Enterprise AI projects have a quiet failure mode that does not show up in the demo. The Copilot looks impressive on screen. The agent gives fluent, confident answers. The pilot goes well. And then, when the tool reaches real employees working on real problems, the answers are subtly but persistently wrong.

Not wrong in an obvious way. Wrong in the way that only becomes visible to someone who already knows the answer. The agent recommends the discontinued product. It pulls figures from the old financial model. It suggests contacting the person who left the company eight months ago. It gives an answer that is technically grounded in real company data but misses the business reality that any experienced employee would immediately understand.

This is the context gap. And it is the primary reason that enterprise AI programmes continue to underdeliver despite access to genuinely capable models.

Microsoft announced its answer to the context gap at Build 2026: the Microsoft IQ stack. Understanding what the context gap actually is, and how each component of the IQ stack addresses a specific dimension of it, is the most practical way to evaluate whether these tools matter for your organisation. This article explains both.

What Is the Context Gap?

The context gap is the difference between what an AI model knows about the world in general and what it needs to know about your organisation specifically in order to be useful.

Every large language model is trained on vast amounts of text. It has broad knowledge of business concepts, industry practices, writing conventions, and reasoning patterns. What it does not have is any knowledge of your organisation: how your teams are structured, what your revenue figures actually mean, which documents are current and which are superseded, who the right person to involve in a given decision is, or how a word like 'customer' is defined inside your specific business.

When an agent is deployed into an enterprise environment without that organisational context, it fills the gap with its best guess. It reasons from general knowledge rather than specific knowledge. The outputs can sound convincing while being subtly disconnected from operational reality.

The hard part of enterprise AI is no longer the model. It is the context. Agents that start from zero every time, with no shared understanding of how an organisation works, cannot reliably reason, coordinate, or act.

Research consistently confirms the scale of the problem. MIT's NANDA State of AI in Business study found that 95% of enterprise AI pilots deliver no measurable business impact, even as organisations invested tens of billions in them. The researchers did not identify this as a model quality problem. They identified it as a context and alignment problem. Gartner has projected that 60% of AI projects will be abandoned through 2026 due to a lack of AI-ready data and context.

The context gap has three distinct dimensions, and resolving it requires addressing all three.

• The data dimension: Does the agent have access to the right enterprise data at the moment it needs it, in a form it can reason over?

• The meaning dimension: Does the agent understand what that data means in the context of your specific business, including your terminology, definitions, and the relationships between your entities?

• The human dimension: Does the agent understand how work actually happens in your organisation: who the relevant people are, what the current priorities are, and how collaboration and decision-making flow?

Each of the three IQs in Microsoft's stack addresses one of these dimensions. That is not a coincidence. It is what makes the architecture worth understanding in concrete terms.

Why Existing Approaches Have Not Solved It

Organisations have tried to close the context gap through various means, and each approach has hit a ceiling.

Retrieval-augmented generation, commonly called RAG, allows agents to search a document store before answering. It helps. But traditional RAG treats every document equally, cannot reason about relationships between documents, requires custom integration for every new data source, and does not understand your business definitions. An agent that can retrieve a document is not the same as an agent that understands what the document means in the context of your operations.

Uploading documents to a knowledge base gives agents access to specific content, but it does not give them a map of the business. The agent knows what is in the policy document. It does not know that this policy was superseded by a newer version, that the team responsible for enforcing it restructured last quarter, or that the figures it references use a different revenue definition than the one your finance team currently applies.

Fine-tuning models on company data is expensive, requires continuous maintenance as data changes, and still does not produce the real-time, permission-aware understanding of live organisational context that agents need to operate reliably.

None of these approaches fail entirely. They all deliver partial value. But they all leave a version of the context gap open, because they address the symptom (the agent does not have the right information) without addressing the underlying cause (the organisation does not have a shared, governed, machine-readable context layer that agents can draw from).

That is exactly what Microsoft IQ is designed to be.

What Microsoft IQ Actually Is

Microsoft IQ is not a product you install or a feature you enable on a single system. It is a shared intelligence layer that sits between your enterprise data estate and your AI agents, translating raw organisational information into something agents can reason about.

Think of it as the connective tissue that was missing from the enterprise AI stack. Models are capable. Agent frameworks exist. Deployment infrastructure is mature. What was absent was a governed, shared, always-current understanding of how the organisation works that any agent could draw from, regardless of which platform built it or which system it runs in.

Microsoft IQ has four components. Here is what each one actually does.

Work IQ: The People and Process Context Layer

Work IQ addresses the human dimension of the context gap. It captures the signals generated by everyday collaboration inside Microsoft 365 and turns them into organisational intelligence that agents can query.

When you send an email, attend a meeting, share a document, update a Teams message, or collaborate on a file, those interactions leave signals. Work IQ reads those signals to build a dynamic, permissions-aware picture of the organisation: who works on what, who the relevant experts are for a given topic, what the current status of a project is, which documents are active and which are stale, and how decisions typically get made within a team.

This is the layer that allows an agent to say more than 'here is a document about this topic.' It allows the agent to say 'the person most recently involved in this decision is in your extended team, and the relevant document was updated three days ago by the lead on that project.' That is a qualitatively different capability.

Work IQ is built on Microsoft Graph and extends it with reasoning and memory capabilities. The Work IQ APIs became generally available on 16 June 2026, making this context programmable for developers building on Microsoft Foundry and Copilot Studio.

In plain terms:

Work IQ gives agents a real-time understanding of how your organisation actually operates, including who does what, who works with whom, and what is currently relevant, without the agent needing to be told all of this from scratch every time.

Fabric IQ: The Business Data Meaning Layer

Fabric IQ addresses the meaning dimension of the context gap. It solves a problem that sits beneath most enterprise AI failures and rarely gets named clearly: data without meaning is not enough.

Every large organisation has the same underlying tension. Data is stored in systems across finance, operations, CRM, ERP, and dozens of other platforms. Each system uses its own terminology, its own definitions, and its own data model. When three different departments define 'revenue' differently, any agent querying across those systems will produce inconsistent answers. When 'customer' in one system does not map cleanly to 'client' in another, the agent's outputs will be unreliable in ways that are difficult to detect and impossible to govern.

Fabric IQ resolves this through ontologies: structured, machine-readable definitions of your business concepts, the relationships between them, and the rules that govern them. Define what 'Customer', 'Order', or 'Revenue' means once, inside Fabric IQ, and that definition travels consistently to every agent, every report, and every AI experience that draws from your Fabric data estate.

Fabric IQ is built on top of OneLake, Microsoft Fabric's unified data foundation. It works with the Power BI semantic models that many organisations already have in production, extending their business logic beyond dashboards and making it available to agents. The semantic layer that your data team maintains for reporting purposes becomes, with Fabric IQ, the shared meaning layer that grounds all your AI agents.

In plain terms:

Fabric IQ makes sure that every agent speaks the same business language as your analysts. It turns raw data into understood data, so agents reason about customers, revenue, and orders the way your organisation defines those terms, not the way a model might guess.

Foundry IQ: The Knowledge Retrieval Layer

Foundry IQ addresses the data dimension of the context gap, specifically the problem of unstructured enterprise knowledge: the policies, procedures, contracts, technical documentation, regulatory guidance, and institutional knowledge that organisations hold in documents, SharePoint sites, knowledge bases, and storage systems but that no structured data model has ever captured.

Every enterprise has this problem. The most important knowledge in the organisation often lives outside databases entirely. It lives in a Word document that was last updated six months ago, in a SharePoint folder that three teams have contributed to inconsistently, in a PDF that contains the definitive interpretation of a regulatory requirement. When agents need to answer questions that depend on that knowledge, they need a way to find it, retrieve it accurately, and reason over it in a permission-aware way.

Foundry IQ provides a managed knowledge layer, powered by Azure AI Search, that connects all of these sources through a single, governed retrieval endpoint. Instead of every development team building their own retrieval pipeline for every agent they create, with its own chunking, indexing, and permission logic, they connect their agents to a shared Foundry IQ knowledge base that handles all of that centrally.

The retrieval process in Foundry IQ is not a simple keyword search. It uses agentic retrieval: the system plans the query, searches multiple sources in parallel, validates answers against source documents, and returns results with citations so the agent can trace exactly where each piece of information came from. This is what makes Foundry IQ useful in regulated environments where explainability is not optional.

In plain terms:

Foundry IQ gives agents secure, governed access to all the documents and knowledge your organisation holds, without every developer having to build the same plumbing from scratch. It is the layer that allows agents to understand your policies, procedures, and institutional knowledge, and to show their working when they do.

Web IQ: The Real-Time External Context Layer

Web IQ is the fourth component of the Microsoft IQ stack, announced at Build 2026. It extends the intelligence layer beyond the organisation's own data by grounding agents in real-time information from the web, powered by Bing.

The practical purpose is straightforward. Many enterprise use cases require agents to reason over both internal context and external signals at the same time. A procurement agent assessing supplier risk needs internal contract and spend data alongside current market intelligence on supplier stability. A compliance agent evaluating a regulatory question needs internal policy documents alongside the latest published guidance from the relevant authority. A sales agent preparing for a client meeting needs internal account history alongside current news about the client's business.

Without Web IQ, these combinations require custom integrations for every use case. With Web IQ available as a knowledge source inside Foundry IQ, the external context layer becomes as accessible as the internal one. The agent accesses both through the same retrieval endpoint, with the same permission model and the same citation capability.

In plain terms:

Web IQ means agents are not limited to what is inside the organisation. When a task requires current market context, regulatory updates, or external intelligence, the agent can access that from the same place it accesses internal knowledge, without a separate integration for every source.

How the Four Layers Work Together

Each IQ component is independently valuable. Most enterprise agents will draw on one or two of them depending on their use case. But the architecture is designed for them to work in combination, and the compound effect is significantly more powerful than any single layer.

Consider a practical example. A business development agent is preparing for a meeting with a major prospect. To be genuinely useful, it needs several distinct types of context simultaneously.

It needs to know who internally has been involved with this account recently, what conversations have taken place, and what the current relationship status is. That is Work IQ: the organisational and people context layer.

It needs to understand the commercial performance data: revenue from existing engagements, deal stage information, and account health metrics using the definitions your finance team applies, not generic ones. That is Fabric IQ: the structured business data meaning layer.

It needs to access the relevant proposals, service agreements, and technical documentation that have been shared with or discussed in relation to this account. That is Foundry IQ: the unstructured knowledge retrieval layer.

And it needs current intelligence on the prospect's recent announcements, competitive landscape, and market position. That is Web IQ: the real-time external context layer.

None of those four types of context required a custom integration or a separate development effort. Each one was available through a shared, governed context layer that the agent accessed via a single architecture. The result is a level of organisational intelligence that would previously have required hours of manual research, compressed into a moment.

The organisations that will derive the most from the agent era are not those with the most agents. They are those whose agents start from the richest, most accurate understanding of how the business actually works.

What This Means for Your Organisation

Microsoft IQ became generally available at Build 2026 across GitHub Copilot, Microsoft Foundry, and Copilot Studio. For organisations already operating within the Microsoft ecosystem, the infrastructure is available now. But availability is not the same as readiness, and the value of the IQ stack depends directly on the quality of the data and semantic models underneath it.

• Your Fabric and OneLake investment is now your AI foundation. If you have built semantic models in Power BI and governed your data estate in Fabric, those investments now have a direct line to AI capability. The quality of your agents' business reasoning is bounded by the quality of your semantic layer.

• Data definitions that live in people's heads are now a liability. The meaning layer in Fabric IQ is only as good as the ontologies and semantic models it draws from. If your business terminology is inconsistently defined across systems, your agents will inherit that inconsistency.

• The ROI case for Microsoft 365 and Copilot becomes stronger. Work IQ is built on Microsoft 365 signals. Organisations that have invested in Teams, SharePoint, and M365 Copilot now have that investment serving as the organisational context layer for every agent they deploy.

• Developer productivity on agent projects increases materially. Foundry IQ's shared knowledge bases mean teams are not rebuilding the same retrieval infrastructure for each new agent. The shared context layer reduces time from prototype to production.

• Governance and compliance become more tractable. Every piece of context returned by the IQ stack is permission-aware. Agents cannot surface information that the user they are operating on behalf of does not have access to. This resolves one of the most persistent concerns about deploying agents in regulated environments.

The Context Gap Is Solvable. But It Requires Investment in the Right Foundation.

The Microsoft IQ stack is the most comprehensive answer to the enterprise context gap that any vendor has produced. It addresses all three dimensions simultaneously: the human and process dimension through Work IQ, the business data meaning dimension through Fabric IQ, and the unstructured knowledge dimension through Foundry IQ, with Web IQ extending context beyond the organisational boundary for use cases that require it.

But the stack is infrastructure, not a shortcut. The organisations that will realise its full value are those that have already invested in a governed data estate, clear semantic models, and well-maintained knowledge assets. The IQ stack makes that investment productive for AI. It does not substitute for the investment itself.

The context gap is not a problem that better models will solve. It is a data and architecture problem. And unlike the model quality question, which has largely been resolved by the industry, the context architecture question is one that every organisation has to answer for itself. Microsoft IQ is the framework for doing so within the Microsoft ecosystem. Whether that framework delivers value depends on the quality of what sits underneath it.

Want to understand where your organisation stands on context readiness, and what it takes to activate the Microsoft IQ stack effectively?

VE3 works with enterprise teams across the UK and beyond to assess data estate readiness, build the semantic and governance foundations that the IQ stack requires, and implement Microsoft AI architectures that close the context gap in practice. Get in touch to start the conversation.