AI is Not the Solution; It’s a Tool

-

 


Partly created with AI

Most organizations don’t start AI projects with a clear problem.

They start with a statement like:

“We should use AI for this.”

That sounds forward-thinking. But it’s often where things go wrong. AI is not an objective. It’s not even a solution. It’s a tool; and like any tool, its value depends entirely on how and where you use it.

The Pattern: Starting with the Wrong Question

A common pattern in AI initiatives looks like this:

  • A repetitive process is identified
  • AI is suggested as the improvement
  • A solution direction is quickly sketched
  • Only later, the real complexity emerges

What’s missing is not technology. It’s problem definition and system thinking.

A Real Case: Automating Bill of Lading Processing

We’re currently exploring an initiative to process bill of lading documents received via email.

The objective is straightforward:

Reduce manual effort by automatically extracting structured data from PDFs.

The initial idea looked simple.

This model is intuitive; and incomplete.

It assumes:

  • Documents are consistent
  • Data can always be extracted correctly
  • Output can be trusted

In reality, none of these assumptions are guaranteed.

What Becomes Clear Very Quickly

Even in early exploration, the focus shifts from “How do we apply AI?” to:

  • What variations exist in document formats?
  • What defines a “correct” extraction?
  • When can we trust the output?
  • What happens when the system is unsure?

These are not AI questions. They are system design questions.

The Missing Layer: Decision-Making

The biggest gap in the initial model is the absence of a decision layer. Not every document can, or should, be processed automatically. Instead, the system needs to distinguish between different types of outcomes:

  • ✅ Good cases (high confidence, fully automated)
  • ⚠️ Unsure cases (require validation)
  • ❌ False positives (incorrect data accepted)
  • ❌ False negatives (valid data rejected)


Example Decision layer

This is where the real complexity lives. Not in extracting text; but in deciding what to do with it.

Early-Stage Reality: Working with Unknowns

At this stage, many things are still unclear:

  • The exact behavior of upstream systems
  • The stability of document formats
  • The reliability of extraction methods
  • The cost and dependency on external vendors

Instead of forcing conclusions, the focus should be on making unknowns explicit.

From Solution Thinking to Hypothesis Thinking

Rather than stating:

“AI will solve this”

A more effective approach is:

  • ⚠️ Hypothesis: Document variability may impact extraction reliability
  • ⚠️ Hypothesis: Not all cases can be automated without human validation
  • ⚠️ Hypothesis: The decision layer will determine overall system performance

This shifts the project from implementation to learning.


The Key Insight

AI doesn’t replace a process. It depends on a well-defined process. The success of an AI initiative is determined by everything around it:

  • Input quality
  • Process structure
  • Decision-making rules
  • Human interaction points

Closing Thought

In early-stage projects, the most valuable work isn’t building the solution. It’s structuring the problem. AI can be a powerful tool; but only when it’s placed in the right system, solving the right problem, with the right expectations. Until then, it’s just potential.





Reacties

Een reactie posten

Populaire posts van deze blog

Are AI-Generated Diagrams the End of Business Analysts?

-
Afbeelding
Or the moment they finally get to focus on what actually matters? Partly created with AI For decades, visual modeling has been one of the most visible outputs of a Business Analyst (BA): process flows, system landscapes, data models, swimlanes, architectures. Diagrams were proof of work. Now AI can generate them in seconds. Paste in a paragraph, a meeting transcript, or a rough list of steps, and tools will instantly output: Process flows System interaction diagrams Architecture overviews UML-like structures So the provocative question arises: If AI can create diagrams for us, what’s left for Business Analysts? Let’s unpack that — and why this shift is not the end of BAs, but a necessary evolution. The Rise of AI-Generated Diagrams AI-assisted diagramming has quietly crossed an important threshold. Today, you can: Describe a process in natural language Upload meeting notes Paste requirements or user stories …and receive a structured, syntactically va...
Meer lezen

From Chaos to Clarity – How to Design Process Diagrams That Actually Work

-
Afbeelding
Partly created with AI Introduction: The Illusion of Alignment We've all been in those meetings where everyone nods in agreement about a process—until someone says, "Let me draw it out." Suddenly, what seemed like alignment crumbles into a dozen different interpretations. Whether you're dealing with warehouse logistics, exception handling in ERP, or municipal BI dashboards, one truth remains: if it isn't visualized, it isn't aligned. Process diagrams are more than just boxes and arrows. They're tools to challenge assumptions, expose gaps, and foster meaningful discussion. But only if they're done right. The Hidden Mess Behind the Scenes Take a typical warehouse intake process. On paper, it's just "Receive, check, store." In reality? You might have incoming goods exceptions, missing barcodes, warehouse receipts that don’t match the PO, or cross-team dependencies that no one documented. Or imagine a BI project for a municipality: stak...
Meer lezen