When AI Lacks Context, Employees Pick Up The Slack
Luke Norris is the co-founder and CEO of Kamiwaza AI, driving enterprise AI innovation with a focus on secure, scalable GenAI deployments.
gettyEnterprise AI costs have a visibility problem. AI compute expenses are easy to see. Companies can review invoices for models, infrastructure and tooling. What doesn’t show up is the cost of operating AI without the unique context of the business, or a shared understanding of how an organization’s data, rules and decisions relate to each other.
When that unique understanding of your organization is missing, the task of connecting work produced from disconnected data and systems falls to employees. That burden adds to AI operating costs, from lost work time and slower decision-making to inconsistent outputs that are difficult to audit or explain.
Most leaders watch AI spending closely, but few understand the cost of the hidden work that AI hands back to their people, even though both determine whether their AI investments are paying off.
The burden of business context for enterprise AI falls on employees. Your compliance team may use one AI tool to review contracts, while the finance team uses another to model risk and legal to flag exceptions. Each tool works well within its own silo, but most business decisions require integrated input from multiple departments, like contract renewals, regulatory filings, budget approvals and vendor assessments.
When each team’s AI is working from its own data and definitions, no single output answers the necessary questions to make a business decision. That means human workers must assemble the full picture manually, comparing outputs, resolving conflicting definitions, tracing missing sources and reconstructing how decisions were reached. In regulated work, that adds time and introduces inconsistency into decisions that must be explainable and auditable, making humans just another bottleneck.
The problem is widespread and getting harder to ignore. McKinsey has identified fragmented data pipelines, siloed AI teams and LLMs’ inability to branch logic and track context over time as primary barriers to scaling agentic AI. Meanwhile, access to enterprise data is becoming a competitive territory in its own right. SAP’s 2026 API policy, for example, routes third-party agentic access through SAP-controlled architectures, a sign that the systems containing critical business data may increasingly operate on their own terms. Signals such as these should further reinforce for business leaders the importance of solving for connected business context if AI is to operate reliably at enterprise scale.
To overcome the barrier posed by cross-department data silos for scaled AI, conventional wisdom once dictated that all data must be consolidated and normalized so that enterprise AI solutions could understand and interpret the data they ingested. That instinct was understandable because data access challenges have familiar solutions. IT has been funding data warehouses and migration projects for decades, and when AI can’t work across systems, it looks like a data access problem.
But when it comes to AI, the context that gives data its meaning often gets lost in transit. For example, a contract value recorded in a CRM reflects what was signed. The same figure in a revenue recognition model may be split across time periods and subject to different accounting rules. AI pulling that number from one system without understanding how the other interprets it can draw the wrong conclusion. When unable to understand the business meaning of data inputs and their dynamics, it’s no surprise that companies still rely on manual labor to manage their AI investments, and thus fail to unlock AI’s full value.
For AI to have that connected context and work reliably across systems, organizations need a structured map of how information, business rules and decisions relate to one another. With a map of your unique business structure as its guide, dependable enterprise AI no longer requires data centralization or excessive employee intervention.
Before building that structured map, leaders need to know where manual integration work is happening today. A few questions can help surface it:
Look for places where separate teams run their own AI-assisted analyses and merge the results via a spreadsheet, a slide deck or an email thread. Each instance represents an integration your AI platform can’t perform.
Watch what happens when data analysis work moves from one team or system to the next. If the person receiving it has to ask what a figure means for their department or where it came from, the context of those results didn’t travel with the data.
When critical business information sits behind architectures you don’t control, your AI can only reason across it on the vendor’s terms. Organizations are likely making up for that limitation through direct employee intervention.
This work rarely shows up in a project plan, but it adds up quickly. In regulated industries, it also carries risk. For compliance purposes, decisions must be traceable and explainable long after they’re made.
Once you identify the manual work, quantify it. Count hours spent reconciling outputs, days added to approval cycles and decisions revisited because of inconsistent data. Without those numbers, the true cost of AI stays hidden.
Without connected context, the manual work of combining AI outputs will only increase as more teams adopt more AI tools and more vendors set their own terms for data access. Without measuring that work and its effect on decision speed and quality, leaders will keep misreading their AI investments as more efficient than they are. The longer that goes on, the more those work-arounds become routine, and the harder they are to recognize as a cost at all.
Treat connected context as infrastructure, and plan, fund and govern it with the same seriousness as compute. Then judge AI not only by how much work it takes off people’s plates, but also by how little it hands back.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?