One source of truth — for every dashboard, app, and AI agent
Define each business metric once, in code. Humans, applications, and AI agents consume it through the protocol that fits them — so the same question never returns three different numbers, and no AI ever writes raw SQL against the warehouse.
How it works · Universal Semantic Layer
One source of truth
Metrics-as-Code · Git-versioned definitions
Analytics SL
SQL · ODBC / JDBC
API Supergraph
HTTP · GraphQL
Why it matters
“Where are we today on EMEA bookings against commit?”
Three systems answer within ten minutes — $248M, $229M, $261M — each “correct” against a different definition. Every day we can’t trust our own pacing number is a day of strategic latency.
With the semantic layer
$243M · 61%
One governed number — pacing 3 points behind, traced to bookings_net_emea v4.2.
Why not just centralize the data?
M&A velocity
A new acquisition every 1–2 quarters — and each data migration runs 12–18 months. By the time one lands, the next deal has closed.
Diverse workloads
Hardware telemetry SaaS apps services billing. No single database is optimal for time-series, relational, and transactional at once.
Data sovereignty
GDPR · PIPL · DPDP · US state laws. Some data is legally forbidden from leaving its region — centralizing it isn’t even allowed.
Our data will never be physically centralized — so we stop trying. We centralize something far more powerful: meaning.
Foundation — state
An immutable, vendor-agnostic bedrock of metric definitions in YAML. It defines and stores; it never executes — enabling CI/CD, testing, and peer review of business logic.
Execution — compute
Interchangeable engines — BI tools, AI agents, NLQ — read the YAML and compile it to dialect-specific SQL. They hold no state, so execution can innovate without risking core logic.
The API track
The API Supergraph
A single GraphQL gateway that uses entity federation to stitch legacy REST APIs, microservices, and the Analytics SL into one unified schema. To an app — or an AI agent — there is only one API, and one version of the truth.
One schema, many sources
Entity federation merges legacy REST APIs, M&A systems, microservices, and the Analytics SL behind a single typed endpoint — no point-to-point pipelines.
The only safe tool for AI
LLMs can’t reliably write SQL against thousands of tables, but they reliably call a typed API — so agents get governed data with no raw DB access and no hallucinated numbers.
Zero-friction M&A
Wrap an acquired company’s API as a subgraph and attach it — its data is live to BI and AI agents in days, not a 12–18 month migration.
One query, planned & stitched
From raw data to governed meaning
Governance is built in, not bolted on
Access
ABAC via Open Policy Agent — field-level masking row-level filters, with Git-versioned policies.
Lineage
“Change revenue.sql — which 47 dashboards and 12 AI prompts are affected?” answered from OpenLineage.
Trust
PII classification, per-metric SLAs (freshness · P95), and a regulator-ready audit trail by default.
Platform selection · POC
Choosing the engine
Four candidates scored across 11 pillars — governance, performance, AI readiness, and persona-fit — by building one shared “Golden Metric” (e.g. revenue) in each tool.
Cisco constraint: cloud data stores may not connect to on-prem stores — on-prem deployment may be the only path, which shapes the down-select.
The payoff · baseline → target
“Revenue” definitions in dashboards
AI hallucination on metrics
Quarter-close cycle
M&A data integration
