Worldwide spending on IT will reach a staggering $6.31 trillion by the end of 2026, according to Gartner. That is just for 2026, on top of the roughly $5.56 trillion spent in 2025. All of that technology makes up the backbone of how our businesses are run and how internal processes get automated.
On top of that, and far less measurable, organizations have internal knowledge repositories defining the specific rules and logic for their own business purposes. They have specific rules on how REVENUE is calculated, what constitutes an ACTIVE CUSTOMER, when an OPPORTUNITY becomes a BOOKING, and how hundreds of other business concepts should be interpreted.
Yet when organizations begin deploying AI against enterprise data, they often encounter a fundamental problem:
The AI doesn't know any of this business knowledge.
Consider a seemingly straightforward question:
"What were our bookings last quarter?"
A modern AI system can discover the relevant database. Technical metadata can tell it which tables exist, what the columns are called, how data types are defined, and perhaps how tables relate to one another.
But what exactly is a BOOKING?
Does it include renewals? What about services? Are cancelled orders excluded? Is the metric based on total contract value or annual contract value? Which date determines the quarter? Are there exceptions for certain product categories or geographies?
Those answers are not contained in the database schema.
They are part of the organization's business knowledge: the definitions, rules, calculations, exceptions, relationships, and institutional understanding that give data its meaning.
As enterprises move from traditional BI toward natural language analytics, copilots, and autonomous AI agents, discovering and governing this knowledge is becoming one of the most important, and most underestimated, challenges in enterprise AI.
The Knowledge Already Exists. It Just Isn't in One Place.
Over the past six months we have had the opportunity to showcase Codd AI to numerous executives and leaders across multiple industries. Organizations are now coming to the realization that AI is powerful but it needs help to become accurate and consistent. That help comes in the form of CONTEXT.
One of the first challenges they bring forward is that "our business knowledge and rules are not well documented." The resulting concern is that it will take years to document everything, and that AI can't be used pervasively until that work is finished.
The good news is that it does not have to take that long.
Much of the knowledge already exists. The problem is that it has accumulated over years or decades across different systems, applications, documents, and people.
Some of it may exist in data catalogs, business glossaries, policies, operating procedures, requirements documents, SharePoint sites, or Confluence pages.
Some is embedded in data pipelines, stored procedures, and SQL.
A significant amount may exist inside BI platforms, where analysts have created calculations in reports and dashboards, along with filters, groupings, custom hierarchies, and other logic to make raw data reflect the way the business actually operates.
And some of the organization's most valuable knowledge may not be documented anywhere.
It resides in people's heads.
An experienced sales operations analyst may know that three categories of transactions need to be excluded when calculating bookings. A finance analyst may know that a particular customer hierarchy must be used for revenue reporting. A supply chain expert may know that orders with a particular status should not be considered open orders even though the database technically categorizes them that way.
People apply this knowledge every day without necessarily thinking about it. It is the same expert reasoning we described in From Dashboards to Decisions, where the real value sits in the judgment layered on top of the data rather than in the data itself.
The organization therefore "knows" these things.
Its AI doesn't.
This creates a very different challenge from simply cataloging data.
The enterprise knowledge problem is rarely that the knowledge doesn't exist. The problem is that nobody knows exactly where all of it exists.
Technical Metadata Tells You Where the Data Is. Not What It Means.
Technical metadata is an important part of the equation, but it is also the relatively straightforward part.
Modern platforms can automatically inspect databases and identify schemas, tables, columns, data types, primary keys, foreign keys, and other structural information.
That gives AI a map of the data environment.
But a map is not the same thing as understanding the territory.
Knowing that CUSTOMER_ID connects an orders table to a customer table doesn't explain what the organization considers an active customer.
Knowing that a column called ORDER_AMOUNT contains a decimal value doesn't tell AI whether that value represents revenue, bookings, gross order value, or something else entirely.
Technical metadata might tell AI how the data is structured.
Business knowledge tells AI what the data means.
For enterprise AI to answer business questions reliably, it needs both. This is the difference we explored in AI with Context vs AI with a Contextual Semantic Layer: having access to information is not the same as understanding it.
And this is where knowledge discovery becomes substantially more difficult.
Why "Just Mine the Query Logs" Isn't Enough
One increasingly popular approach to discovering business logic is to analyze database query logs.
At first glance, this makes perfect sense.
Query logs contain the SQL that users, applications, reports, and dashboards actually execute against production systems. Instead of asking people how they use the data, why not simply observe what they do?
Query logs can certainly be valuable sources of information. But treating them as a complete representation of enterprise business knowledge introduces several problems.
First, execution does not necessarily explain meaning.
A SQL statement can reveal which tables were joined, which filters were applied, and which calculations were performed. It doesn't necessarily explain why those rules exist or which business concept the query is intended to represent.
Second, some of the most important business logic may never reach the database.
Consider a BI dashboard. The underlying SQL query might retrieve relatively straightforward transactional data. But the report developer may then create local calculations, custom measures, filters, groupings, hierarchies, and exceptions inside the BI platform.
If knowledge discovery stops at the database query log, that logic remains invisible.
The AI sees the query.
It doesn't see the complete business interpretation applied to the results.
Third, query logs can introduce enormous amounts of noise.
A popular dashboard might execute substantially the same query hundreds or thousands of times. Slight variations may be generated for different users, filters, or reporting periods.
Simply ingesting every query doesn't necessarily produce more knowledge.
It may just produce more data for AI to sift through.
Instead of asking AI to find a needle in a haystack, we have given it a much bigger haystack.
Stop Thinking "Knowledge Project." Start Thinking "Discovery System."
At this point, many organizations may reach an uncomfortable conclusion:
"Are you telling us we have to document everything our business knows before we can deploy AI?"
That sounds like a multi-year knowledge management or data governance initiative.
And if organizations approach the problem using traditional methods, it could become exactly that.
Imagine attempting to manually interview hundreds of subject-matter experts, inspect thousands of dashboards, analyze millions of queries, read years of documentation, identify every business rule, reconcile every contradiction, and manually enter everything into a knowledge repository.
That isn't a particularly attractive AI strategy.
Fortunately, AI itself changes how organizations can approach the problem.
Instead of thinking about business knowledge as a massive documentation project, organizations can begin thinking about business knowledge discovery as an automated and continuous system.
The key is recognizing that different knowledge sources require different discovery techniques.
Rather than asking one general-purpose AI system to "understand the enterprise," specialized scanners or agents can be assigned bounded discovery tasks.
The basic pattern becomes:
Discover, Extract, Structure, Validate.
This creates a fundamentally different approach to building enterprise context.
Different Knowledge Sources Require Different Discovery Methods
Consider the different places business knowledge can reside. In each case, AI can help with the discovery and curation process.
| Knowledge source | What discovery looks like |
|---|---|
| Data platform | Inspect databases and automatically extract schemas, tables, columns, relationships, views, stored procedures, and other technical metadata. |
| Query logs | Examine SQL execution history. Rather than indiscriminately ingesting every query, identify recurring patterns, cluster similar queries, remove duplication, and surface potentially important calculations and business rules. |
| BI platforms | Inspect analytical artifacts from platforms such as Power BI or Tableau. The job isn't simply to find the SQL behind a report. It is to find the business logic created inside the analytical layer: calculated measures, filters, custom groupings, relationships, hierarchies, and other report-specific logic. |
| Document repositories | Examine business glossaries, policies, requirements documents, data catalogs, SharePoint content, Confluence pages, PDFs, and other unstructured sources to identify definitions, rules, terminology, exceptions, and business relationships. |
Another interesting area is getting knowledge from subject-matter experts that has never been documented before.
Imagine an AI-powered SME survey engine tasked with interviewing the organization.
It might ask sales operations experts:
"How do you define a booking?"
It could ask customer success teams:
"When do you consider a customer active?"
It might ask finance:
"Which transactions are excluded from recognized revenue?"
Or supply chain experts:
"What conditions determine whether an order is considered backlogged?"
Instead of organizing weeks of workshops and manually documenting responses, AI agents could conduct structured discovery campaigns across dozens or hundreds of subject-matter experts.
This changes the economics of business knowledge discovery dramatically.
But it also exposes another problem.
Bonus Benefit: Discovery Reveals How Little Organizations Agree
Suppose five subject-matter experts are independently asked:
"How do we define an active customer?"
Three provide roughly the same definition.
One uses a different time period.
Another says that a particular product category should be excluded.
Which definition is correct?
This is where business knowledge discovery becomes particularly interesting.
The AI hasn't failed.
In fact, it may have discovered something extremely valuable: the organization itself does not have a consistent definition.
The same problem can occur when analyzing existing systems. A query-log scanner may identify one calculation for gross margin. A Power BI scanner may discover three slightly different versions. A finance document may define the metric another way. And an SME may explain that the official definition changed six months ago.
This is exactly the condition we called analytical drift, where the same question produces different answers depending on who asks it and which system answers.
Automated discovery can therefore generate enormous volumes of potentially useful information.
But more knowledge is not automatically better knowledge.
Organizations can encounter duplicates, outdated rules, conflicting definitions, department-specific interpretations, personal conventions, obsolete calculations, and outright mistakes.
The problem is now shifting.
Originally, the challenge was: how do we find our business knowledge?
Now it becomes: how do we determine which business knowledge should be trusted?
The Missing Step: Distillation
This is why enterprise knowledge discovery cannot simply follow a Discover, Ingest, Use model.
A more robust process looks something like this:
Discover, Extract, Normalize, Compare, Reconcile, Certify, Use.
Once knowledge has been collected, AI can play another important role in distilling it.
It can identify duplicate definitions.
It can recognize that twenty SQL statements represent variations of essentially the same business rule.
It can compare calculations discovered in different BI reports.
It can flag contradictory definitions.
It can identify terminology that appears to represent the same business concept.
It can distinguish frequently used rules from isolated examples.
And it can highlight areas where different parts of the organization appear to disagree.
This is an important distinction for organizations building knowledge bases, knowledge graphs, or semantic layers for AI.
The objective should not be to collect the largest possible amount of business knowledge.
The objective should be to create the most relevant, consistent, governed, and trustworthy representation of how the business works.
That requires curation.
And curation requires governance.
Why Human-in-the-Loop Becomes More Important, Not Less
There is a natural temptation to automate the entire process.
If AI can discover business rules, compare definitions, identify contradictions, and recommend a preferred interpretation, why involve humans at all?
Because determining how a business should define something is fundamentally different from discovering how people currently define it.
Consider our five definitions of an active customer.
AI can identify the differences. It can determine which definition is most common. It can show which dashboards and reports use each definition. It can identify which definition appears in official documentation. It can even recommend a canonical definition.
But deciding which definition should become the authoritative enterprise definition is a business and governance decision.
A qualified data steward, analyst, or subject-matter expert should make that determination.
AI can accelerate knowledge discovery. It shouldn't quietly appoint itself the authority on how your business works.
This principle becomes even more important as organizations move toward autonomous agents.
A questionable metric in a dashboard may result in a bad chart.
A questionable business rule used by an autonomous agent could result in an incorrect action.
Human-in-the-loop governance therefore shouldn't be viewed as friction in the AI process. It is part of the mechanism that turns discovered information into trusted enterprise context, and it is a core part of the blueprint for trusted enterprise AI we outlined for Chief Data Officers.
Business Knowledge Discovery Isn't a One-Time Project
There is another reason organizations shouldn't approach this as a traditional documentation exercise.
Business knowledge doesn't stand still. And in the AI-powered world, it is going to move even faster.
New products are launched. Pricing changes. Organizations restructure. Customer classifications evolve. Regulations change. New dashboards are built. Metrics are revised. Acquisitions introduce new terminology and systems.
Even if an organization could perfectly document its business today, that documentation would begin becoming outdated within a year.
Business knowledge discovery therefore needs to become a continuous operating capability.
This also changes how organizations can get started.
There is no requirement to document the entire enterprise before AI can deliver value.
The practical approach is to start with a high-value domain such as sales.
Discover the definitions, calculations, rules, relationships, and terminology required to understand bookings, pipeline, revenue, customers, opportunities, territories, and products.
Validate and govern that knowledge.
Make it available as trusted context for analytics and AI.
Then expand into finance, customer success, supply chain, operations, or other domains.
The goal isn't to finish documenting the enterprise.
The goal is to establish a repeatable system for continuously discovering, validating, and governing the knowledge the enterprise depends on.
From Data-Aware AI to Business-Fluent AI
The enterprise AI conversation has understandably focused on models.
Which LLM should we use? How large should it be? How should we implement RAG? How do we connect agents to enterprise systems?
Those questions matter.
But the quality of enterprise AI will increasingly depend on something more fundamental: the quality of the context surrounding the model.
AI needs more than access to enterprise data.
It needs to understand what that data means within the context of a specific organization. That is the gap between a query generator and a business-fluent AI system, and it is also why governing models alone isn't enough.
That understanding already exists throughout most enterprises. It is encoded in databases, queries, BI reports, calculations, documents, policies, processes, and decades of accumulated human expertise.
The challenge is finding it. Quickly.
Modern AI gives organizations an opportunity to rethink how that happens. Specialized agents and scanners can continuously search different knowledge sources, extract relevant business logic, eliminate duplication, identify contradictions, and surface institutional knowledge that might otherwise remain hidden.
But discovery is only the beginning.
The knowledge must then be distilled, reviewed, reconciled, and governed before it becomes trusted context for analytics and AI.
At Codd AI, this is the philosophy behind how we are approaching business knowledge discovery: using specialized AI-driven discovery mechanisms to help organizations find and organize the business context they already possess, while maintaining human-in-the-loop governance before that knowledge becomes part of the trusted semantic foundation used by AI.
The broader principle is simple:
- Use AI to discover the knowledge.
- Use humans to govern the knowledge.
- Use the resulting trusted context to make AI business-fluent.
Organizations don't need to spend the next three years documenting everything they know before they can deploy enterprise AI.
They need a better way to discover what they already know, and a governed process for deciding what their AI should trust.
About Codd AI
At Codd AI, we believe enterprise AI should reason from a shared, trusted understanding of the business, not from isolated prompts or disconnected data sources. Our AI-powered contextual semantic layer combines technical metadata, business knowledge, governance, and human certification into a reusable enterprise foundation for conversational analytics, AI copilots, and agentic workflows.
By helping organizations capture and govern business context once, Codd AI enables every AI and BI application to deliver more consistent, explainable, and trusted insights, today and as the AI landscape continues to evolve. To learn more, visit www.codd.ai or schedule a conversation with one of our co-founders.