Offering · AI-ready data foundation

Data your people and agents can trust, starting with one workflow.

A faster platform does not make data usable, trusted or governed. We build the data products, shared definitions and controls the first workflow needs, on the platforms you already own.

Request a working sessionWhat you get

Call when

  • Two teams get different numbers from the same AI.
  • Your data is on a modern platform, but the business still does not trust it.
  • An agent needs definitions and sources it can cite.

You get

  • Certified data products for the workflow, each with an owner and a contract.
  • Shared business definitions the business can test: the semantic layer, or what the market now calls the context layer for AI.
  • Labels and permissions that travel with the data.
  • A governance model built from the center out, inside delivery.
  • Relevance, not just accuracy: every source has an owner, an effective date and a current or retired status, and retired material leaves what the AI can search.

How it runs

Length
Scoped to the first workflow, then extended product by product.
You bring
Data owners, your platform team and the definitions currently in dispute.
It ends with your decision
Certify, extend to the next workflow, or stop.
Fee
Fixed fee per phase, stated in a written proposal before you commit.

The foundation

Production AI needs information, controls and ownership the business can trust.

Access & orchestration

APIs, agent tools, identity and controlled interfaces

Data products

Governed, reusable information with owners and contracts

Business meaning

Semantic models and ontologies that define what the business means

Knowledge

Policies, research and expertise, made usable and traceable

Trust & control

Classification, permissions, lineage, evaluation, monitoring and audit evidence

Already in place

Cloud infrastructure, data platforms, warehouses and the models you already license

The brain

Meaning the business can test.

The brain turns raw data into answers the business agrees with. Each layer can be tested on its own.

Physical informationGoverned tables and documents in the platforms you already run
OntologyThe entities and relationships the business agrees on: a customer, a policy, a claim, a party
Tested viewsGenerated views over the sources, checked against known answers
Semantic modelsApproved metrics and definitions, versioned, with an owner
OrchestrationSelects the right definition and context for each request

The meaning layers give AI its context. They do not decide who may see what. Access is enforced underneath, at the data itself, and every search index built from that data filters each query by the user’s entitlements.

Labels and tags

One vocabulary, attached to the data itself.

Use one governed classification vocabulary. Propagate its labels to each approved serving path, verify the corresponding enforcement and measure drift and revocation lag. A label in a catalog alone does not restrict access.

How labels become enforcement

Two catalogs, two jobs

Know where the record lives, and where the rule is enforced.

JobThe system of record: what exists, what it means, where it came fromWhere access rules are enforced, at query time
ScopeEnterprise-wide: ontology, lineage, and the classification of tables and documentsThe data inside the platform, and every view derived from it
Exampledata.world (ServiceNow)Snowflake Horizon

An enterprise metadata catalog is not an enforcement point. Products named are examples, not endorsements. Richard has worked closely with several vendors named here as a customer and practitioner. No vendor reviewed or paid for this content.

What examiners and auditors ask

The data questions behind the review.

US banks

At large banks, BCBS 239 principles on accuracy and integrity, completeness, timeliness and adaptability, and the lineage supervisors use to test them, and traceable inputs behind every adverse action reason under Regulation B.

US insurers

An inventory of external data, algorithms and models, with oversight of the vendors who supply them, under the NAIC bulletin and Colorado’s rules.

Every AI capability

Which definition produced this number, from which version of which source, and who approved it.

Read next:

Governance

Governance is built from the center out.

Each ring depends on the one inside it, and a tool bought for the outer ring cannot supply what the inner rings have not established. That does not mean finishing every ring first: for a first workflow, build the slice of each ring it needs, together.

Ready for AIAgent identity, enforcement at the data, evidence a regulator can inspectTrusted analyticsCertified data products, a semantic layer, metrics people can defendGovernance as an operating modelNamed owners, decision rights, certification, release gatesData foundationOwned definitions and a catalogQuality rules, master and reference dataEvery outer ring depends on it
Governance as four rings: data foundation, governance as an operating model, trusted analytics, and ready for AI. Swipe sideways to see the whole drawing.

The Data and AI Operating Model

AI across the data lifecycle

AI proposes. The business keeps the decisions only it can make.

From discovering what exists to retiring what is no longer needed, AI can draft the definitions, mappings, tests and evidence. Meaning, rights to use a source, acceptance of a version and safe withdrawal stay with named owners.

AI proposesBusinessdecidesDiscoverrelated products,current-state mapoutcome, value,urgencyDefinedefinitions,contracts, examplesmeaning,authorityAcquiresource mappings,schema impactrights to usethe sourcesPreparetransforms,matches, testsidentity,reconciliationAssuretest and policyevidence packetsaccept a versionand its scopeServeapproved context,stated limitsconsequentialuseOperatetriage, affectedconsumers, fixespriority,restorationRetiredependency andretention packetsafe withdrawal,kept obligationsEach activity needs a defined interface, a named decision owner and a test of whether the automation helps.These are automation opportunities, not new certification states: an agent cannot certify by updating metadata.
Across the lifecycle, AI proposes and the business keeps the decisions only it can make. Swipe sideways to see the whole drawing.

Rewiring Data Management with AI

A product, not a project

A relabeled dataset is still a dataset.

A project optimizes for delivery. A data product optimizes for adoption, reuse and value that lasts after the delivery team moves on.

Confirm demand first

A draft contract, a sample or a catalog listing shows whether people will use the product before anyone builds pipelines. A governed product nobody adopts is the most expensive kind.

Owners measured on adoption and value

The owner answers for use, consumer experience and value as well as quality, and governance asks whether a product creates value, not only whether it complies.

Retire what is not used

Usage, satisfaction and maintenance cost decide whether a product is enhanced, consolidated or retired. Reuse is rewarded, so teams stop rebuilding their own copies.

Data Products in Practice Worked guide: a governed data change

Next step

Bring the numbers that do not agree.

A free sixty-minute session with Richard to examine one workflow, what is blocking it and whether there is a useful next step.

Request a working session