← Forward Deployed Engineer
18 min

Scoping, Success Numbers & the One-Page Spec

Turn a vague wish into a spec, one success number and a clear edge around the work.

Lesson 3 of 10
0 done
In plain English

Scope says what you will build, what you will not build and what done means. Pick one success number and measure it before you start. Write a one-page spec the customer can say yes or no to. Plan a small pilot with a date to decide.

Scope
The edge of the work: what is in and what is out.
Success number
One measure, such as minutes saved, that shows the project worked.
Pilot
A small, real test with an end date and a decision.
Picture it
Scope the project in five moves
1
Measure the baseline
Today's number
2
Pick one success number
What will move
3
Write the one-page spec
In and out of scope
4
Agree the test
Real cases, a pass mark
5
Plan a small pilot
With a decision date

After discovery, you must agree what you are building and how everyone will know it worked. Without this, every project drifts. This lesson shows how to turn a vague wish into a one-page spec, a success number and a clear edge around the work.

Why scoping matters

AI projects drift easily. The model can do many things, so people keep adding wishes. Scope is your protection. It says what is in, what is out and what done means. It keeps the customer happy and keeps you sane.

Pick one success number

Choose one number that shows value. It should be something the customer already tracks or can measure easily. Examples are the minutes to answer a ticket, the share of invoices processed without a person, or the number of leads followed up in a day.

Measure it before you build. That is the baseline. Without a baseline, you cannot prove improvement.

  • Time: minutes or hours saved per task.
  • Quality: errors found, or answers a reviewer accepts.
  • Volume: how many items one person can handle.
  • Money: cost per task, or revenue from faster follow-up.

The one-page spec

A spec for an AI project can fit on a page. It must be clear enough that the customer can say yes or no. Use the template below.

one-page-spec.md
PROJECT: {{name}}
SPONSOR: {{who pays and decides}}
USERS: {{who will use it daily}}

PROBLEM (2-3 sentences)
{{what happens today, and what it costs}}

WHAT WE WILL BUILD
{{one paragraph in plain words}}

SUCCESS NUMBER
Baseline today: {{number}}   Target: {{number}}   Measured by: {{how, who}}

IN SCOPE
- {{...}}
OUT OF SCOPE
- {{...}}

DATA AND SYSTEMS
- Data we need: {{...}}   Owner: {{...}}
- Systems we touch: {{...}}   Access level: {{read only / write}}

RULES AND RISKS
- Must never happen: {{...}}
- Needs human approval: {{...}}
- Privacy and legal notes: {{...}}

HOW WE WILL TEST
- {{a test set of real cases, who judges}}

TIMELINE
- Prototype: {{date}}   Pilot: {{date}}   Decision: {{date}}

Draw the edge

The out-of-scope list is as important as the in-scope list. Write down things people will ask for later. Examples are other languages, other departments, voice, and real-time dashboards. When a new wish appears, point to the page. Say, that is a good idea. It is not in this phase. Shall we add it to the next one?

Define done with a test

Agree how you will test before you build. Collect a set of real cases and the right answers, judged by the people who know. Decide the pass mark together. For example, a reviewer accepts at least nine out of ten drafts without edits.

⚠Do not promise perfect accuracy. AI makes mistakes. Promise a measured level, a way to catch errors, and a person in the loop for risky steps.

Plan the pilot, not the whole project

Start with a small pilot: one team, one process, a few weeks. Set a date to decide. At the date, the customer chooses to expand, change or stop. A pilot with a decision date builds trust, because nobody is trapped.

Key takeaways
  • ✓Scope protects everyone. It says what is in, what is out and what done means.
  • ✓Pick one success number the customer already tracks, and measure the baseline before you build.
  • ✓Write a one-page spec the customer can say yes or no to. Include risks and who must approve what.
  • ✓Agree a test set and a pass mark in advance. Plan a small pilot with a decision date.
Hands-on exercise

Write a one-page spec

  1. Use the problem from your mini discovery, or choose a new one.
  2. Fill in every section of the template. Keep it to one page.
  3. Write at least four out-of-scope items.
  4. Show it to one person and ask: could you say yes or no to this? Fix what is unclear.
Check yourself
Q1Why measure a baseline before building?show answer

Without a number from before, you cannot prove improvement. The baseline turns a feeling into evidence.

Q2Why list what is out of scope?show answer

New wishes keep appearing. The list lets you say not in this phase, and moves them to the next one without a fight.

Q3What should you promise about accuracy?show answer

A measured level on an agreed test set, a way to catch errors, and a person in the loop for risky steps. Never promise perfection.

← Discovery: Finding the Real ProblemPrototyping on the Customer's Real Data →
0 / 10 finished0%