The AI Context You Already Paid For

Bruno Billy, president and CEO of APGAR North America, has put an inconvenient claim in front of every executive about to sign for another data layer: most enterprises already own the technology needed to give AI context, and simply do not operate those capabilities as one coherent system ^[1][2]. If that holds, the line item that matters is integration and decision rights, not procurement.

The scaling numbers he cites are not his own. McKinsey researchers recently reported that fewer than 10% of companies have fully scaled AI, and that more than two-thirds of high-performing companies name data as the primary obstacle to enabling it ^[3][4]. That leaves more than 90% of companies short of full scale ^[5]. KPMG researchers landed in the same place, with 58% of executives citing data readiness and access as the biggest challenge to deploying AI agents ^[6]. Neither figure describes a missing product category.

Billy breaks context into three parts. Semantics, or business meaning, sits in business glossaries, governance tools, master data models, semantic models, policy repositories and data dictionaries ^[7][8]. Operational state sits in master data management platforms, hierarchy and relationship management, reference data and stewardship workflows, data quality rules and exception management ^[9]. Traceability sits in catalogs, metadata management, lineage tools, stewardship histories, governance workflows, approval records and audit trails ^[10]. Most large organisations have bought into all three columns already, usually to serve governance programs, regulatory initiatives or transformation projects ^[11].

The vendor answer to a context gap is a new layer built for the agentic era, and Billy concedes some of those tools may be useful or even necessary ^[12]. His objection is sequencing. Because each capability was bought for its own program, business definitions end up in one platform, stewardship processes in another, lineage somewhere else, and operational rules buried inside applications ^[13]. Each piece produces value alone; AI needs them working together ^[13].

The part that cannot be purchased is the operating model. Billy argues that definitions and metadata alone cannot keep context accurate, consistent and trustworthy, because definitions change, source systems disagree, policies evolve and business rules collide ^[14][15]. Someone has to decide which definition applies, which source is authoritative, which relationship matters and how exceptions are handled, and those decisions are what let context support models reliably ^[15]. That is a question of people, processes and decision rights, not licences ^[14].

His prescription is an inventory that goes further than counting tools: establish where business meaning, operational state and traceability actually live, whether they are concentrated in a few trusted systems or scattered, which sources are authoritative, where conflicts exist, and how the pieces could connect into a common context layer ^[16]. Then attach a cross-cutting operating model with common ownership, decision rights and change processes so context can be produced, made accessible to AI systems, governed and improved over time ^[17]. Understanding alignment and gaps, he writes, may be the best and most cost-effective place to start ^[18].

Worth noting the incentive: this is an audit-first recommendation from a firm that advises on enterprise data programs ^[1]. The survey evidence he leans on comes from elsewhere ^[3][4][6], and it does not distinguish between an enterprise that lacks tooling and one that has never made its tooling agree with itself.