Fix the Process, Then Add AI

AI is not a strategy or a magic wand; it simply accelerates existing processes. Its real advantage—handling unstructured data—demands that companies first structure and formalize the underlying workflows. Design clear triggers, transformations, and outputs, then apply AI to scale speed under human governance.

Key Points

• There is no standalone “AI strategy”; effective AI use is a subset of Business Process Optimization.
• AI doesn’t make organizations smarter—it accelerates whatever process exists, good or bad.
• AI’s unique value is handling unstructured data, which reveals and requires structuring previously ad hoc processes.
• You can’t automate what you haven’t designed: define triggers, transformations, and structured outputs before applying AI.
• Human governance still sets definitions and guardrails (e.g., what counts as risk); AI delivers speed, not wisdom.

Sentiment

The community overwhelmingly agrees with the article's core message. Commenters draw on extensive personal experience with broken processes across industries — hedge funds, SaaS, enterprise IT, ERP implementations — to validate that process optimization must precede technology adoption. The few dissenting voices raise valid points about the article's narrow scope and AI's growing capabilities, but they represent a clear minority. The discussion is constructive and experience-driven rather than argumentative.

In Agreement

• Documenting processes forces clarity and consensus — just writing down what a step does often reveals that teams disagree on what it should be doing
• Automating a bad process only produces bad outcomes faster; process optimization must come before technology adoption
• Most tech debt is actually organizational debt introduced by leadership decisions to cut corners, not engineering failures
• AI's real value is handling unstructured data, but processes relying on unstructured data are themselves usually undocumented and unstructured
• The process-first principle applies to every technology cycle, not just AI — the same pattern has repeated with every buzzy technique for decades
• Good process should make talented people faster, not slower; when process stifles people, the problem is bad process design, not process itself

Opposed

• The article has a narrow BPO-centric view and misses that most companies want AI strategy for customer-facing products and new revenue, not just internal optimization
• AI can actually help companies learn and improve their processes faster — LLMs are already speeding up documentation and test case creation that never existed before
• The claim that 'risk assessment still requires human governance' becomes less true as LLMs internalize millions of past assessments, potentially outperforming humans who have seen only thousands
• The article conflates AI with LLMs — broader AI applications like medical imaging can genuinely transform entire process steps, not just accelerate them
• Trying to fully legibilize and document every process can damage organizational culture and flexibility