Your company isn't ready for AI (and it's a data problem)

The pressure on companies to adopt AI is intense. Boards want an AI strategy. Investors ask about it in every meeting. Competitors are announcing AI-powered features every quarter. The fear of falling behind is real, and it's driving a wave of spending on AI tools, pilots, and "digital transformation" initiatives.

The results, so far, are sobering. McKinsey's 2026 State of Organizations report found that 86% of leaders feel their organisations are not prepared to adopt AI in day-to-day operations. Only 5.5% of companies are seeing real financial returns from their AI investments. The majority are stuck in what McKinsey calls "pilot purgatory": interesting experiments that never translate to business impact.

The common explanation is that AI isn't mature enough, or that the organisation lacks the technical talent to implement it. These explanations miss the actual bottleneck, which is simpler and more fixable: the knowledge that AI needs to be useful is scattered, unstructured, and inaccessible.

The data readiness gap

AI tools, whether they're coding assistants, internal chatbots, customer service agents, or analytical tools, all share a common dependency: they need context to be useful. A generic AI tool with no access to your company's specific knowledge produces generic outputs that require extensive editing and verification. An AI tool with access to comprehensive, current, searchable knowledge about your company produces outputs that are grounded in your reality and immediately useful.

The gap between these two outcomes is the data readiness gap, and for most companies, it's wide.

IBM research shows that 68% of enterprise data remains completely unanalysed. An estimated 80% of enterprise data is unstructured "dark data" that AI tools can't access or reason about. And 61% of companies openly admit their data assets are not ready for AI applications.

The knowledge exists. It's in Slack threads, email archives, meeting recordings, shared drives, wikis, CRMs, project management tools, and individual employees' heads. But it's scattered across dozens of systems that don't share context, in formats that AI tools can't easily consume, and without the structure or searchability that would make it retrievable when needed.

Why tools aren't the fix

The instinct when facing the AI readiness gap is to buy an AI tool and hope it performs. This is like buying a racing car and expecting it to perform without fuel. The tool (the model, the interface, the integration) is the engine. Your knowledge (the context, the documentation, the institutional understanding) is the fuel. An engine without fuel sits there looking impressive and producing nothing.

Most AI pilots fail because they're deployed into knowledge vacuums. The internal chatbot can't answer questions about your company because the answers aren't documented anywhere it can search. The coding assistant can't follow your conventions because your conventions aren't written down. The analytical tool can't provide useful insights because the context it needs is locked in people's heads rather than in a system it can access.

The companies in the 5.5% that are seeing real returns from AI share a common characteristic: they invested in making their knowledge accessible before deploying AI tools on top of it. The knowledge infrastructure came first. The AI came second.

What AI readiness actually looks like

Preparing your company for AI is primarily a knowledge management exercise. The specific steps:

Consolidate your knowledge sources. Connect the tools where knowledge currently lives, Google Drive, Slack, GitHub, Gmail, Notion, meeting recordings, into one searchable layer. The knowledge stays where it is. The search spans everything.

Make institutional knowledge explicit. The tribal knowledge that lives in people's heads, the decisions, the reasoning, the context, needs to be captured in a form that AI can access. Self-writing documentation that captures knowledge from Slack conversations, PRs, and meetings makes this happen automatically rather than requiring a special documentation effort.

Enable semantic search. Keyword search doesn't serve AI tools well because AI needs to find relevant information by meaning, not by exact phrasing. Semantic search across your consolidated knowledge base lets AI tools find what they need even when the user's question doesn't match the exact words used in the documentation.

Create AI access through open protocols. MCP (Model Context Protocol) makes your knowledge layer accessible to any AI agent without custom integration. The knowledge serves as a shared context layer for every AI tool in your stack, which means your investment in knowledge infrastructure multiplies the value of every AI tool you deploy.

The compounding advantage

The knowledge infrastructure that makes AI useful has a property that AI tools themselves don't: it compounds over time. Every day the system runs, it captures more knowledge, resolves more connections, and builds a richer understanding of how your company operates. The AI tools that read from this layer get better not because the models improve (though they do) but because the context they can access gets richer.

A company that starts building its knowledge infrastructure today will have six months of compounded context that can't be fast-forwarded or bought. That context is what turns generic AI tools into tools that are specifically useful for your company, your decisions, your market, your customers.

The 86% of companies that aren't ready for AI are, in most cases, companies that haven't invested in making their knowledge accessible. The fix is more prosaic than most AI strategy documents suggest: before you worry about which model to use or which AI features to build, make sure the knowledge your AI needs is consolidated, searchable, and current. Everything else follows from that.

Frequently asked questions

We've already deployed AI tools. Should we go back and fix the knowledge layer? You can do both in parallel. Connect your knowledge sources to a unified, searchable layer now. The AI tools you've already deployed will immediately benefit from better context, and the improvements will grow as the knowledge layer gets richer.

How long does it take to see improvement? Connecting sources takes hours. The search layer begins producing results immediately. The AI improvement is noticeable within weeks as the context available to your tools expands. The compounding effect becomes significant over three to six months.

Is this an IT project or a leadership project? Both. IT handles the technical connection of sources. Leadership sets the expectation that knowledge should be captured and made accessible. The cultural shift, from "knowledge in my head" to "knowledge in the system", needs leadership sponsorship to take hold.

Related reading: The cost of scattered knowledge, The knowledge scaling problem, How to break down information silos, The memory is the moat. Related pages: Self-writing docs, Connections, One search, MCP.