Works with Snowflake Cortex Code

Ask Cortex Code why the number moved.

The FactorPrism skill takes Cortex Code from a business question ("why did weekly revenue drop?") to ranked, reconciled factors, inside your Snowflake account.

Already a Snowflake customer? Install from Snowflake Marketplace

Five-minute start

Install the skill, then ask in plain words.

In Cortex Code, paste the line on the right. The skill activates when your question matches, or call it with $factorprism-explain-change.

  1. Add the skill with the prompt, or from Cortex Code's agent settings.
  2. It checks for the app before touching any data, and asks before creating a read-only binding.
  3. Ask the question: the metric, the period, the source, and the dimensions your team manages by.

The skill is submitted to Snowflake Labs' skills catalog as a partner skill and awaiting review. Until it is listed, install it from the FactorPrism copy shown here. It works once the FactorPrism app is installed. View the submission.

Install the skill at https://github.com/davidrimshnick/coco-skills/tree/factorprism-partner-skill/skills/factorprism-explain-change

Then ask, for example:

$factorprism-explain-change Why did weekly revenue drop across region and product?
$factorprism-explain-change Explain the gross-margin miss versus last month.
$factorprism-explain-change Which payer and facility locations drove the denial-rate increase?
The missing tool for agents

The "why" is one call.

General-purpose agents write and run SQL well. Explaining a movement across region, product and channel at once, without counting anything twice, and tying it to the reported number, is a multi-step computation. FactorPrism makes it one callable procedure.

From Cortex Code, a Cortex Agent, a Snowflake task or an alerting pipeline, the same procedure returns the ranked factor table and saves every run. API reference.

CALL FACTORPRISM.API.RUN_DECOMPOSITION(
  source_ref    => 'SOURCE_VIEW',
  date_field    => 'ORDER_DATE',
  metric_field  => 'REVENUE',
  use_row_count => FALSE,
  rollup        => 'WEEK',
  hierarchies   => PARSE_JSON(
    '[["REGION","DIVISION","STATE"],["CATEGORY","PRODUCT"],["CHANNEL"]]'),
  period_start  => '2025-01-06',
  period_end    => '2025-06-30');
Built entirely on Snowflake

No outside services. One read-only grant.

Native App Framework

The whole engine installs into the customer's account. Versioned releases pass Snowflake's security review.

Marketplace

Install in a few clicks; billing runs through the existing Snowflake relationship.

Warehouse compute

Aggregation and analysis run on the customer's warehouse, inside the account.

References

Read-only grants on just the table or view the customer chooses.

Cortex

Optional plain-language interpretation of each Finding, inside Snowflake's boundary.

Agents and tasks

A callable procedure that Cortex Code, Cortex Agents, tasks and pipelines read back with SELECT.

Proof, not adjectives

Built to be automated.

  • It reconciles exactly. The ranked factors add up to the change, with no balancing figure, and the same input gives the same result every run.
  • Measured against known answers. A published blind benchmark of 280 runs. See the benchmark.
  • Asked for by a customer. Our first trial customer asked for the analysis as a callable procedure so they could wire it into their alerts pipeline as an automatic second-stage explainer.
  • Live on Snowflake Marketplace. Install, connect a source, and ask.
FAQ

Questions about the skill

What does the FactorPrism skill do in Snowflake Cortex Code?

It guides Cortex Code through checking the source, running FactorPrism, and explaining the ranked places in the business that drove a metric change.

Does FactorPrism need to be installed?

Yes. The skill guides Cortex Code; the FactorPrism Snowflake Native App does the analysis. Install it from the Snowflake Marketplace, or request a guided evaluation first.

Does customer data leave Snowflake?

No. FactorPrism runs in the customer's Snowflake account with read-only access to the one table or view the customer chooses.

Ask one real "why" question.

Bring a recurring KPI question and judge the result on a sample, a sanitized extract or a source you choose.