Simple effects hide at the wrong level
A regional cut can look harmless while one state, division, or product line drives the change. Unless you choose that level, the effect stays buried in the aggregate.
The manual investigation problem
You cannot find what you do not know to slice.
Even a simple factor may sit at one specific level: state instead of region, SKU instead of product. You find it only if you drill to the right place. Add independent dimensions and their intersections, and the number of plausible cuts grows too quickly to search one by one.
The total hides opposites
A broad lift can conceal a local decline. A beat and a miss can cancel. “Flat overall” can contain two urgent, opposing stories.
The factor lives in a combination
Region may look normal. Product may look normal. The specific region × product intersection can still contain the decisive movement—and no separate one-dimensional cut can isolate that answer.
Example: South Jewelry sales doubled
Was the movement a company-wide lift, strength across the South, growth in Jewelry everywhere, or something specific to South × Jewelry?
Company-wide movement?→Would appear across products and regions
South-wide movement?→Would affect other products in the South
Jewelry-wide movement?→Would affect Jewelry in other regions
South × Jewelry factor?→Would exist only in this combination
A manual search has no reliable stopping point: every explanation can fit part of the visible pattern, and you cannot know whether an unchecked level or intersection contains the real factor.
The FactorPrism approach
Search the full problem, not one cut at a time.
FactorPrism examines the relevant levels and intersections in one analysis. It searches beyond the hypotheses an analyst happened to think of, finds a compact set of factors that explains the observed movement, locates each one where it acts, and measures its contribution.
- Broad factors stay broad. A movement affecting the whole business is reported once instead of smeared across every segment.
- Local factors stay local. A factor specific to one region, product, or meaningful intersection appears at that level.
- Gains and declines remain visible. Offsetting movements do not disappear inside the rolled-up total.
- The result closes. Contributions reconcile to the reported change, giving the team one answer to inspect.
Under the hood
Separate the signals before explaining the story.
FactorPrism draws on methods from signal processing, adapted for business metrics, to separate overlapping movements across dimensions and their intersections.
Measurement comes first.
The analytics engine determines where the movement occurred and how much each factor contributed before AI helps communicate the result. The narrative follows the measurements; it does not invent them.
The answer remains reviewable.
Each Finding links the headline to ranked factors, timelines, business locations, comparisons, prior runs, and a focused Investigation Plan.
The practical difference: you do not need to guess every useful cut in advance, and you do not need to hand-build an explanation from a pile of overlapping percentages.
From question to Finding
Three steps, one repeatable investigation.
Choose the governed source
Grant read-only access to one Snowflake table or view and select the metric, time grain, and business dimensions.
Define the movement
Choose the comparison window or plan column that represents the business question the team needs to answer.
Open the Finding
Review the measured factors and evidence, then carry the strongest signals into focused next investigations.
Start with a question that is still unresolved
What metric is your team still slicing?
Bring one real business surprise to a guided evaluation. We will test which factors FactorPrism can surface automatically across the dimensions.