Your AI is only as smart as your data: Why AI readiness starts with an ECM 

We’re all in the middle of it. Companies are investing in artificial intelligence to analyze information  faster and automate processes.  

Yet many AI applications hit their limits as soon as they have to work with business information that  is outdated or scattered across different systems. Without a clearly valid version of a document or  the business context around it, the risk of wrong answers and poor decisions goes up. 

A Gartner study shows how big this challenge is: 63% of the organizations surveyed either don’t have  the right data management practices for AI or aren’t sure whether they do. In February 2025, Gartner  also predicted that through 2026, organizations will abandon 60% of AI projects that aren’t supported  by AI-ready data. 

That makes the systematic preparation and management of business information a key requirement  for successful AI. Intelligent document processing (IDP) can recognize and structure the relevant  content in documents.  

Enterprise content management (ECM) makes sure this information is available with the right  context, a traceable version history and controlled access rights. Together, they create a reliable  information foundation that AI applications can draw on in everyday business processes. 

Where AI initiatives get stuck in practice 

AI pilots usually work with carefully selected data. Things get harder when the same application has  to access a company’s actual document repository, where information is often much more difficult to  find and interpret than in a test environment. 

Unstructured documents 

A large share of corporate knowledge sits in PDFs, scans and email attachments. Modern AI can  read these documents, but that doesn’t mean the information inside them is automatically put in the  right place. 

Take a scanned supplier agreement. The AI needs to work out which deadline applies to termination.  If the contract contains several dates, simply extracting them isn’t enough. The AI also has to  understand what each date refers to, so the wrong deadline doesn’t end up in the system. 

Scattered systems 

In many large companies, the paperwork for a single transaction lives in several places. The original  contract sits in the contract management system, while later amendments were saved to a project  drive. 

If a procurement manager asks an AI assistant which supply contracts expire in the next 90 days,  the assistant may only consider the contracts it can find. If it misses an amendment that changed the  contract term, its answer still sounds plausible. The mistake may only come to light once a notice  period has already passed. 

Missing context 

Finding a document doesn’t guarantee the right answer either. What matters, for example, is whether  it’s the valid version of the contract or an older draft.

That information often sits in the document’s metadata. The same goes for access rights: an AI  assistant shouldn’t use confidential content just because it can technically retrieve it. 

For AI readiness, companies therefore need to make their content usable for AI and make sure the  right information is actually being used. Intelligent document processing and enterprise content  management play different but closely connected roles here. 

IDP: Turning documents into usable data 

Intelligent document processing uses AI to classify documents and extract the relevant information  from them. A scanned supplier agreement, for example, becomes structured data on the contract  term and the notice period. Before this data flows into downstream processes, it can be checked  against defined rules. 

IDP made its name in the automated processing of incoming invoices. Today, the technology also  handles more complex documents such as contracts or HR files.  

For AI readiness, this means information no longer has to be pulled out of the original document  again for every application. Once processed, it’s available for approval workflows and AI assistants  alike. 

A reliable information base needs one more step, though. The extracted data has to stay linked to  the original document so its origin can be traced and its use controlled. This is where enterprise  content management comes in. 

ECM as the information foundation for AI 

An enterprise content management (ECM) system manages business documents across their entire  lifecycle and links them to the business processes they belong to. The Doxis Intelligent Content  Automation platform shows how this works in practice. It combines ECM with AI-powered document  processing, so incoming documents can be classified automatically and their key information  captured before they’re used in business processes. 

For AI applications, what matters most is that information stays usable in its business context. An  ECM contributes to this in several ways: 

• Reliable sources of information: An ECM makes documents from different systems  accessible through a single search, without having to move every file to one central location. 
• Context: Metadata and version histories help identify valid documents and assign them to  the right transaction. 
• Permissions: When an AI application is properly integrated, the ECM can make sure it only  accesses information the user in question is authorized to see. 
• Traceability: Logging features record access and changes. Actions taken by AI agents can  be recorded too, provided they run through properly integrated interfaces. 

This becomes especially important with AI agents. When an agent edits documents or triggers  processes on its own, its permissions and actions have to stay under control. The experiences of  two organizations with very different requirements show what this means in practice. 

Lessons from the field: What organizations can learn on the road  to AI

What does preparing information for AI look like inside real organizations? That question was at the heart of a  recent conversation titled “Is Your Data Ready for Agentic AI?”, recorded live at an industry event in Munich  and released as Episode 8 of The Enterprise Content Show, a podcast about how organizations manage and  use their business information.  

The episode brought together guests from two very different worlds: TGW Logistics, which builds highly  automated warehouses for large companies, and the Natural History Museum in London, which looks after a  collection of around 80 million specimens. 

Both organizations use Doxis, though with different priorities. TGW relies on the platform to manage its project  documentation, while the Natural History Museum is rolling it out as part of a larger IT program. Despite their  different starting points, both are working on the same question: how to make their information usable for AI. 

AI can help improve the data foundation 

Duplicate records are a familiar problem for almost every organization, and the Natural History Museum is no  exception. The museum’s database might contain ten entries for Charles Darwin, for example, even though  they all describe the same person. 

Before building anything on top of that data, the museum is putting AI to work on cleaning it up. As Dr. Steen  Dupont, IT Programme Manager at the museum, puts it in the episode: 

“Instead of trying to get good outputs from good data, we’re going to use AI to fix the data first.” 

To do this, the AI compares details such as dates of birth and collection locations to flag likely duplicates.  These analyses can now run directly in the system, where the data already lives. 

In a next step, AI agents are meant to correct the errors they find on their own, which only works with clearly  defined permissions and rules that determine which changes an agent is allowed to make.  

Business teams need to be on board 

At TGW, keeping the document repository in good order is part of preparing for AI. Just as important is the  question of which applications fit within the company’s existing governance rules. 

Successful implementation, however, starts with the business teams. When IT spots a potential AI use case,  process owners need to be involved early. After all, they know how the work actually gets done, and they’re  the ones who will have to embrace the changes in their day-to-day work. 

Where to start 

Both guests in the episode came back to the same advice: AI readiness is built in small, deliberate  steps. For your document landscape, that means starting where you can solve a concrete problem: 

• Review your documents: Find out where the documents for an important business process  are stored and whether they’re complete and up to date. 
• Pick a meaningful use case: Focus on a process where AI delivers measurable time  savings. 
• Structure documents early: IDP can capture and validate relevant information as soon as  documents arrive. 
• Provide information in a controlled way: An ECM can make documents accessible across  systems without giving up existing permissions.
• Involve the business teams: Develop the use case together with the people who own the  process. 
• Improve data quality: Check whether AI can help you detect duplicates or faulty records  first. 

Foundation first, then agents 

A powerful AI model won’t help much if it finds outdated contracts or misses important amendments.  What really counts is whether the information it draws on is complete and reliable. 

Anyone who wants to put AI to good use in their organization should therefore start with how they  handle their own documents. That’s where it’s decided whether a promising pilot becomes an  application employees can rely on every day.