Recommended resources
The freely available resources this lesson will direct you to.
Hello. This course begins with a question that comes before choosing any AI tool: where, exactly, does work get done, delayed, checked, or passed to someone else? A workflow map makes those points visible. In this lesson, you will map a process at the level of tasks, decisions, and handoffs, then use the map to locate possible AI interventions without assuming every bottleneck needs AI.
By the end, you should be able to describe a current workflow clearly enough that someone who performs it could correct your map—and clearly enough that a proposed intervention has a specific place in the work.
Set the boundaries before drawing the map
“Improve order fulfillment” is too broad to map usefully. A workable boundary might be from an approved order reaching fulfillment to its dispatch. That gives you a start event, an end condition, and a way to decide whether an activity belongs on this map or in an adjacent process.
Within that boundary, record who participates, what information or materials enter, and what counts as a completed output. Do not start by drawing every detail. First establish the main route; then add the branches and delays that change how work actually proceeds.
What Is the Best Way to Create and Document Work Processes for a Team? | APQC
Read APQC’s guide for a practical way to scope a map before adding operational detail. Its distinction between a high-level map and a swimlane map will be useful throughout this lesson.
In “How Organizations Create and Document Team Work Processes,” read the five numbered stages, beginning with “1. Determine the process scope” and ending with “5. Map the process.” The passage begins with the mapping stages. Focus on why scope, inputs and outputs, and roles come before a detailed drawing.
For our order example, a first-pass map might say that customer service checks the order, planning schedules work, the warehouse prepares it, and shipping dispatches it. That is a useful outline, but not yet a useful map for finding an intervention. It does not tell us what happens when information is missing, whether work waits in a queue, or when responsibility changes hands.
Show tasks, decisions, handoffs—and time spent waiting
A task is an action, such as “check address” or “book carrier.” Labeling tasks with a verb and an object helps distinguish observable work from vague stages such as “processing.”
A decision selects a path using some criterion: “Is the delivery address complete?” Record both the question and what happens for each answer. If the criterion is currently subjective, preserve that fact rather than writing a precise rule that nobody actually uses.
A handoff changes responsibility to another person, team, or queue. It matters because the receiving party may need to find context, check the work, or wait before continuing. A wait state is different from a task: no one may be actively working on the order while it sits in a batch-release queue or awaits collection. Finally, a loop shows work returning to an earlier check after missing information or an error is corrected.
Watch “Process Mapping” from LeanOhio for a compact demonstration of how to build and inspect a swimlane map. Treat its colors and sticky notes as optional conventions; the distinctions between work, decisions, waits, and ownership are what matter.
Watch purpose and scope for why a map needs boundaries. In building the map, notice how tasks sit in the lanes of the roles doing them, while branch points and delays receive separate markers. Finish with inspecting friction, paying particular attention to rework and transfers of responsibility.
Now compare the small first draft at the top of this worked example with the larger observed current state. The draft names the major activities. The observed map adds an address-completeness check, a correction loop, a planning wait, a shipping wait, and lanes that reveal which team owns each activity.
The image is valuable partly because it resists an easy conclusion. A planning batch wait is not the same problem as repeatedly correcting incomplete addresses; a carrier collection wait is not necessarily something software can shorten. The map tells us where to investigate, not yet which technology to buy.
Build a map that someone can test against a real order
For each point in a workflow, try to capture five things: the actor, the action or decision, the input used, the possible path or wait, and the record showing what happened. You can place these in a diagram, a short step ledger, or both. The format is less important than whether the account matches the work.
The following ledger translates the order example into questions you could check with the people involved. It is deliberately more precise than “fulfill order,” without pretending that the image resolves every timing detail.
| Point in the workflow | Task or decision to map | Path, wait, or handoff to capture |
|---|---|---|
| Customer service | Check whether the approved order has a complete address. | If information is missing, correct it and repeat the completeness check. Record how the correction is obtained. |
| Planning intake | Receive the order for planning. | Record when planning takes ownership and whether the order waits for a batch-release window. |
| Planning | Plan the work. | Record what information planning uses and how the work is assigned to the warehouse. |
| Warehouse | Pick material and pack the order. | Record when shipping can act and what confirms packing is complete. |
| Shipping | Book the carrier and prepare for dispatch. | Record any collection wait, then the event that confirms dispatch. |
This is a current-state ledger: it describes what happens now, not an ideal AI-enabled design. Some apparent ordering also needs verification. For instance, does carrier booking require a packed order, or can those activities overlap? A plausible-looking map is not a substitute for checking a recent order.
Business Process Mapping for AI Automation: 5-Step Guide
Read the guide’s treatment of branches and delays to make the ledger more faithful to real work, including work that pauses and later resumes.
In “Step 3: Map Decisions, Handoffs, Wait States, and Exceptions,” start where the author explains that real processes are not straight sequences. Read the standard path, decisions, and waits, stopping at the end of the “Handoffs and Wait States” subsection. Focus on the distinction between a final outcome and a temporary status that returns work to an earlier check.
Two additions make this kind of map especially useful for business decisions. First, mark the information source for each check: an order form, an approved customer record, a scheduling system, or a message from another team. Second, note where human judgment or authority matters. “Check address completeness” might mean spotting blank required fields, or it might mean interpreting inconsistent address details across messages. Those are materially different tasks.
Locate intervention points without labeling every box “AI”
An intervention point is the precise place where changing how work is done might improve the process. Look for repeated interpretation of variable information, searching across documents, re-entering data, avoidable rework, or decisions made without timely information. Then ask what creates the friction.
IBM Technology’s case-management example illustrates one such point: a worker spends time finding facts scattered across arriving documents. The proposed assistance is a summary and extraction of relevant facts at the point the worker needs them. It is an example of a workflow insertion, not evidence that summarization will be accurate or valuable in every setting.
3 examples of AI transforming business processes
Watch the case-management portion of “3 examples of AI transforming business processes” from IBM Technology. Use it to identify the specific manual task being assisted, rather than accepting broad claims about productivity.
In case management, notice the distinction between receiving documents, searching them for relevant details, and using the extracted information to advance a case. Ask which of those activities the proposed assistance changes and which still belongs to the worker.
Applied to the order map, the promising points are conditional, not conclusions:
| Mapped point | Possible intervention to investigate | What could make AI unnecessary or unsuitable? |
|---|---|---|
| Address check and correction loop | Help staff find address details in varied incoming text and draft a clarification when details conflict. | If the issue is simply blank required form fields, validation at entry may be enough. An authoritative check still matters. |
| Planning intake | Surface the order information planners repeatedly search for across records. | If information is already structured and accessible, a better screen or integration may solve the problem. |
| Carrier booking | Transfer confirmed order details into a booking system. | If the transfer follows fixed rules, conventional automation may be more appropriate than AI. |
| Batch-release or collection wait | Investigate why work waits and who controls the release or collection schedule. | Faster text generation cannot, by itself, remove a scheduling policy or make a carrier arrive sooner. |
Notice the distinction between where a candidate sits and whether it merits investment. A proposed address assistant belongs at the completeness check and correction loop, not vaguely “in customer service.” Its possible output might be a highlighted inconsistency for staff to verify. At this stage, you have located and bounded the intervention; later lessons will help you choose the method and evaluate its value.
Validate the map before trusting the opportunities
A map drawn from a procedure manual may describe intended work rather than actual work. Check it against a few recent cases: one ordinary order and at least one that took a different path, such as an address correction or a missed collection window. Ask the people who performed and received the work where the draft differs from what occurred. If accounts conflict, mark the point as unknown until you can inspect further evidence.
Business Process Mapping for AI Automation: 5-Step Guide
Return to the guide for its advice on recording existing judgment and checks, then validating a draft against actual cases. Read these as instructions for describing the present process—not for adding new AI controls yet.
In “Step 4: Record Human Judgment and Existing Checks,” read from the “Human Judgment and Restricted Actions” subsection through “Quality Checks and Retained Evidence”: judgment and checks. Then, in “Step 5: Validate the Current-State Map,” read the validation walkthrough. Focus on checking a draft against cases and retaining unresolved questions.
As you validate, record just enough evidence to distinguish active work time from elapsed waiting time, and frequent problems from memorable exceptions. Timestamps, queue statuses, correction messages, and dispatch records may help; if those records do not exist, note the gap. Avoid assigning impressive savings to an intervention before you know how often the relevant path occurs.
Key takeaways
A useful workflow map has a clear start and end, shows who performs each task, states what selects each decision path, and makes handoffs, waits, and rework loops visible. Build it as an account of current work, then test it against real cases. AI intervention points are most credible when they name a specific task and information problem; a visible delay alone is not an AI use case.
In the next lesson, you will take these located intervention points and decide which call for generative AI, predictive methods, optimization, conventional automation, or a non-AI process change.