Skip to content

Lesson 1 of 6

What a Proof of Concept Is For: One Question, One Week

Learn what a proof of concept has to prove, how to cut a customer's wish down to one question a week can answer, and why the pass test is agreed before any result exists.

5 minForward Deployed Engineering

After this lesson you will be able to say what a proof of concept is for, turn a customer's wish into one question, and write the one-page brief that keeps a week of work on course.

A forward deployed engineer (FDE) is often asked to "show us it can work" before a customer commits to a full project. That piece of work is a proof of concept. This course follows one from the first conversation to the write-up, a day at a time.

The case

One made-up customer runs through every lesson. Brightwell Leisure runs four leisure centres with gyms and pools. Members pay monthly and can cancel at any time. Too many do.

The head of membership has a small team who could phone members and offer help: a session with a trainer, a different plan, a pause. They can make about 15 calls a week. The trouble is that nobody knows whom to call until the cancellation arrives, and then it is too late.

Her request to your company is: "Can you use our data to tell us who is about to leave?"

What a proof of concept is for

A proof of concept exists to remove one doubt, cheaply, before real money is spent. It is finished when the customer can make a decision: build it properly, change direction, or stop.

That gives it a different shape from a small project.

  • It answers a question. It does not deliver a product.
  • It runs once, on your machine, on a copy of the data. It does not have to run every night.
  • Most of it may be faked or done by hand, as long as the part being tested is real.
  • It ends on a fixed day, whatever state it is in.

The commonest way these go wrong is that nobody wrote down the question. The work then grows for three weeks, ends in a demo that looks good, and leaves the customer no nearer a decision.

Cut the wish down to one question

"Tell us who is about to leave" is a wish. To find the question inside it, ask what the customer would do with the answer, and what data exists.

At Brightwell the answer is used for one thing: the team's 15 calls a week. And the one record Brightwell has of what members do is the turnstile, which logs every visit. So the doubt to remove is narrow.

"From the visit records we already have, can we make a weekly list of up to 15 members who are much more likely than most to cancel in the next four weeks?"

That question can be answered yes or no. It names the data, the size of the list and the period. It leaves out a great deal on purpose: why members leave, what to say on the call, and whether calling helps.

Agree the pass test before you start

Next, agree what "yes" means, in numbers, before any result exists. Ask what is true today. About one member in eight cancels in a typical four weeks, so a list chosen with no information would be right about one time in eight.

The head of membership agreed this test: of the members on the list, at least one in three must go on to cancel within four weeks, measured on a period the list was not built from.

Fixing the number first protects both sides. If you choose it after seeing the result, you will choose whatever you got, and the customer will be right to doubt it.

Real, faked and left out

Sort every part of the idea into one of three lists.

  • Real: the thing in doubt. Here, the rule that picks members, and the test of that rule against what happened.
  • Faked: whatever a finished product would need but the question does not. Here, the data is one export made by hand, the list is a file, and nothing runs on a schedule.
  • Left out: whatever is somebody else's question. Here, the calls themselves.

Write the lists down. They are how you say no during the week, and how the customer knows what they are looking at on demo day.

The brief

Put all of it on one page and get the customer's agreement in writing.

# Proof of concept brief: Brightwell Leisure

## Question
From the visit records we already have, can we make a weekly list
of up to 15 members who are much more likely than most to cancel
in the next four weeks?

## Pass test
Of the members on the list, at least 1 in 3 cancel within four
weeks, measured on a period the list was not built from.
Today's rate with no list: about 1 in 8.

## Real
The rule that picks the members. The test against what happened.

## Faked
One export of members and visits, made by hand. A list in a file
and one page to show it. Nothing scheduled, no logins.

## Left out
The calls, what to say on them, and whether a call keeps a member.

## What we need from Brightwell
By day 1: an export of members and visits for the last 16 weeks.
Someone who can answer questions about the data the same day.
One hour on day 5 for the demo.

## People
Head of membership: owns the question and the decision.
Membership team lead: makes the calls; checks the list makes sense.
Systems administrator: makes the export; answers data questions.

## Dates
Start Monday. Demo and write-up Friday.

The week

A week is short, and one part of it is outside your control: the data. Ask for the export before the week begins, and treat a late export as a late start, in writing.

  • Day 1: get the data and find out what is wrong with it.
  • Days 2 and 3: build the thinnest thing that works from end to end.
  • Day 4: test it against what really happened.
  • Day 5: demo it, and write up what it showed.

The next five lessons follow those days.

Set up for this course

You need Python 3.9 or later and nothing else. Create a folder called poc-week and keep every file from this course in it. Where the course writes python, use python3 if that is the command on your machine.

Key takeaways

  • A proof of concept answers one question well enough for the customer to decide what to do next.
  • Write the question so that the answer can be yes or no, and agree the pass test before you start.
  • Sort everything into three lists: built for real, faked, and left out.
  • A week is enough only if the data arrives on day one, so ask for it before the week begins.
  • Say at the start what the proof of concept will not show.

Try it

Write a proof of concept brief for a different case. A firm of surveyors spends hours each week reading building reports to find the ones that mention damp, and wants to know whether software could find them. Use the headings of the Brightwell brief. A good result has one question that can be answered yes or no, a pass test with a number in it, at least two things that will be faked, and one thing the proof of concept will not show.

Official documentation

Ready to use this in a job?

Rolebound applies to matching roles for you.

Start now

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.

    Cookies on Rolebound

    We use necessary cookies to sign you in and keep the site working — those cannot be turned off. We would also like to set analytics and advertising cookies so we can see which pages help people get interviews and which adverts brought them here. Nothing optional is set unless you say yes, and you can change your mind at any time from “Cookie settings” in the footer. Read our Privacy Policy.