment | net_revenue |
Refunds are deducted in the month the refund is issued |
“Which explore should I use?” is another question that comes up all the time, and being able to browse every explore with its metrics and dimensions helps people pick the right explore or data agent before they ask Conversational Analytics anything.
What this shows
To be clear about what this is: the catalog is a prototype built on our own data, and we’re not offering it as a product. A version for a client would be shaped around their data model, their terminology and the questions their teams actually ask.
What it does show is what’s now possible inside Looker with Data Apps and some careful engineering: tools that sit in the platform people already use, read from the same semantic layer as everything else and don’t need another subscription. We’ve already used the same approach for a client’s metric explorer with live figures, and it’s the kind of work we can take on as part of a Looker or analytics engineering project.
Interested? Find Out More!
Rittman Analytics is a boutique data analytics consultancy that helps ambitious, digital-native businesses scale up their approach to data, analytics and generative AI.
We’re authorised delivery partners for Google Cloud along with Oracle, Segment, Cube, Dagster, Preset, dbt Labs and Fivetran and are experts at helping you design an analytics solution that’s right for your organisation’s needs, use-cases and budget and working with you and your data team to successfully implement it.
If you’re looking for some help and assistance with your AI initiative or would just like to talk shop and share ideas and thoughts on what’s going on in your organisation and the wider data analytics world, contact us now to organise a 100%-free, no-obligation call, we’d love to hear from you!