Prasinus AI & software for manufacturers
Index
Contact
Licence CC BY 4.0
Phases
01–04
Artefacts
4 — one per phase
Scoped at
1–2 weeks, then 4–8 weeks
Revised
2026-08-07

The method, in four phases.

This is the whole method, written down. We publish it so that you can hold us to it, use it yourself, or give it to a different supplier.

Each phase has a written entry condition, a written exit condition, and a document that you keep. If a phase needs more time, we tell you in that week’s note.

Run the deployment check first

  1. 01

    Assess

    We walk the floor and speak with the operators. We watch a shift, read the SOPs, and ask what fails most often. Then we put the candidates in order, by return and by risk.

    Enters when
    One named process owner can answer questions the same week, and the cost problem has been named.
    We ask
    What does this cost you today, in scrap, hours or missed ship dates? Who has already said no to something like this, and why? Which system holds the correct data: the ERP, the MES, the historian, or a spreadsheet?
    Exits when
    The list puts every candidate in order. Each one has an estimated return, a named constraint that can stop it, and a deployment mode.
    Scoped at
    1–2 weeks

    You keepThe written, ranked opportunity list

  2. 02

    Pilot

    One scoped problem. We agree one success measurement in writing before we start.

    Enters when
    The person who owns the outcome signs the success measurement. We also hold a copy of real data.
    We ask
    What number, measured how, over what period, would make you deploy this? What number would make you stop? Who reviews the cases the system is unsure about?
    Exits when
    The system runs on real data against a held-out evaluation set, and the measured result is recorded beside the agreed metric, whether it passed or failed.
    Scoped at
    4–8 weeks

    You keepThe pilot system and the evaluation set

  3. 03

    Deploy

    We make the system ready for production. We connect it to the ERP, MES and historian. We add monitoring, we record the failure modes, and we train the operators on their shift. We record the model by the checksum of its weight file.

    Enters when
    The pilot met its metric, the hardware or the account exists, and change control has a window.
    We ask
    What happens on the line when this is wrong? What happens when it is unavailable? Who gets paged, and what do they do before they page us?
    Exits when
    The system has completed one full production cycle with monitoring on. We have walked the failure-mode list with the shift that operates it.
    Scoped at
    Per site. It depends on your change-control window.

    You keepRunbook, monitoring and the failure-mode list

  4. 04

    Hand off

    Your team runs the system. The source code, the documentation and the training are part of the price.

    Enters when
    Somebody on your side has been named to own it and has been in the room since Deploy.
    We ask
    Can your named owner redeploy it from the repository, without us, on a call we are not on? Can they roll a model version back? Do they know what the evaluation set is for?
    Exits when
    Your owner has done one deploy and one rollback, from your repository, without help from us.
    Scoped at
    Part of the engagement

    You keepThe repository and the admin credentials

What this method does not claim

The method does not claim that every factory needs AI. Sometimes the assess phase finds no candidate that is good enough this year. We write that down.

Prasinus is new and has no client history. This document describes how the engagement operates.

Licence

The method, in four phases · Prasinus · Rev C Licensed CC BY 4.0 — creativecommons.org/licenses/by/4.0/ You may copy it, adapt it, and use it commercially, including to run the engagement without us. Keep the attribution line and say if you changed it.

The format follows published method documents such as thoughtbot’s Playbook and Basecamp’s Shape Up.

Questions: ltplace@umich.edu