Frequently Asked Questions

Semantic layers define what your metrics mean. Fig maps what causes what — and enforces those definitions and rules on every AI agent in your stack. They solve different problems. You can keep your semantic layer and add Fig as the decision and governance layer on top of it. Fig's agents will use your existing definitions as input when available.

Cortex and Genie are AI features inside their respective platforms — they translate questions to SQL using that platform's data. They don't have your business's approved metric definitions, causal model, or policies. Fig is an installable skill that gives any AI agent — including those tools — the business context to make governed, consistent decisions using your definitions and rules.

No. Fig reads your warehouse as-is. There's no data modeling required upfront, no tables to create, and no schema changes. Fig's agents analyze what's already there and build the knowledge layer on top of it. You can connect and have a working initial model the same day.

Yes. Each organization can connect multiple warehouse destinations — for example, a BigQuery production environment and a Snowflake environment for a specific business unit. Fig normalizes data across sources and resolves entity conflicts (the same customer appearing in both systems) automatically.

Fig exposes a Model Context Protocol (MCP) server. Any MCP-compatible AI agent — Claude, Gemini, OpenAI's agents, or a custom internal agent — connects to it by adding Fig's endpoint to their configuration. Once connected, the agent gains access to your approved metric definitions, business policies, and causal model. No changes to the agent's code.

Fig sits between the agent and your data. When an agent queries Fig, the response is already governed — it uses your approved definitions, respects your policies, and returns only what the policy allows. An agent can't return a metric using an unapproved definition because Fig doesn't serve unapproved definitions. The governance is enforced at the data layer, not the prompt layer.

Yes. Many teams start with Fig as the governed analytics layer for their human operators — defining metrics, mapping causal relationships, and using Flows to automate recurring analyses. The AI agent skill is an add-on, not a prerequisite. You get full value from Fig before a single AI agent connects to it.

You assign named owners to each metric — typically the person or team responsible for that number (finance for revenue metrics, marketing for CAC, and so on). They review and approve proposed definitions. The approval step is intentionally human — the goal is that the right person has explicitly signed off, not that a system auto-approves on their behalf.

No — and that's the point. Fig enforces one approved definition per metric across the organization. If two teams have different views on how a metric should be calculated, that conflict is surfaced and resolved through the approval process rather than silently producing different numbers at different meetings.

Any change creates a new version. The previous version is preserved and auditable. Every change shows who made it, who approved it, and when. Agents and Flows automatically use the current approved version — you don't need to update them manually when a definition changes.

Most teams connect a warehouse, review the initial metric map, and approve their first definitions within a day. Flows and monitoring can be set up in hours. The causal model deepens over time as your team enriches it, but you're not starting from a blank slate — Fig builds the initial model automatically from your data.

No. The model evolves incrementally, not rebuilt from scratch. Adding a new metric or relationship is a single change. When something significant happens — a new product line, a reorganization — you update the affected parts of the model. Version history means you can always see what the model looked like before and after any change.

You own everything you've built in Fig. Before canceling, you can export your metric definitions, business policies, and relationship structure in standard formats. Nothing is locked in a proprietary format that only Fig can read.