Walk Through a Complete Workflow Analysis

In the last two chapters you built the method: name the pieces, map the workflow from start to finish, then read the map for where it is likely to break. Now you get to watch that whole method run once, end to end, on a fresh everyday effort. Follow each move closely, because in the very next chapter you will run it yourself.

Meet the Green Thumbs Plant Sale

Green Thumbs is a student environmental club. Each spring the members grow seedlings and sell them to fellow students to raise money for a small campus garden. Nobody on the team is a technology expert, and everything runs on free, everyday tools that anyone can open.

Before mapping anything, list the components: every tool and every person the effort touches. Here they are.

The tools:

• an online order form, where a buyer picks their plants and a pickup time ;

• a shared spreadsheet, the master order list, with one row per order and a column for payment status ;

• a group chat, where volunteers sort out who staffs the pickup table ;

• a free AI assistant, used to draft each buyer’s confirmation message ;

• a free peer-to-peer payment app, where buyers send their money.

The people:

• the buyers, the orders volunteer, the treasurer, the comms volunteer, and the pickup volunteers.

That inventory is the raw material for the map. With every piece named, you can now walk the path they form.

Map the Workflow from Order to Pickup

The workflow starts when a student decides to buy a plant and ends when that plant is collected and paid for. Walk it step by step, naming the actor, the input, the output, and the transformation at each step, and marking every handoff, decision point, and dependency as it appears.

• Step 1. Actor: the buyer. Input: their plant choices and a pickup slot. Output: one submitted order. Transformation: filling in the form ;

• Step 2. Actor: the order form itself. Input: the submitted order. Output: a new row in the master spreadsheet. Transformation: the form drops each response straight into the sheet. Handoff: form to spreadsheet ;

• Step 3. Actor: the treasurer. Input: payment notifications in the payment app, plus the order rows. Output: each row marked paid or unpaid. Transformation: matching each payment to its order by hand. Handoff: payment app to spreadsheet ;

• Step 4 (decision point). Actor: the orders volunteer. Input: the order and the current stock count. The question: is every plant in this order still in stock? Output: the order is cleared to confirm, or flagged for a substitution ;

• Step 5. Actor: the AI assistant. Input: one order row (buyer name, plants, pickup slot) plus the instruction to write a warm confirmation. Output: a drafted confirmation message. Handoff: spreadsheet to AI assistant. This step depends on Step 2 having copied the order correctly ;

• Step 6. Actor: the comms volunteer, acting as the human-in-the-loop. Input: the drafted message. Output: an approved message sent to the buyer. Transformation: reading it and fixing anything wrong. Decision point: send as is, or correct first ;

• Step 7. Actor: the pickup volunteers. Input: the paid, confirmed orders. Output: plants handed over. Transformation: matching each buyer to their order at the table. This step depends on Step 3 and Step 6 both being done.

Notice the dependencies, because they tell you the order that truly matters. The AI draft in Step 5 cannot be right unless Step 2 landed a correct row, and the handover in Step 7 cannot happen cleanly unless both the payment in Step 3 and the confirmation in Step 6 are complete. That short list is the whole map: one clear start, one clear end, and every actor and handoff named in between.

Locate Where AI Acts

In this entire workflow, AI does exactly one job, at Step 5. Its AI model input is one order row plus the instruction to write a warm confirmation; its AI model output is the drafted message. The assistant does not check whether that row is right. It will happily produce a fluent message that names the wrong plant or the wrong pickup time, either because the row itself was wrong or because it filled a gap with a confident guess.

That is why Step 6 keeps a person in the loop. The comms volunteer reads each draft before it goes out, catching a confirmation that does not match the real order. The habit to build is simple: find the AI step, then find who checks it. Here someone does, which is a good sign, but only if that person genuinely reads rather than skims.

Diagnose the Fragility Points

With the map drawn and the AI step located, read the workflow for its weak spots. Each one below names a familiar pattern and the evidence you would look for to confirm it is real rather than a hunch.

• Manual re-keying at Step 3: the treasurer matches every payment to an order by hand, so a payment can be tied to the wrong buyer or missed entirely, leaving a paid order marked unpaid. Evidence to look for: a buyer insists they paid while the sheet says otherwise, a direct mismatch between the payment app and the spreadsheet ;

• Stale data at Step 4: the stock count is only refreshed now and then, so two buyers can both be cleared to order the last fern. Evidence to look for: the stock column reads one, yet two order rows name that same plant ;

• Unchecked AI step across Steps 5 and 6: if the comms volunteer skims instead of reading, a confirmation goes out naming a plant the buyer never chose. Evidence to look for: a buyer replies confused about their order, tracing straight back to a detail the assistant invented ;

• Single point of failure at Step 3: only the treasurer can open the payment app and mark orders paid, so if they are away on pickup day, no one can say who has paid. Evidence to look for: the whole pickup stalls whenever the treasurer is offline ;

• Unclear ownership of the stock count: no single person is responsible for updating it after each sale, so it quietly drifts out of date and feeds the stale-data problem above. Evidence to look for: ask who updates the stock count, and the answer is a shrug.

Mostly, yes, and that is worth noticing. An automatic handoff removes the by-hand copying that invites typos, so Step 2 is genuinely sturdier than the manual matching in Step 3. But automatic is not the same as unwatched: if a buyer leaves a field blank or the form’s questions change, the sheet can still fill with gaps. You mark an automatic handoff as lower risk, not no risk.

Finally, do not treat all five weak points as equal. Rank them by how likely a failure is and how much damage it would cause. The payment handling at Step 3 is both a single point of failure and manual re-keying, and a wrong paid status can send someone home without the plant they paid for, so it comes first. The unchecked AI step is likely but cheaper to get wrong, an awkward email rather than lost money, so it comes next. Sorting the map this way tells the club exactly what to shore up before sale day, instead of worrying about everything at once.

Let’s recap

• A full analysis starts by listing every tool and person, then walking the workflow from its first step to its last ;

• Naming the actor, input, output, transformation, and handoff at each step turns a vague effort into a map you can inspect ;

• Marking where AI acts, and who checks its output, shows at a glance whether the AI step is guarded or exposed ;

• Reading the map surfaces the familiar fragility patterns: manual re-keying, stale data, unchecked AI steps, single points of failure, and unclear ownership ;

• Ranking those weak points by likelihood and damage tells you what to fix first, before the workflow is put under pressure.

In the next chapter the method is yours to run. You will map a short workflow and find its fragility points on your own, then check your work against a full model solution at the end of the chapter.