Stop tweaking your to-be process. Give AI its own lane.

Every process improvement follows the same shape. Document how work runs today, the as-is. Design how it should run, the to-be. Close the gap. The method is decades old and it still works. What has changed is the size of the gap you should be aiming for.

The classical loop, in short

For anyone who has not run this before, the method has five steps.

1. Map the as-is. Model the process as it actually runs, not as the handbook says. Talk to the people doing the work, walk the handovers, name the systems.
2. Agree on it. Review with everyone in the process until they say "yes, that is how it works". Without that, everything after is guesswork.
3. Design the to-be. Remove steps that add no value, automate what can be automated, integrate the systems that currently break the handover, streamline the flow. Attach the numbers: cycle time, cost, error rate.
4. Compare and implement. Put both versions side by side, understand what changes for whom, then implement the changes phase by phase.
5. Publish the new reality. Once the changes are live, the to-be becomes the new as-is. Publish it as the new way of working and share it with every stakeholder.

Then you go again. This is a loop, not a line: people comment on the published process, spot what still hurts, and the next round starts from there.

The weak point in this loop is step three. Most to-be processes are the as-is with fewer boxes. Sensible, incremental, and far too timid for what is now possible.

Why AI changes step three

Before AI, automation could only do what you could specify exactly. Anything that required judgement stayed with a person. That constraint shaped every to-be process you have ever seen.

AI removes it. A model can read an unstructured email and decide what to do with it. It can weigh a case against a policy. If your to-be still assumes the old constraint, you are designing for a world that no longer exists.

A more radical way to draw the to-be

Give AI its own lane. In BPMN, a lane is a role. Treat AI as one. Open your as-is diagram, add a lane called AI, and start moving tasks into it. Not "could AI assist here" but "AI owns this step". Where a person needs to review the outcome, that review is a task in the human lane, right after the AI one. This is where BPMN earns its keep: the diagram makes explicit what runs autonomously, what requires a human check, and where the handover between the two sits. Human and AI collaboration stops being an intention and becomes something drawn on the page. And it forces the question at every task instead of letting you skip past it.

Start with the BPMN decision gateways. Where the decision follows from information already in the process (a request, a document, a customer record), it is a candidate for the AI lane. Flag every gateway and work through them one by one.

Then look at the tasks for reasoning. Which tasks exist because a human has to read, interpret, classify or draft something? Reviewing a contract, triaging a ticket, preparing a summary for approval. These could never be automated before and can be now.

Be aggressive on plain automation, too. AI changed the economics of the boring stuff. Scheduled jobs, data transfers, format conversions: the scripts write themselves in an afternoon. Steps you left manual because integration was too expensive belong back on the table.

The result is a to-be that looks different from the as-is, not just shorter. A new role. Handovers between humans and AI that everyone can see and discuss. Decisions in seconds instead of days. That is the step change to design for.

What that requires from your tooling

You need modelling that is easy but not loose. The as-is has to be quick to capture or it never gets done, yet structured enough to map your entire operations, not just today's process. Crismo is built for that balance: real BPMN, fast to model, organised into value streams so it scales across the company and never gets in your way.

You need to see the delta clearly. When half the tasks moved lanes, a side-by-side diff is the only way to explain the change to the people it affects. Crismo's milestone comparison does this: mark the as-is, mark the to-be, get the visual diff plus the attribute delta (cost, time, effort) at process level.

You need the people in the room. An AI lane changes jobs. Invite every relevant stakeholder, let them comment, get their approval on the target state, or it will not survive contact with reality. Crismo's collaborator access, comments and reviews exist for this, and they keep the loop turning after the first round.

You need a standard. BPMN is what makes a diagram readable to the next team, importable into a workflow engine, and interpretable by an AI. Sketch tools cannot do this. Real BPMN can.

Start with one process

Pick a process you know well. Model the as-is, get it approved, set a milestone. Add the AI lane and see how much moves. Our bet: more than you expect.

👉 Try it in the free Crismo playground — no signup required — or create a free account to model inside a workspace with milestones and collaborative review.