A first engagement

How a first sprint runs

A sprint is two weeks and it ends with something someone uses. What makes that possible is the work before it starts: naming the number, testing it against your data, and agreeing in writing what done means for every item. You decide again at every planning, so what you are committing to stays small.

Before the sprint starts

The first conversation

You tell us which number is under pressure and who owns it. If we cannot name that number between us, we say so and leave it there.

A scoping session

On site, with the people who own the number and the people who hold the data. Half a day.

We test it against your data

Before anything is scoped we check the assumptions hold. If one does not, you hear it the same day. A quick no is a result, and it is the cheap one.

The done-when, in writing

Every item gets a sentence that says what done means for it. We write them with you, so neither of us can move the line once the work starts.

Your go

A statement of work within a week of the scoping session: the scope, the impact we are chasing, the price and the start date. Two pages, not twenty.

The sprint: two weeks

Planned with you on the first day, reviewed with you on the last. What comes out is something a person uses, not a document about it. Usually several of these, and they do not have to be the same kind of work:

Every item says what done means

An item is done when its done-when holds, not when the code is merged. We write these with you before the work starts, so neither of us can move the line afterwards. Each item also names the number on your P&L that it serves. The done-when shows the step is finished. The number shows whether it paid. For example:

A dashboard for one production siteFor first pass yield on that site

The person who owns the number sees it against budget and against the previous period, refreshed by 07:00 with no manual step. They have used it in two consecutive weekly reviews and confirmed it works. The extract it replaces is switched off.

More reliable ingestion of one sourceFor electricity cost

The source lands every fifteen minutes instead of once a day, with retries, a freshness test on every load and an alert when a load fails. Ten consecutive days at that cadence with no failed load going unnoticed. The extra platform cost is measured and reported against the old one.

A new demand forecastFor days of finished goods inventory

Accuracy measured on a holdout period agreed at planning, with the same metric for the old forecast and the new one. The new one beats the old one. If it falls short we report the accuracy we reached and what it would take to close the gap, and you decide whether to carry on.

Retiring an app nobody usesFor platform cost

Its users are told, with the replacement named. Its data is archived where its owner says. Its pipelines and routes are removed. A week after the switch-off nothing else has broken, and the cost line is gone from the platform bill.

What we need from you

The person who owns the number

At the planning and at the review, with the authority to decide what goes into the sprint.

The domain expert

An hour at the start to set the problem statement, then the weekly reviews. Their time is the main input to the work, and the end date moves with it.

Access

The repositories, the development and production environments, and the platform cost data, by the start date.

What you commit to

Which number would you put first?

Tell us the number and who owns it. If we can name it between us, the statement of work writes itself, and you have it within a week.