Analyze a Workflow on Your Own
You have watched the method run twice now, on the weekend food drive and on the Green Thumbs plant sale. This chapter hands it to you. Using only the moves from the last three chapters, you will map a fresh workflow from start to finish and find where it is most likely to break. Give it a genuine attempt before you look at the model solution, because the thinking you do on your own is where the skill actually settles.
Your Turn! Map and Diagnose a New Workflow
Read the scenario below, then work through the three tasks in order. Everything you need is the method from the last three chapters: name the pieces, map the path, then read the map for its weak spots.
Context. Each spring, the Maple Street Community Garden assigns its raised beds to residents who want to grow vegetables. A small volunteer committee runs the whole effort with free, everyday tools, and no one on it is a technology expert.
Here is how a single plot request travels from start to finish:
A resident fills in an online request form, choosing a plot size and listing what they hope to grow ;
A committee volunteer reads each new form entry and types it into a shared spreadsheet, one row per request, with a column left blank for the plot number ;
The plot coordinator looks at the garden’s master plot map, decides which specific bed each resident gets, and writes that plot number into the row. The coordinator is the only person who keeps the plot map, and it is updated by hand every few days ;
A free AI assistant reads each spreadsheet row and drafts a warm confirmation message giving the resident their plot number and a few starter tips for the vegetables they chose ;
A volunteer copies the drafted messages into email and sends them out. The committee assumed someone would read the drafts first, but no one was ever clearly given that job.
The residents are happy and the committee is capable. Yet last spring two residents were assigned the same bed, one confirmation named a plot that had already been taken, and one resident was told to plant something they had never asked to grow.
Your task. Work through these three moves, writing your answers down as you go.
List the components: every tool and every person the workflow touches.
Map the workflow step by step. For each step, name the actor, the input, the output, and the transformation, and mark every handoff, decision point, and dependency. Then note where AI acts and who, if anyone, checks its output.
Diagnose at least three fragility points. For each one, name the pattern (single point of failure, manual re-keying, stale data, unclear ownership, or unchecked AI step) and point to the specific clue in the scenario that gives it away.
Deliverable. A short ordered map of the workflow, plus a list of at least three fragility points, each matched to its pattern and the clue you used to spot it.
Then check your answer against the model solution found at the end of this chapter.
Let’s Recap!
Running the method yourself is what turns a set of definitions into a skill you can actually use ;
Every analysis follows the same three moves: list the components, map the path step by step, then read the map for its weak spots ;
Marking where AI acts and who checks it shows at a glance whether that step is guarded or exposed ;
The familiar fragility patterns give you a checklist to scan against: manual re-keying, stale data, single points of failure, unclear ownership, and unchecked AI steps ;
Naming the clue behind each weak point keeps your diagnosis grounded in evidence rather than a hunch.
That completes the analysis half of the method. A short Part 1 quiz now lets you confirm what has landed, and then Part 2 turns from reading a workflow to judging it: evaluating outputs, troubleshooting failures, and deciding what to accept, fix, escalate, or override.
Model Solution
Here is one sound way to analyze the Maple Street plot-assignment workflow. Your wording will differ, and that is fine; what matters is that your map names the pieces in order and that each fragility point rests on a real clue.
The components.
Tools: the online request form, the shared spreadsheet, the master plot map, the AI assistant, and email ;
People: the residents, the committee volunteer who copies entries, the plot coordinator, and the volunteer who sends the messages.
The workflow, step by step.
Step 1. Actor: the resident. Input: their plot size and planting wishes. Output: one submitted request. Transformation: filling in the form ;
Step 2. Actor: the committee volunteer. Input: the form entry. Output: one new spreadsheet row. Transformation: typing it across by hand. Handoff: form to spreadsheet ;
Step 3. Actor: the plot coordinator. Input: the request plus the master plot map. Output: a plot number written into the row. This is a decision point, and it depends on Step 2 being complete. Handoff: plot map to spreadsheet ;
Step 4. Actor: the AI assistant. Input: one spreadsheet row plus the instruction to write a warm confirmation with starter tips. Output: a drafted message. Handoff: spreadsheet to AI assistant. This step depends on Steps 2 and 3 being correct ;
Step 5. Actor: the sending volunteer. Input: the drafted messages. Output: emails sent to residents. Transformation: copying each draft into email and sending it.
Where AI acts: only at Step 4. Its input is one row, its output is the drafted confirmation, and it does not know whether that row is right. Whether anyone catches a bad draft depends entirely on Step 5, where no one was given the reading job, so the AI step is effectively unguarded.
The fragility points. Any three of these earn the check.
Manual re-keying at Step 2: the volunteer types every request in by hand, so a mistyped plot size or a dropped detail feeds the whole workflow a wrong row. Clue: one resident was told to plant something they never asked to grow ;
Single point of failure at Step 3: only the coordinator holds the plot map, so if they are unavailable, no assignments can be made at all. Clue: everything downstream waits on that one person’s map ;
Stale data at Step 3: the plot map is refreshed only every few days, so two residents can be handed the same bed before the map catches up. Clue: two residents were assigned the same bed ;
Unchecked AI step across Steps 4 and 5: with no one assigned to read the drafts, a confirmation naming an already-taken plot goes straight out. Clue: one confirmation named a plot that had already been taken, and no one caught it ;
Unclear ownership of the proofreading step: the committee assumed someone would check the drafts, but no one was given the job, so it falls through. Clue: no one was ever clearly given the job of reading the drafts first.
If you mapped the five steps in order and named at least three fragility points with a real clue for each, you have run the full analysis method on your own. That is exactly the skill Part 1 set out to build.
