Skip to content 
Every executive team is asking the same question right now: how do we turn our AI investment into better business decisions? The ambition is there; so are the budgets and the tools. But the results do not match expectations.
When looking for the reason, people tend to think the model is wrong or there is not enough data.
But the reality is that most AI analytics initiatives fail because they are built on top of a foundation that is not designed to support them.
Look around your organisation. If your business logic is fragmented, your metric definitions are not consistent, and data means different things depending on who you ask, this article is for you.
More data, less confidence
This is the paradox many business leaders are quietly living right now. Organisations have never had more data, more analytics tools, or more AI capabilities available, and yet confidence in the numbers is lower than ever.
"You do not have an AI problem, you have a data trust problem."
We have said this phrase in prospect calls more times than we can count over the past year.
Executives receive conflicting reports for the same KPI. Analytics teams spend more time validating data than generating insights. Business users are not sure which dashboard represents the real answer. And when AI surfaces an insight, the first instinct is to verify it manually before acting on it.
The numbers behind this problem are stark:
| $12.9M | The average annual cost of poor data quality to an organisation (Gartner, 2021) |
|---|---|
| 43% | Of Chief Operations Officers identify data quality as one of their most critical challenges (IBM, 2023) |
| 25%+ | Of organisations estimate losses exceeding $5M per year as a direct result of data quality issues (IBM, 2023) |
There is also another, deeper issue, namely what happens when AI enters that environment. At the enterprise level, agents expose fragmented foundations faster, with higher-stakes decisions attached.
Why AI analytics projects actually fail
To understand the root of the problem, it helps to look at what is happening inside most organisations.
Business definitions drift over time
Take something as fundamental as revenue. Depending on which team you ask, it might mean gross revenue, net revenue, bookings, ARR, or recognised revenue.
Each definition might be valid in context; the problem is when those definitions live in different dashboards, different SQL queries, and different reports with no single source of truth.
If people cannot agree on what a number means, AI has no chance of giving a trustworthy answer.
Business logic gets scattered everywhere
As organisations grow, logic gets embedded across:
- BI tools
- Data pipelines
- Custom queries
- One-off reports
Every new dashboard is an opportunity for metrics to drift and every new analyst inherits assumptions they may not fully understand.
Governance becomes difficult for humans, and nearly impossible for AI systems trying to interpret the business.
Traditional dashboards were not built for the questions you ask AI
Business users want to know why revenue declined, what is driving churn, and which markets are underperforming. These questions require exploration and context, not pre-built charts. When AI is layered on top of disconnected dashboards, it inherits all of their limitations.
AI does not inherently understand your business
Large language models understand language but they do not understand your specific metric definitions or your business rules. They do not even know which data a given user should be allowed to see.
Without that context, every AI-generated answer is an inference, and every inference is a potential risk.
Between data and AI
Clean data is a good start but enterprises need far more than that.
There is a missing layer between the data and the AI agents that connects business meaning to data, consistently, in a governed way, which both humans and AI systems can rely on.
This is what a semantic layer does.
A semantic layer sits between raw data and the systems that consume it. Instead of forcing every dashboard, analyst, and AI tool to independently define business logic, it centralises those definitions in one governed place.
This changes a lot:
- Revenue means the same thing everywhere.
- You define business rules once and they are applied consistently.
- There is a foundation for AI capabilities that already understands how your business works.
Introducing the semantic layer
The semantic layer sits between raw data sources and every consumer: dashboards, self-service analysts, AI chat, and embedded or MCP-connected tools.
This is not a new concept but it has become the most critical infrastructure decision that enterprises can make as AI moves to the centre of how they operate.
Omni is the fastest way to create a semantic layer for your business. It provides a governed, version-controlled foundation where metrics, business logic, AI context, and permissions are defined once and shared consistently across every dashboard, analyst workflow, AI application, and embedded experience.
Different answers to the same question?
“How was the total revenue by product and date?” It is an everyday business question.
In a traditional AI analytics environment, the AI may have access to dashboards, reports, and raw tables. Yet, it has no shared understanding of how the business defines revenue, it will not be able to give a trustworthy answer.
This does not mean that there will be no answer.
Without a semantic layer, AI agents sum a raw order-amount column, the closest match it can find, and answer confidently from numbers that include refunds and tax. However, Finance reports a governed product-revenue figure that excludes both. The totals do not match.
Did AI answer the question? Yes, it did.
Did it use the wrong number, confidently? Also yes.
Can the business user tell which number is right? Not at all.
How Omni tackles the metrics problem
To be able to get correct answers from AI, you need to make sure it understands the business context behind the data.
Here is what the difference looks like:
With Omni’s semantic layer and AI context, AI uses the governed Total Product Revenue definition and returns the right figures by product and date.
Here is how it works:
The semantic layer gives the model a governed definition of revenue, consistent business logic, and the relationships between metrics. AI context adds the documentation, explanations, and business meaning that help the model interpret results correctly.
This shift is what turns AI from a system that sounds convincing into a system decision-makers can trust; from AI you have to double-check, to AI you can actually rely on.
How Omni builds that contextual foundation
Omni was built around the idea that analytics and AI should share the same trusted foundation, not operate as separate systems with separate interpretations of the business.
At the centre of Omni’s platform sits a governed semantic model where subject-matter experts define core metrics, dimensions, and business logic once, and make them available consistently across dashboards, self-service analytics, embedded applications, and AI-powered experiences.
When a business user asks a natural language question, when an analyst builds a custom query, when an AI agent pulls data to support a decision, they’re all working from the same definitions.
Let’s see what users say:
“Choosing Omni is solving for more than just BI. We’ve also primed ourselves to leap forward into AI because the semantic model is at the heart of the platform." – Mike Doll, VP of Data at Guitar Center
“Our big lesson with AI is that it’s about control. When you constrain it and give it context, like Omni’s semantic layer does, you get predictable, reliable results that drive action.” – Edward Mancey, Director of Data at Synthesia
The shift that Omni enables turns agentic output from something you have to double-check, to insights you can actually rely on.
Omni AI analytics tools in practice
Omni’s AI capabilities are powerful because they are built on top of a governed foundation and use a context shared across teams and functions in the organisation.
Example:
AI chat lets any user ask questions in plain language, e.g. “What drove the drop in conversion last week?” or “Which regions are underperforming versus target?”, and gives them answers grounded in the same metrics your analysts use.
Because of the shared context built from the start, users get trustworthy answers rooted in shared logic, not hallucinated inferences. Context carries over, so the conversation can go deep.
Here, we can see that the answer is grounded in the governed revenue definition, not a raw column, and links back to the metric and the SQL that ran.
Modelling agent
The Omni Modelling Agent helps data teams build faster. It handles a lot, including:
- Generating metrics
- Adding AI context to the semantic model
- Incorporating existing documentation
- Accelerating the work of defining business logic, without replacing the governance layer.
For example teams at Cribl use it to support self-service analytics for over 700 monthly users with a lean data team.
Agent skills
Agent Skills are pre-packaged agent behaviours that bring Omni’s capabilities into your workflow, in the chat interface and directly in your IDE, to build models, run queries, manage content, and more.
They make the power of the semantic model accessible without requiring users to understand what is underneath.
MCP Server
This is where things get genuinely interesting for the AI-forward organisation.
Omni’s MCP Server lets you query your governed data from any AI platform, Claude, ChatGPT, Cursor, or your own custom chatbot. Your semantic model becomes the trusted data layer for whatever AI surface your team is using.
For instance, Photoroom’s Head of Data described the impact directly: a team of three enabling more than 100 people to get answers in seconds.
Branch mode
Omni’s Branch mode lets teams tune AI context in a safe environment before anything goes live, meaning you can improve how AI understands your business without risking what is already working. Combined with Git and version control, it closes the loop on governance.
The pattern across all of these tools is the same: the AI is constrained by, and empowered by, the semantic model. That is the design choice that makes the difference.
Turning data into a strategic asset
Most organisations treat data as a technical resource, something IT manages and analysts query.
In contrast, the organisations that thrive on AI agents treat data as a strategic asset that creates competitive advantages and compounds in value over time.
The difference between those two realities comes down almost entirely to trust. When business definitions are standardised and governed, teams stop debating which number is correct and start focusing on what to do about it. Analytics scales because self-service becomes safer and teams see AI as a genuine force multiplier rather than another system that they need to validate.
The organisations that succeed with AI will not necessarily be the ones with the most data or the largest budgets but they will surely be the ones that build a trusted foundation first.
They all share the following:
- Their metrics are governed
- Logic is centralised
- Every AI system touching the data works from the same shared understanding of the business
AI is only as trustworthy as the foundation beneath it. Right now, building that foundation is the most valuable investment most organisations can make.
This is the first of two articles in a mini-series on Omni’s capabilities and the importance of building a semantic layer. Part 2 is published and focuses on giving AI agents the business context they need via the Omni semantic layer architecture.
