Natural-language analytics with governed metrics

This worked application uses explicit planning assumptions to demonstrate FIELD. Quantities below are inputs, calculated implications or proposed targets; they are not reported customer results. The engagement decision remains conditional on the evidence described.

Planning basis and engagement boundary

F — Frame the outcome

Give analysts faster access to authorized, correctly defined metrics without turning natural language into uncontrolled warehouse work.

Value measure: Time and warehouse cost per accepted answer, with clarification and access-denial rates.

I — Investigate the workflow

Agree metric definitions and data grain with the business owner. Observe recurring queries, ambiguity and which cohorts users may access.

First evidence task: Test ambiguous revenue plus a one-to-many join against a small gold dataset.

E — Engineer the solution

Use governed semantics, query validation, read-only restrictions and byte/slot/time limits; size scans and joins separately from model latency.

The implementation assessment below makes the authority, workload, resource and failure assumptions concrete. Its numbers are calculations from the stated scenario, not a vendor performance guarantee.

Contract

Establish authoritative metric definitions, grain, dimensions, freshness and user scope. Ask what “revenue” means when recognized, gross and net revenue differ. Define an interactive deadline and resource budget; a 5TB warehouse does not mean every question should scan 5TB.

Implementation plan

Authenticate→resolve intent/metric→retrieve authorized schema/semantic definitions→construct query→validate→execute in a constrained read-only environment→check results→explain with assumptions. Use approved views/row and column policies, not arbitrary database access. Restrict functions/procedures and external access as well as write verbs.

Correctness

A one-to-many join can double-count order values. Aggregate child tables to the correct grain or use existence joins; compare against small hand-calculated fixtures. Syntactic validity and query execution success do not establish the business definition. For unknown metrics or unclear filters, request clarification before running a costly wrong query.

Sizing

A 5TB full scan has a physical lower bound of 500s at an illustrative effective 10GB/s, ignoring pruning, parallel architecture and cache. This demonstrates the need for partitions/materialized aggregates rather than asserting real warehouse throughput. Set maximum scanned bytes/slots, concurrent queries, result rows and execution time. Model generation and query planning/execution get separate latency budgets.

Failure/authority

A query for a forbidden cohort is denied by data policy even if the model produces valid SQL. Timeouts cancel warehouse work where supported and release permits; a slow query cannot monopolize all slots. Cache only within identical metric, data-version and authorization scope.

Evaluation/rollout

Grade result sets/numbers and permissions across paraphrases, ambiguous metrics, join traps and expensive scans. Shadow query plans before allowing broader access, then pilot known analyst workflows. Compare with a templated dashboard or parameterized-query baseline.

L — Launch with evidence

Pilot a small approved metric set against hand-calculated fixtures and existing reports. Measure correct task completion and query cost including clarifications.

Critical decision condition: Cross-scope data access, writes or incorrect metric grain block the affected capability.

Attach raw artifacts, the tested configuration, population, denominator, date and limitations to the release record. Missing or stale evidence holds the dependent decision. A favorable value forecast cannot offset a failed critical condition.

D — Develop ownership and reuse

The analytics owner maintains definitions and the warehouse owner operates quotas. New metrics follow an explicit semantic review.

Candidate reusable asset: Semantic metric contract and adversarial SQL fixtures.

Commercial decision and next commitment

Accountable decision owner: Analytics product owner with warehouse/data owner.

The initial commitment is the bounded journey above, with the stated exclusions. Confirm access, customer staffing and acceptance ownership during discovery; price and schedule unresolved dependencies explicitly. A new source, tenant class, language, action or operating envelope requires an impact review rather than silent scope growth.

Approve only the next exposure level supported by the evidence. The next useful investment is: test ambiguous revenue plus a one-to-many join against a small gold dataset.

Method and references

Apply the FIELD method, economic model and evidence rules. Methodological and technical influences are listed in the source register.