Lesson 5 of 6
Write the Build Plan: Success Measure, Scope, Milestones and Risks
Turn the chosen option into a one-page plan with a number that defines success, a clear edge to the scope, milestones the customer can see, and an early check for each risk.
5 minForward Deployed Engineering
After this lesson you will be able to write the plan itself: a success measure with a baseline, a scope with a clear edge, milestones that each end in something visible, a list of what you need from the customer, and risks with an early check for each.
Define success as a number
"Handle order emails better" cannot be checked, so nobody can say when the work is finished. A success measure names one thing that is counted, where it stands today, and where it should stand.
For Harrow, the count that matters to the operations director is same-day delivery. So the measure is: of the email orders received before 12:00, how many are entered before 12:00? In the sample from lesson 3 the answer was 8 of 13.
The target is the customer's to agree, and your part is to propose one. You might suggest that, for the three regular senders, at least 19 of every 20 such orders are entered in time. Then add a second measure that guards against making things worse: the share of email orders corrected after entry must not rise above what it is today, about a third in the sample.
Two rules keep a success measure honest. It must be something you can count from data the customer already has, here the same export the IT contractor produced. And the baseline must be measured before the build starts, on more than two days, which is one reason the plan asks for a month of orders.
Draw the edge of the scope
"In scope" says what will be built. "Out of scope" says what will not, and it is the list you will be glad of in week three.
Write the out-of-scope list from the things people actually asked for or are likely to ask for. For Harrow that means phone orders, emails from customers other than the three regular senders, replies to customers, any change to the order system itself, and removing the person who confirms each draft. For each one, give a short reason or say when it might be looked at again. An item that is out of scope "for this piece of work" is easier for a customer to accept than a flat no.
Plan in steps the customer can see
A milestone is a point where the customer can look at something and say whether it is right. "Parser finished" is not a milestone, because nobody at Harrow can look at a parser. "The order desk compares twenty drafts with what they typed" is one.
Make the first milestone small and near. If the first thing the customer sees is six weeks away, six weeks pass before anyone can tell you that you misunderstood. For Harrow, the first milestone takes one sender's emails and produces draft orders that are shown beside what the order desk typed. Nothing is imported. It can be ready in the first week, and it tests the riskiest idea, that these emails can be read reliably, before anything depends on it.
Give each estimate as a range and say what it depends on. "One to two weeks, provided we have the example emails and the import format by Wednesday" is useful. "Two weeks" alone becomes a promise the moment you say it, including on the day the examples arrive late.
Say what you need from the customer
Work for a customer stalls far more often on waiting than on anything technical. List what you need, from whom and by when, and put it in the plan so that it is agreed along with everything else:
- A month of orders from the same export, from the IT contractor, before work starts.
- A month of real emails from the three regular senders, in the first week.
- The format of the order import file, from the IT contractor, in the first week.
- A test area of the order system where imports can be tried safely.
- One named person on the order desk for an hour a day, to check drafts and say what is wrong with them.
The named person matters most. A build that the order desk has checked every day for three weeks is one they already trust on the day it goes live.
Name the risks and check them early
A risk in a plan is not a worry. It is a specific thing that could make the plan wrong, paired with the earliest and cheapest way to find out. For each risk write what it is, how you will check, and who can answer.
- A file import might not run the credit check. Ask the IT contractor and test it in the test area in week 1. If it does not, the plan changes before anything is built on it.
- A sender might change its email layout. The tool refuses any email it cannot read with confidence and leaves it for the order desk, as happens today.
- Customers might use their own names for products. The order desk keeps a list that maps those names to product codes, and the draft shows the original wording next to the code.
- The two-day sample might not be typical. Compare it with the month of orders before the target is agreed.
The plan on one page
Here is the finished plan for Harrow. Read it as a model. It is short enough to read in a meeting, and each line can be traced to something you heard or measured.
# Build plan: Harrow Building Supplies
## Problem
Email orders are typed in by hand. Orders that arrive in the morning
are often entered too late for same-day delivery, and many need
correcting afterwards.
## Evidence
Two-day sample of 24 orders, 14 by email:
- 5 of the 13 email orders received before 12:00 were entered too late.
- A typical email order waited about 2.5 hours to be entered.
- 5 of the 14 email orders were corrected after entry.
- Three customers sent 10 of the 14 email orders. Their emails are
sent by purchasing systems and always have the same layout.
## Success measure
For the three regular senders: at least 19 in 20 email orders received
before 12:00 are entered before 12:00 (sample: 8 of 13 for all email).
Corrections after entry do not rise.
## In scope
A tool that reads order emails from the three regular senders, writes
each as a draft order, and imports it once a person on the order desk
has confirmed it.
## Out of scope
- Emails from other customers (measured in step 4, decided after).
- Phone orders. Replies to customers.
- Changes to the order system. Removing the person who confirms.
## First piece of work
Week 1: drafts for one sender, shown beside what the order desk typed.
Nothing is imported.
## Milestones
1. Week 1: drafts for one sender, compared by the order desk.
2. Weeks 2-3: the order desk confirms and imports that sender's orders.
3. Weeks 3-4: the other two senders.
4. Weeks 5-6: measure against the baseline. Count the emails the tool
could not read, and decide whether an AI model for them is worth it.
## What we need from the customer
- A month of orders (export) and a month of the three senders' emails.
- The import file format and a test area of the order system.
- One named person on the order desk, an hour a day.
## Risks
- Import may skip the credit check: confirm with IT in week 1.
- A sender changes layout: tool leaves the email for the order desk.
- Customers' own product names: mapping list kept by the order desk.
- Sample not typical: compare with the month of orders before agreeing
the target.The plan never says how the emails will be parsed, which language the tool is written in, or where it runs. Those are your decisions, and they belong in your own notes. The plan is the part the customer agrees to: what will change for them, by when, what it needs from them, and what could go wrong.
Key takeaways
- A success measure names what is counted, where it stands today and where it should stand, and it is agreed with the sponsor.
- The "out of scope" list is what protects the plan later, so write it as carefully as the scope.
- Each milestone ends in something the customer can see, and the first is one or two weeks away.
- Give estimates as ranges and say what each one depends on.
- Every risk gets an early check and the name of the person who can answer it.
Try it
Complete your own plan.md for Harrow, with all nine headings filled in using your own words and not the example's. Then give it to somebody who has not read this course and ask them two questions: what will Harrow be able to see at the end of week 1, and name one thing that will not be built. A good result is a reader who answers both correctly from the page alone, in under five minutes.
Official documentation
Written with AI and checked automatically. Nobody has run these steps by hand, and tools change, so a command may behave differently in the version you install. Spotted a mistake? Tell us.