The question your exec actually asked was "why did the number move?" — not "which segments look interesting?" Here's where each tool fits. Both Snowflake Cortex TOP_INSIGHTS and FactorPrism live in your warehouse and help you investigate a moving metric — but they're built to do different jobs.
Cortex's TOP_INSIGHTS (and the Contribution Explorer experience built on it) is a genuinely useful, native part of Snowflake.
If your job is exploratory triage — "show me the segments that stand out" — Cortex TOP_INSIGHTS does that well, and it's right there in the platform you already pay for.
The trouble starts when "interesting segments" has to become "the answer." Three gaps tend to show up.
FactorPrism is built specifically to close those gaps: an exact decomposition, reconciled to the total, down to factor intersections, in your warehouse, in seconds.
| Snowflake Cortex TOP_INSIGHTS | FactorPrism | |
|---|---|---|
| Exact reconciliation to total | Ranks/surfaces notable segments; not designed to sum exactly to the observed change | Every contribution reconciles exactly to the total move — zero unexplained residual, no balancing plug (verified: residual 0.00e+00, 13/13 regression tests) |
| Intersection-level attribution | Primarily single-dimension surfacing | Attributes down to factor intersections (e.g., region × product × channel) |
| Runs in-warehouse | Yes — native Snowflake, no data movement | Yes — Snowflake Native App; SELECT-only access to one table/view, zero external network calls |
| Output you can defend to an exec | A ranked shortlist to investigate | A reconciled waterfall: "these factors sum to the full move" |
| Setup | Native Snowflake function | Native on Snowflake; standalone desktop app (Win/Mac/Linux) for non-Snowflake teams |
| Best for | Exploratory triage — where to look | Root-cause answer — exactly why, ready to present |
This isn't either/or. A clean workflow:
Same warehouse, sharper math, a total you can defend. FactorPrism complements Cortex; it doesn't replace it.
Doesn't Cortex TOP_INSIGHTS already do this — why would I add another tool?
TOP_INSIGHTS is excellent at surfacing interesting segments to investigate. It isn't designed to produce an exact decomposition that reconciles to the total or to attribute down to factor intersections. FactorPrism does exactly that, in the same warehouse. Most teams use Cortex to find where to look and FactorPrism to produce the answer they present.
Does my data leave Snowflake to use FactorPrism?
No. FactorPrism runs as a Snowflake Native App: your data never leaves your Snowflake account. You grant read-only (SELECT-only) access to a single table or view, and the entire analysis — aggregation and the AI-written narrative (via Snowflake Cortex) — executes inside your account, on your warehouse, under your existing Snowflake roles and governance.
What does "reconciled to the total" actually mean?
Every factor's contribution — including intersections like EMEA × Enterprise × Paid-Search — sums exactly to the observed change, with zero unexplained residual and no balancing plug. Small drivers beyond the top ones are grouped into a labeled "Other smaller factors" rollup — itself a sum of real attributed drivers, not a fudge factor.
Is this just AI guessing at drivers?
No. The decomposition is a math result (a log-mean/Divisia decomposition) that adds up to the total exactly. AI assists the experience and writes the narration, but the answer is an exact, reconciled attribution — not a generated guess.
Why did the number move? Answered — exactly, in seconds, in Snowflake.
