What to automate first · SOP to workflow

The SOP with the decisions missing

Find the decisions a five-line invoice SOP leaves to its builder, write each as a rule, and decide whether the 12-hour build starts on Monday.

critiqueeasyExercise · 3 min · 11 of 75

On this page

A builder has quoted 12 hours to automate the invoice run from a five-line SOP. Before saying yes, find what the five lines leave the builder to guess.

The situation

The operations manager wrote this SOP for the Friday invoice run, the task the arithmetic of a task prices at $520 a month: 40 invoices a week, 6 minutes each, staff at an assumed $30 an hour. A freelance builder has quoted 12 hours to automate it from the page as written.

Invoice run — Fridays
1. Export the week's timesheet from the time tracker.
2. Add up each client's hours.
3. Multiply by the client's rate.
4. Create the invoice in the accounting system.
5. Send it to the client.
Done: all invoices sent.

What the operations manager knows and did not write down. About 1 timesheet line in 8 is tagged internal and is never billed. Four clients are on monthly retainers of 20 hours, with hours over that billed at the hourly rate at month end. Two clients reject any invoice without their purchase-order number. On a typical Friday, two or three people have not submitted their timesheets by 15:00. Assume a wrong invoice costs $50 of rework: a credit note, a reissue, a call.

The question

Find the decisions this SOP leaves to the builder, write the rule for each, and decide whether the builder starts on Monday.

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Four decisions are hidden.

Step 2 hides which hours count. Rule: bill only lines tagged to a client project; internal lines never reach an invoice.

Step 3 hides which rate. Rule: the four retainer clients get no weekly invoice; on the last Friday of the month they get the retainer fee plus hours over 20 at the hourly rate. Everyone else is hours × rate.

The trigger hides whose timesheets. Rule: run at 15:00 on submitted timesheets; a client with unsubmitted hours is held to Monday 10:00, and the operations manager gets the list of who is missing.

Step 5 hides what makes an invoice sendable. Rule: a purchase-order client's invoice with no PO number goes to the exception queue, not the client.

The done line cannot fail. Replace it: by 17:00 Friday, every client with submitted billable hours has exactly one invoice sent or one exception queued, and invoiced hours equal the export's billable hours.

Do not start Monday. Built literally, internal lines are billed and retainer clients are invoiced by the hour. If even a quarter of invoices carry one internal line, 10 a week go out wrong: $500 a week, about $2,170 a month of rework against a task that costs $520 by hand. An hour with the operations manager writing the four rules, $30, comes first; then the 12 hours buy a page that can be tested, as SOP to workflow sets out.

What would change that: if the export already drops internal lines and retainers are billed in a separate monthly run, two decisions disappear and the build can start once the other two are written.

Rubric

finds the internal-hours decision in step 2 and the retainer-rate decision in step 3
writes each decision as a rule a builder could code without asking a question
treats missing timesheets as a trigger decision and states one rule for it
routes the purchase-order case to an exception queue rather than to the client
replaces "all invoices sent" with a count that balances by a stated time
decides not to start, with the cost of building the SOP as written

Where answers go wrong

"Let the builder use judgement, or put a model on the edge cases." These are the agency's billing policies, not judgement calls. A model guessing a retainer cap makes a $50 error per guess, and nobody can test a build against a rule that was never written.

"Add a step: the operations manager checks every invoice before it goes." A check with no rules behind it is the original 6 minutes done again, and at that length the automation never pays back.

"Start Monday and fix it in testing." The first Friday's 40 invoices go to real clients, so the test costs $50 for every wrong one.