Automating Chaos Is Manufacturing Chaos at Scale
An AI agent executes tasks quickly. If the process behind it has unnecessary steps, rework, or bottlenecks, the AI will repeat all of that at high speed. The result: you turn a slow problem into an industrial-scale problem.
Hypothetical example: a credit department requires three sequential approvals for proposals under R$ 50,000. An AI agent starts automatically collecting documents and filling out forms. The process gets faster, but the three approvals continue to clog the queue. The AI accelerated the input, not the output. The bottleneck became critical.
The Difference Between As-Is and To-Be
The as-is map shows the process as it is today: all steps, handoffs, waits, redundancies. It's a faithful portrait of the current flow, including the problems.
The to-be map is the redesign: it eliminates steps without value, parallelizes what can happen simultaneously, adjusts approval authorities, and defines where people decide and where AI executes.
Automating directly from as-is means perpetuating inefficiencies. Redesigning before automating means winning twice: a leaner process and more effective automation.
Three Redesign Moves: Eliminate, Parallelize, Reallocate
Eliminate is the most powerful move. Every step that doesn't add direct value to the customer or the decision can go. Hypothetical example: if the system already validates the tax ID, there's no need for manual verification.
Parallelize reduces cycle time. If two approvals are independent (finance and legal, for example), they can happen at the same time. In the as-is, they're often in sequence out of habit, not necessity.
Reallocate means deciding what stays with people and what goes to AI. AI is good at checking, comparing, filling out, notifying. People are irreplaceable in judgment, context, exceptions, and final decisions. Putting AI where it shouldn't be creates frustration. Leaving people doing repetitive work wastes talent.
Where People Should Stay: Authority and Decision-Making
Automation isn't about removing people from the process. It's about putting people in the right place. People should be where there's authority: approving, rejecting, authorizing, escalating.
Hypothetical example: an AI agent checks onboarding documents, validates data against external databases, and builds a checklist. If everything's OK, it forwards for approval. If there's a discrepancy, it flags it. The person with authority looks at the context, decides, and records the reason. The AI prepared, the person decided.
This clear division avoids bottlenecks (people aren't checking commas) and maintains control (critical decisions aren't in the hands of a script).
The Risk of Automating Without a Map
Without an as-is map, you don't know where the problems are. Without to-be, you don't know what to improve. Automating in the dark is betting that AI will "solve" it on its own. It won't. It executes what you tell it to, well or poorly.
Companies that put AI on top of unmapped processes typically face: customer complaints (the automated process got faster but is still bad), exception overload (the AI doesn't know how to handle what wasn't anticipated), and rework at scale (errors that were once isolated are now systemic).
How IN Does This in Practice
IN · Business Intelligence generates the as-is map from a recorded meeting or process description. It identifies pain points, times, and costs. Then it proposes the to-be redesign: eliminates steps, parallelizes where possible, and suggests where to place AI agents (to check and prepare) and where to keep people (to decide, with clear authority). Everything in BPMN 2.0, exportable to Camunda, and with SHA-256 verifiable records. You redesign before automating, and automate knowing exactly what you're doing.
Informational content. Numerical examples are hypothetical.