Introduction
This blog is written by Jerome Fortias. He began working in AI in 1991 in the field of vision. After founding an AI lab for Sopra Steria in Brussels, he founded a startup, then joined the Amexio group.
During a discussion with a professor friend specializing in artificial intelligence, we touched on the “AI winters” — periods between 1960 and 2010 when investment in AI slowed significantly. He made a surprising observation: “If there’s one domain that’s safe, it’s AI applied to documents.”
This statement deserves closer examination.
Since the legendary Library of Alexandria (288–48 BCE), we have associated knowledge and intelligence with libraries, books, and documents. In the business world, this role is fulfilled by tools like CMS, DMS, and ECM, such as MS SharePoint, IBM Filenet, and OpenText.
However, are these tools being used to their full potential? I doubt it. If not, why is that the case?
Why Enhance ECM with AI?
In a context marked by significant geopolitical, economic, and technological uncertainty, organizations must bolster their resilience. Enhancing their operational intelligence with AI is, in my view, an essential step.
LLM, RAG, and CAG: Challenges and Opportunities
The advent of ChatGPT marked a genuine technological revolution, albeit accompanied by exaggerated promises and unrealistic expectations. Companies now aspire to create their own “enterprise ChatGPTs.” Naturally, they turn to their ECM tools for content.
Is it simple? Not quite.
While it’s easy to work with a few PDFs for a POC or demonstration, integrating AI at the scale of an ECM system requires deep expertise in disciplines like document management, semantics, and ontology.
Steps to Improve AI Integration into ECM
- Step 1: Define Strategic Objectives
Not all documents in a repository hold equal value. Depending on the company’s goals, it is crucial to define a clear orientation for AI-generated responses, for instance, by fine-tuning system prompts. - Step 2: Sort and Prioritize Documents
Document repositories often include duplicates, outdated files, or poorly versioned content. Identifying the most relevant materials using symbolic AI or rule-based systems is essential. This process can be optimized by distributing tasks across a grid computing system (e.g., Amexio Grid). - Step 3: Tailor Responses to the Audience
Just as Google customizes search results for individual users, an ECM system should adapt to user roles, security profiles, and behaviors. - Step 4: Enhance Search Capabilities
Traditional search functionalities (e.g., ElasticSearch) can be enhanced using technologies like semantic search, vector databases, and graph-based solutions. However, these options must be evaluated objectively to avoid choices driven by hype. - Step 5: Optimize the User Interface
An enterprise solution often requires a combination of search engine-like interfaces (Google) and chat-style interactions (ChatGPT). Ergonomics and enabling users to refine or enrich AI-generated responses are critical elements.
Beyond RAG, CAG, and LLM
Building an “enterprise ChatGPT” is a significant investment, particularly with a “Big Bang” approach. A more gradual strategy focusing on incremental ECM improvements (e.g., better document organization, improved search functionality, image analysis) can deliver quick value while laying the groundwork for future AI projects.
Improving Document Search: A True Victory
Even without implementing an enterprise chatbot, enhancing document search capabilities is a major asset. The benefits are manifold:
– Within office suites, to quickly locate relevant content.
– In email clients, to share the most pertinent documents.
– In ticketing tools, to access complementary information.
By improving search performance and usability, organizations can achieve significant efficiency gains.
Conclusion
Does this seem overly complex? It’s not. It’s no harder than implementing a BPM or RPA project.
The unrealistic promises surrounding AI in 2023 led to widespread disappointment. While building a POC with a few PDFs is straightforward, scaling these concepts to millions of documents requires genuine expertise.
AI is a specialized field built on decades of practice. The same applies to document management. At AmeXio, we leverage this expertise to help our clients strengthen their organizational intelligence.
