It's Always the Process, Stupid (its.promp.td)

AI isn’t a magic strategy — it’s an accelerator for well-designed business processes. The piece argues that simply adding LLMs, agents, or generative tools to messy workflows only amplifies bad decisions: automate a flawed approval chain and you get faster bureaucracy. The one genuine breakthrough is AI’s ability to consume unstructured data (emails, PDFs, images, Slack threads), which exposes a hidden truth: many enterprise processes themselves are unstructured and undocumented because humans historically handled the mess. For practitioners and ML teams this has immediate technical implications. Before deploying models, map the workflow: identify the trigger (where the mess originates), define the transformation (what to extract or deduce), and specify the structured output that feeds ERP/CRM systems. Remember models are pattern-matchers that accelerate throughput, not replace human judgment—human governance remains required to define “risk,” edge cases, and quality metrics. A concrete example: AI can scan 50 contracts in minutes to extract risk clauses, but only if the review process and risk definitions are rigorously specified first. Bottom line: prioritize Business Process Optimization, structure unstructured workflows, then use AI to safely multiply velocity and scale.