You defined the metric once, carefully. The org still answers it three ways in ad-hoc SQL.

You built the dbt models, wrote the tests, and codified what “active,” “revenue,” and “churn” actually mean. But your governed definitions get bypassed the moment someone opens a warehouse tab or points an LLM at raw tables — and when two reports disagree, you can’t point to which one is authoritative. Spotonix is the AI analyst that interprets before it queries: it reads a business question, resolves it against the definitions available to the deployment and exposes the resulting plan before SQL executes—so you can review whether governed logic was used instead of discovering a fork after the fact.

The alternative

You built it right. Here’s where it still leaks.

Each of these is what happens when a governed definition has no path into the moment someone actually needs the number. The models are correct; the org just doesn’t reach them.

Bypassed by ad-hoc SQL

Your semantic layer is the thing they route around.

You codified the metric in dbt with tests and docs. When someone needs a number now, they open a warehouse tab or ask a chatbot that writes SQL against raw tables — and your definition is exactly the layer that just got skipped.

No authoritative report

Two dashboards disagree and both cite “the data.”

Marketing’s number, finance’s number, and the ad-hoc pull all differ. Your canonical definition exists, but nothing at query time enforces it — so you spend the week reconciling numbers instead of building models.

Logic that walks out

Accepted answers live in someone’s notebook.

The hard-won cohort logic and the “this is how we actually compute retention” nuance sit in a saved query or a Slack thread. It isn’t versioned, it isn’t reused, and when that analyst leaves, the reasoning leaves with them.

Cost of delay Every month the governed layer stays bypassable, more decisions get made on numbers that never touched your definitions — and your models become documentation nobody queries.

The outcome

Your definitions get used. Not replaced, not forked.

Where supported connectors make your definitions available, they can become primitives in the Context Graph and the visible analysis plan. Connector scope remains an explicit proof item.

01

Your models become the primitives.

Supported semantic definitions can become Segments, Calculations, and Analysis Patterns that Intent Algebra composes from. dbt and LookML ingestion are engagement-specific until evidenced live.

02

It asks instead of guessing.

When unresolved business meaning could materially change the analysis, Spotonix can surface the ambiguity. In Copilot mode, the visible plan remains yours to confirm before generation continues.

03

Governed meaning is inspectable.

The Context Graph represents domain concepts and relationships explicitly. Stable plan identity, versioning, persistence, and reuse should be tested for the workflow that requires them.

How it works

Connect supported context. Interpret, generate, and validate.

1

Reads your governed definitions.

Connect a supported warehouse and provide the governed definitions available for the deployment. The integration matrix states what is live, coming, or engagement-specific.

2

Interprets, then asks.

Intent Algebra resolves a business question into a visible plan built from your Segments, Calculations, and Analysis Patterns. Where it’s ambiguous against your definitions, it asks instead of hiding the choice inside generated SQL. In Copilot mode, query generation pauses for approval.

3

Generates, validates, and exposes the trace.

The generator receives the question and plan. Semantic checks validate material bindings before execution. Spotonix does not currently claim deterministic plan-to-query compilation; persistence and reuse are evaluated for the named workflow.

Representative questions

See which definitions each phrasing selects.

The plan makes selected Segments, Calculations, and Analysis Patterns visible. That gives your team a review surface; it is not a guarantee that every phrasing produces identical query logic.

  • “How many active accounts did we have last month?”

    Segment Active accounts Calculation Distinct count, MoM Pattern Period snapshot

  • “What’s net revenue retention for the enterprise cohort?”

    Segment Enterprise cohort Calculation Net revenue retention Pattern Cohort over time

  • “Which products drove the churn change quarter over quarter?”

    Segment Product lines Calculation Churn, QoQ delta Pattern Contribution to change

Proof

Watch the plan govern generation and validation.

The differentiator is a behavior you can watch: visible bindings, a Copilot acceptance gate, generated SQL, and semantic validation. Worked traces use the public TPC-DS schema and are illustrative.