Key Takeaways
Enterprise AI can expose problems organizations have lived with for years: conflicting definitions, unclear ownership and data whose meaning changes depending on who’s using it. For Amir Hezarkhani, SVP & Co-Head of Delivery at BlueCloud, building trustworthy AI means confronting that ambiguity and creating the business context machines need to reason reliably, so organizations can turn AI into decisions and actions that drive real business value.
For BlueCloud SVP & Co-Head of Delivery Amir Hezarkhani, trustworthy enterprise AI begins with a problem organizations have struggled with long before AI arrived: agreeing on what their data actually means.
Ask the finance team what “revenue” means and they’ll answer without hesitation. Ask another part of the business the same question and the number may come back different. Both answers can be internally consistent, drawn from trusted systems such as SAP CFIN, Salesforce and used every day by people who understand the context behind them (also known as tribal knowledge). The problem emerges when that context has never been made explicit and system agnostic.
For example, one client Amir worked with operated with two reports built around different definitions of the same business term, each drawing from its own dataset. For the people using those reports, the discrepancy was manageable because they understood the context: which report served which purpose, what a particular term meant in a particular setting, and who to turn to when the numbers didn’t align. Much of that understanding never needed to be made explicit; it lived in the institutional knowledge people accumulated simply by working inside the business.
An AI system doesn’t arrive with that history. When an agent encounters conflicting definitions, it has to reason from the context it has. As Amir describes it, “It’s going to pick up the first definition that the system can grab,” and the answer may or may not reflect what the business actually intended.
That distinction matters because an unreliable answer is easily blamed on the model. Models can and do hallucinate, but some apparent AI failures in the enterprise begin further upstream, with data and definitions the organization itself has never fully reconciled.
Enterprise AI has a way of exposing ambiguity the business has been living with all along.
The Harder Data Problem Is Meaning
That ambiguity becomes part of the challenge as organizations put AI to work. And Amir’s perspective comes from years spent on the delivery side of complex data programs, working at the intersection of what a business wants from technology and what its systems and data can actually support.
As organizations move into agentic AI, some basic questions start to matter a great deal. What data assets do they have? Who has authority over them? What’s the quality of the data? Is it reliable? Does it need to be curated? Organizations don’t need perfect answers to every question before they begin, but they do need to address them as AI moves from experimentation into real business processes.
Good governance only gets an organization so far. The mechanics of moving and transforming data have become dramatically easier, but that doesn’t resolve a more fundamental problem: what the data actually means. Two teams can work from well-managed data and still use the same term differently. A pipeline can move both definitions perfectly, and an agent can retrieve both accurately, without knowing which definition the business intends.
AI is raising the value of something enterprises have historically relied on people to provide: shared understanding. Organizations have to make explicit the definitions, relationships and assumptions people have carried in their heads because a system that organizes ambiguous business meaning perfectly can still produce a well-organized wrong answer.
Giving AI the Context of the Business
Semantic and knowledge layers are one way of making that shared understanding usable by AI, connecting enterprise data to the business meaning behind it. But no architecture can decide that meaning on the organization’s behalf. Someone still has to determine what a metric means, who owns the definition, how competing definitions should be reconciled and what happens when the business changes.
For Amir, that human work is inseparable from the technology. In delivery, he often finds himself between business teams asking for an outcome and IT teams working within the realities of the environment. He describes the job as helping each side understand the other: where the technology is falling short, where expectations may be unrealistic and what can actually be delivered.
That translation is particularly important with AI because the gap between what a business wants and what its data can support isn’t always obvious at the outset. The technology can reason over the context it receives, but the organization still has to create that context. Trust depends on knowing what the system is reasoning over, what those inputs mean and where the underlying business assumptions came from.
AI Readiness Is an Ongoing Transformation Discipline
None of this stays settled for long. New products appear, organizations restructure, regulations shift, systems are added or retired, and the definitions that describe the business evolve along with it.
That’s one reason Amir sees treating AI as a discrete initiative with a defined endpoint as a fundamental mistake.
“They take AI as a project and they don’t take AI as a transformation initiative,” he says. “That’s a big mistake.”
Treating Data & AI as a transformation changes what readiness looks like in practice. The data foundation doesn’t need to be perfected before an organization begins, but it does have to evolve as AI is deployed, the organization learns and the business changes. New use cases expose new gaps, new questions reveal definitions that were never settled, and changes in the business alter the context AI needs to understand.
That’s also why enterprise AI is so closely connected to the broader work of data modernization. On Snowflake, modernizing fragmented data environments creates the technical foundation for analytics and AI, but the technology alone can’t establish the business meaning that makes the data usable and trustworthy.
For Amir, that work continues as the organization moves forward. More capable models will expand what enterprises can do with their data, but they will also place greater demands on the context surrounding it. Organizations have to keep strengthening that understanding as they learn, deploy new use cases and adapt to changes in the business. Because however capable the model becomes, it can’t resolve a business definition the organization itself hasn’t agreed on. Before AI can reliably understand the business, the business has to understand its own data.
Ready to Put Enterprise AI to Work?
Building trustworthy AI requires more than access to data. BlueCloud helps enterprises modernize and govern their data on Snowflake while building the business context AI needs to deliver meaningful outcomes.
Through a BlueCloud AI Discovery Workshop, we work with your team to assess readiness, identify high-value opportunities and define a practical path from idea to production.
