Process redesign is the work of changing a process on purpose. It is broader than it sounds: moving an approval, merging two queues, giving the front line authority to settle a small dispute and rebuilding an entire service activation flow are all redesign. What they share is that the work itself changes, not only the tool that carries it.
How does redesign differ from reengineering and automation?
The terms overlap, and the differences matter when you decide what to do first.
| Approach | Scope | Starting point | What changes |
|---|---|---|---|
| Continuous improvement | Small, ongoing | The current process | Individual steps, a little at a time |
| Process redesign | One process, any depth | The outcome and the current process | Steps, handoffs, decision rights, roles |
| Business process reengineering | One process, radical | A blank page | Everything, including whether steps exist |
| Automation | Specific steps | The process as designed | Who or what performs the steps |
Davenport and Short's 1990 article in MIT Sloan Management Review used the phrase "business process redesign" and paired it with information technology from the start, arguing the two belong together. That pairing is the point. Technology chosen after the redesign fits the work. Chosen before it, the same technology freezes the old work in place.
What changes when the worker is part person, part model?
Classic redesign moved work between people and systems that follow rules. AI adds a third kind of worker, one that can read and judge but is sometimes wrong and does not know it. Designing for that worker raises new questions:
- Decision rights. Which decisions does the model own outright, which does it recommend, and which never reach it?
- Confidence routing. At what confidence does a case go to a person, and does the threshold differ by risk?
- Context at the handoff. When a case reaches a person, does it arrive with everything needed to decide it in one look?
- Feedback. How do the person's corrections flow back to improve the model and the rules?
The handoff between model and person becomes the center of the design. Get it wrong and people spend their day re-checking the model's work. Get it right and they spend it on the cases that need them, the model of human in the loop that holds up in production.
What does a redesign look like in practice?
Consider a transit agency handling fare refund requests. Today each request is read by an agent, checked against the fare system, approved by a supervisor if it exceeds a limit and paid by finance. A redesign might give classification and the fare check to a model, let refunds under the limit pay automatically, send disputed or unusual cases to an agent with the trip record attached, and drop the supervisor step for cases the rules already cover.
Why does evidence come before the whiteboard?
A redesign drawn from interviews describes the process people remember. The real process has variants, loops and workarounds that interviews miss. Process intelligence and process mining show them, which is why the See step comes before the Rethink step in the AI implementation method, and why an AI workflow automation build should rest on the same evidence. The whiteboard still has a place: after the data, where the argument about what the process should become can start from facts.
