Natural-language operations dashboard for a logistics company
Ask an operations question in plain language; get the query, the chart, and the answer — without waiting on the data team.
Four analysts, three hundred askers
Every operational question — “which routes ran late last week?” — became a ticket for a four-person data team. Answers took hours to days; many questions simply went unasked.
The cost of leaving it alone
Ops managers made daily decisions on stale numbers. The data team spent its time on repetitive queries instead of the modeling work it was hired for.
Text-to-SQL with guardrails
A Slack-native copilot that translates questions into governed SQL over Snowflake, renders charts, and always shows its query.
- Semantic layer of vetted table and metric definitions — no raw schema guessing
- Generated SQL runs read-only under a warehouse resource cap
- Every answer includes the query, so analysts can audit and correct
- Unanswerable questions route to the data team with context attached
Stack: Claude · Snowflake · Slack
How it was built
- Week 1–2: semantic layer built with the data team over the 40 most-asked question patterns
- Week 3–4: Slack app, chart rendering, and query audit trail
- Week 5: pilot with ops managers, correction loop, then company rollout
What the numbers say
What happened next
The correction loop matters most: when analysts fix a generated query, the fix updates the semantic layer. The system gets more trustworthy with use, and the data team finally ships models.
This system is an example of Fractional Technical Operations Support work.
Need a similar system?
Let's talk through your version of this — same architecture thinking, scoped to your operations and tools.
30 minutes · no pitch deck · reply within 24h if you write instead