🔣 Simple flow notation
You don’t need full BPMN. Three symbols handle 95% of processes: rectangle for one step, diamond for one decision and arrow for the direction of the flow. Simple enough for anyone to understand.
💡 Practical tip
Start with pencil and paper. A hand-drawn diagram made in 10 minutes is worth more than a perfect BPMN that nobody finishes.
🔄 As-is vs. to-be
Draw two flows side by side: the as-is (how it works today, full of manual steps) and the to-be (what it will look like with AI, streamlined). The visual contrast is the most convincing proof of value — the client sees the steps that disappear.
✗ As-is (current)
- ✗Too many manual steps and screen changes
- ✗Waiting and rework along the way
- ✗Long cycle time, high cost
✓ To-be (future state)
- ✓Manual steps eliminated or combined
- ✓AI handles classification, extraction, and generation
- ✓Cycle time in seconds, low cost
🧱 Find bottlenecks and rework
With the as-is mapped out, the problems become obvious. Bottleneck is the slowest step, which limits the whole process. Rework is a correction loop that could have been avoided. Both are the areas with the highest returns.
Mark the slowest step
The bottleneck
That’s where the queue forms. Speeding it up speeds up the entire process.
Look for arrows that loop back
The rework
Correction loops indicate that something was done wrong earlier. Fix the cause, not the symptom.
Find the wait times
Dead time
Waiting for approval or for someone else is usually the biggest cause of a long cycle time.
💡 Practical tip
Addressing the bottleneck yields more than optimizing any other step. Improving a step that isn't the bottleneck doesn't speed anything up.
🌿 Decision points and branches
Every process branches at some point: approved or rejected, type A, B, or C. These decision points need to be mapped carefully—each branch must have a clear destination, and the conditions must be mutually exclusive.
⚠️ Attention
A decision without a complete exit branch becomes a bug in the automation. Make sure that todo the path has a destination — including "none of the options".
🤖 Where AI fits in the diagram
Not every step needs AI. Mark only the nodes where it really shines in the to-be: language (classify, extract, summarize, generate) and diffuse judgment. Fixed rules? Use regular logic — it's cheaper and more reliable.
✓ AI shines at
- ✓Classify free text (emails, tickets)
- ✓Extract data from unstructured documents
- ✓Summarize, translate, and generate text
- ✓Making decisions with diffuse context
✗ Don’t use AI to
- ✗Calculations one formula can handle
- ✗Fixed rules like "if A, then B"
- ✗Move data from one field to another
- ✗Decisions that require exact precision
🏷️ Mark the intelligent nodes
In the to-be diagram, highlight the AI steps with a distinct color or icon. That makes it clear where the automation’s “brain” is—and where it’s just plumbing.
✅ Validate the diagram with the client
Before writing a line of automation, sit down with someone who knows the process and validate the diagram. Fixing a design takes minutes; fixing a finished automation takes days. The right question is "is this right?" — not "shall I build it?".
🔍 Validation plan
- •Go through step by step with whoever executes it.
- •Ask about exceptions that were missing from the design.
- •Confirm the to-be: does the proposed future state make sense?
- •Define the scope: validated diagram = agreement on what will be built.
💡 Practical tip
Treat the validated diagram as a visual contract. It aligns expectations and protects you from change requests that "were always in scope."
🔑 Key concepts of the module
📊 Module Summary
Next Track:
Track 3 · Solution Design — from the mapped process to the blueprint: architecture, tools, models, and logic.