Every episode

Key takeaways

• -Prioritise in two passes: the process level tells you where to look, the task level tells you what to do
• -Tag processes with AI potential and process maturity, each high, medium or low
• -Shortlist processes with high AI potential and low maturity: lots of manual work, safe to change
• -Rate tasks on value potential and feasibility; the ones high on both are your first initiatives
• -Let AI pre-classify the landscape as a first draft, then have process owners decide
• -Save the current state as a milestone before any change, so you can measure whether AI paid off

Key takeaways

• -The to-be is a sequence of milestones built from the as-is, not one diagram drawn at the end
• -Freeze the as-is as a milestone first so everyone works from the same baseline
• -Each milestone is one change and a work package an implementation team can build
• -Compare milestones side by side instead of writing a spec: added tasks and attribute changes show up
• -Accepting an offer online can trigger the contracts, notify sales and close the opportunity
• -Work left to right: improve upstream first, then move downstream

Key takeaways

• -Design the ideal process on a green field before asking what AI can take over
• -Weekly agent runs with reminders replace the month-end scramble for receipts
• -A dedicated agent lane turns autonomy into an explicit decision, e.g. whether the agent may send money
• -An explicit BPMN model served as the spec, with almost no back-and-forth with developers
• -Agents that report their actions back into the process build the audit trail themselves
• -Train agents on the 50 to 80% standard cases and route the rest to human review

Key takeaways

• -The usual mistake is not wrong BPMN but mixing different things into the same view
• -Lanes add accountability; a simple process with ten departments reveals organisational friction
• -Business level: show responsibilities with departments, teams and functions as lanes
• -System level: show interactions with systems as pools and message flows as interfaces
• -Sequence flow between lanes, message flows only between pools
• -Consistency beats correctness: pick a convention for each level and stick to it

Key takeaways

• -Use value streams as simple segments that organise diagrams around how value flows
• -Levels: the landscape, then value stream segments, then business processes, then system views
• -Greenfield: agree the top-level value streams with leadership first, then model top-down
• -Existing diagrams: build the value stream landscape, then thread your models into it, with AI suggesting where each belongs
• -Link upstream and downstream processes horizontally to follow the journey end to end
• -Don't map everything: start with one core value stream and connect it properly

Key takeaways

• -Tag risks from a central catalogue, then let AI flag every task with a risk but no control
• -Maintain documents once and see every impacted process the moment one changes
• -Task costs plus gateway probabilities give an average cost per run and a weighted list of cost drivers
• -Classify tasks as waste, required or value to show where AI adoption actually pays off
• -Don't attribute everything at once: one process, one attribute type, then the next layer

Key takeaways

• -Get the as-is validated by the people who run the process before changing anything
• -Ask the people doing the work where it hurts most; it builds the case and their buy-in
• -Combine their input with AI analysis of time and cost per activity
• -Compare the approved as-is with the to-be to quantify the gain and get budget approved
• -Implement, then repeat the loop
• -Don't start with your entire process landscape: one process, get it right, then expand

Key takeaways

• -AI fills time and cost attributes from spreadsheet values in seconds and totals them per process
• -Ask the model directly for the most expensive step or where time is lost, then sanity-check the answer
• -AI can surface inconsistent naming and modelling patterns across many diagrams
• -Restructuring a landscape, like extracting an outsourced lane from 100 processes, went surprisingly far
• -AI removes the friction around modelling; the judgement stays with you

Key takeaways

• -A diagram without attributes is just a picture
• -Don't put data on the canvas, put it behind the canvas
• -Start with three attributes, not twenty: time, cost and owner
• -Attributes unlock reporting, system mapping and audit-ready compliance reports
• -Roll attributes up into totals for leadership; the diagram is the audit trail underneath

Key takeaways

• -Pools for external parties (suppliers, customers, partners), lanes for internal departments
• -Without pools, a process falsely signals 100% internal control
• -Default to collapsed pools for suppliers; expand only when you dictate their process
• -Proper pool boundaries let you find every supplier interaction across your landscape
• -A sequence flow crossing a lane already implies a handover; no message events between departments

Key takeaways

• -A process map shows where AI fits in seconds instead of hours of debate
• -Separate agents by goal: one classifies (go or no-go), one synthesises (the business case)
• -Build human review and approval into the flow before giving agents autonomy
• -Structured data input makes a big difference to agent performance
• -Time-per-task attributes help you pick the first pilot when capacity is limited

Key takeaways

• -A diagram can look perfect and still be logically broken
• -A parallel split without a proper join silently duplicates the flow
• -Exclusive gateway or event-based gateway: do we decide, or do we react?
• -Never split without an explicit gateway
• -Small oversights create live loops that never end; token simulation catches them
• -Interrupting events stop the process, non-interrupting ones run in parallel

Key takeaways

• -Sweet spot for scope: about 15 tasks in the happy path
• -Ask 4 questions before modelling: Who owns it? What triggers it? How does success look? Why does it exist?
• -Talk to more people, not fewer
• -A process living as a PDF on a shared drive is already dead
• -Map variants, label them and give them a home on the value chain
• -Agree on the as-is before designing the to-be