Check the query plan and the volume first: usually a full refresh where an incremental would do, a fan-out join from a non-unique key, or a missing partition/cluster filter. Fix by making it incremental, deduping the join key, or partitioning on the filter column. Measure before and after — it feels faster is not a diagnosis.
How to answer it
A twenty-fold slowdown is a change in the shape of the work, not a slow day on the warehouse. Find what changed before touching the SQL.
Diagnose, in order:
Volume: did the input grow suddenly? A backfill, a new source, a duplicated load. Compare row counts to last week.
Plan: look at the query plan or the warehouse's profile. The usual culprits are a join that fans out (a key that stopped being unique on one side), a full scan where a partition filter used to prune, or a spill to disk from a sort or aggregate that no longer fits in memory.
Materialisation: did the model quietly become a full refresh? A changed incremental filter or a missing is_incremental() block rebuilds the whole history every run.
Upstream: is the slow part actually a view being recomputed underneath this model?
Then fix the cause you found:
Fan-out: deduplicate the join key upstream and add a uniqueness test so it cannot recur.
Lost pruning: filter on the partition or cluster column explicitly, and make sure the filter is a literal or a simple expression the planner can push down.
Full refresh: restore the incremental filter with a lookback; test that incremental and full results agree.
Spills: aggregate earlier, drop unused columns, or increase the warehouse size for that model only if the work is genuinely bigger.
Measure before and after, in the same units, and record it in the PR. "It feels faster" is not a diagnosis, and "it is 40 minutes because the data grew" is only acceptable if the row counts say so.
What they are checking: that you look for the change, and that you can name the three usual causes.
Common mistake: adding a bigger warehouse. A fan-out join is 400 times the work, and no size fixes that.