Enterprise Reliability
Case Study: Sales Analytics with Databricks Genie
Stan Konkin
ML Content Developer, DataCamp
From working to trusted
Space now serves all of Sales
VP wants reliability before full rollout
No new tables, just hardening
Reliability is a loop
Review
the conversations and ratings
Diagnose
what went wrong
Curate
the right lever
Verify
the fix in chat
Start by diagnosing in Monitor
$$
$$
$$
Monitor
tab logs real conversations
Filter to the thumbs-down answers
Read each generated SQL
Two failures, two causes
Monthly trend drifts → loose formula
"Premium"
undefined → Genie guesses
Different failures need different fixes
Fix one, lock the trend
Loose formula → lock with a
SQL example
Genie reuses the exact calculation
Same trend, every run
Fix two, define the term
Missing definition →
text instruction
Define the term once, precisely
Applied to every premium question
Keep it reliable over time
$$
$$
$$
$$
Pair questions with known-correct SQL
Re-run after any change
Catch a regression before leadership does
What makes a space production-ready
Locked formulas, clear definitions
Monitored and benchmarked
That's how a space earns trust
Let's practice!
Case Study: Sales Analytics with Databricks Genie
Preparing Video For Download...