AI is only as right as what it knows about your business.
We build the context layer: one definition of every metric, one record for every customer and product, the documents behind them, and the rules on who may see what. Dashboards, analysts and AI agents read the same layer, so they give the same answer, with the source.
“What was gross margin last quarter, by region?”
Context layer
Azure
A context layer holds what your company means by its own data. We build the parts you are missing and connect the ones you have.
Every metric defined once, with a named owner, and reused by every dashboard, spreadsheet and agent.
One record for each customer, property, fund or product, matched across your systems, with how they connect.
What each table means, who owns it, where it came from and what depends on it.
Contracts, filings and procedures indexed, so an answer cites the passage it came from.
Who may see, use and change each fact, enforced the same way for people and for agents.
AI answers tested against known good ones before release, and every reviewed correction kept for the next question.
Governed data from the lakehouse below, your company's meaning in the middle, and every dashboard, copilot and agent reading from it. It runs in your own cloud account on AWS, Google Cloud or Azure.
In your cloud accountEach metric defined once, each customer, property or fund matched to one record across systems, and the documents and glossary behind them indexed with owners.
A golden record for properties, funds, investors, vendors and the chart of accounts; investor report prep fell 55%.
Read the case study
Here the agent answers the questions leadership asks of the numbers. It reads the context layer first: the definition of revenue and margin, the customer and product records matched across your ERP and CRM, and the rules on who may see what. Then it pulls the figures, checks them against the source, ranks the drivers and shows where each number came from. An analyst reviews the answer before it goes up. Switch who does the steps below to compare.
Software pulls, checks and explains the numbers; a person reviews the answer before it goes up.
Foundation
Cloud
among othersMost companies already have one of these parts. We start with the one that is breaking.
From our case studies, each figure as the study states it:
Reviewing a confidential information memorandum
Runs on
InsuranceProducing the weekly risk report
An agent will not catch a wrong number the way a person does. So the controls live in the layer, not in each tool.
An agent sees exactly what the person it works for may see, through the same row-level permissions.
Each figure carries its definition, its source table and when it was last refreshed.
Every metric and glossary term has a named owner who approves changes to it.
Everything is built on the lakehouse and cloud your risk and security teams already approve.
The layer between your data and the AI that uses it. It holds what your company means by its own data: one definition of each metric, one record for each customer or product, the documents and glossary behind them, and the rules on who may see what. Dashboards, analysts and agents all read from it, so they give the same answer.
A semantic layer defines metrics for BI tools. A context layer includes that semantic layer and adds what an AI agent also needs: records matched across systems, lineage and owners, indexed documents it can cite, access rules it must obey, and a record of reviewed answers. An agent will not catch a wrong number the way a person does.
Without one, an agent guesses what revenue means, joins the wrong customer records and cannot show where a figure came from. With one, it reads your definitions and rules first, answers from governed data and cites its source, so an analyst can check the answer in minutes instead of rebuilding it.
Six parts, starting with whichever is breaking: a semantic layer with one definition per metric, master data and entity matching, a catalog with lineage, owners and a glossary, document retrieval with citations, access policies that apply to people and agents alike, and evals that test AI answers before release. We build it on the lakehouse and cloud you already run.
It depends on how many systems disagree. Bridge Investment Group started with a two-month audit of every source system. Continental Properties ran its catalog, access model and dbt work as a 24-week stream, and Circle of Care went from strategy to a production rollout of one set of definitions in about 36 weeks.
Bridge Investment Group spends 55% less time producing investor reports on a golden record for properties, funds and investors. Robbins Brothers closes its books in half the time on a shared semantic layer. Tungsten Automation users spend 80% less time finding the right marketplace listing through a copilot on a knowledge graph.