Floor-First: my way to change an organization, six weeks at a time

The five-part enterprise architecture series summarized the textbook approach. It is sound and complete, but too slow for most factories. By the time the interviews, worksheets and blueprint are done, months have passed, the people on the floor have received nothing, and the plan is already going stale.

This article is my own method. I call it Floor-First. It keeps the textbook principles (business leads, master data lives in one place, there is governance) but packs them into short six-week loops that start from what you can see on the floor, and it uses today's tools: AI, apps that work offline, and standards that can be checked every day.

Four principles

1. Start on the floor, not in a meeting room. I ran machines, repaired them and worked in QC and production control before I wrote code. Real problems show up within a day if you stand at the machine and watch.
2. One decision at a time, not one system at a time. Every loop must answer: who will make which decision better?
3. Something in use, fast. People on the floor believe when their own work gets easier, not when they see a plan.
4. The blueprint is alive. It fills from both directions: management sets goals and zones from the top, the floor adds boxes from the bottom, one loop at a time.

The six-step loop

The drawing at the top of this article is the loop. Each step in detail:

Step 1: one day on the floor

Spend a full day at one production line with a one-page sheet. Write down every time you see someone waiting, walking off to ask for information, or copying numbers from paper, and estimate the minutes lost.

The rows in the drawing are illustrative. At the end of the day, add up the minutes. The biggest row is what this loop fixes. No vote, no meeting; the numbers decide.

Step 2: pick one decision and the number to measure

Write it as one sentence, for example "a technician knows which machine has stopped within 5 minutes, without waiting for a phone call", and set the number you will measure at the end, such as minutes a machine waits for a technician. If it does not fit in one sentence, it is too big for one loop.

Step 3: a one-page worksheet drafted with AI

Use the Design & Track worksheet by Danairat T. (ดนัยรัฐ ธนบดีธรรมจารี) from part 5, but do not start from a blank page. Let AI draft it from the floor notes, then have the job owner and people from the floor correct it together. A prompt that works:

Here are one day of floor notes [paste notes]. Draft a Design & Track worksheet for the decision "[sentence from step 2]", with three measurable intended outcomes, stakeholders, no more than five key processes, the non-technology changes needed, and the technology needed. Where the notes do not give enough information, write a question instead of guessing.

The last sentence matters most. AI has to say what it does not know instead of making up answers. The boxes that come back as questions are what you go back to the floor to ask.

Do not send AI customer data, personal data or trade secrets. If your organization has rules about this, use an AI that runs in-house, or remove names and sensitive numbers first.

Step 4: build a slice in two weeks

Build only what serves the decision from step 2, not the whole system. I keep three rules for floor work:

• Work without a signal and send data once back online, because factory Wi-Fi is not good everywhere.
• Use master data from one place, such as machine codes from the central register. Never create a new set of codes inside the app.
• Ask for as little input as possible: choose instead of type, scan instead of copy, check values as they are entered.

Step 5: run beside paper, measure, update the blueprint

In weeks 4–5, run the app alongside paper, compare the numbers daily and fix the app based on what users say. In week 6, measure against the number from step 2. If it really improved, retire the paper and turn that box on the blueprint green.

Step 6: standards built into the work

The textbook has an architecture board review things every quarter. I move most of the checking into daily work, as checklists and tests that run on every change: does it use machine codes from the central register, does it work offline, are permissions checked on the server? I wrote rules like these down as the jk framework and use them on every system I build. The board is left to decide only the big things, such as new investment or retiring a system.

A blueprint that fills up

Management sets the blueprint's goals and zones on day one, in about half a day. Each loop then adds boxes from the floor. At six weeks a loop, a year gives about eight loops, and every box on the blueprint is real work in use, not a plan on paper.

The blueprint should be a web page or dashboard, not a PowerPoint file. Box colors come from real work status, and anyone in the plant can see what this loop is doing and what is finished.

Who is in the loop

Role	Job in the loop	Time
Sponsoring manager	Sets the goal, approves retiring paper	1 hour at the start and 1 hour at the end
Job owner	Answers the worksheet, decides between options	Half a day a week
Two people from the floor	Try it out, say what does not work	15 minutes a day while running beside paper
One developer	Watches the floor, builds, fixes, measures	Full time

A small team is why the loop is fast. I work alone from finding the cause to deploying the system, and use AI to help write code, which is what makes loops this short possible.

Numbers to track every loop

• Minutes lost per day on the chosen problem, before and after
• Time for data to reach the person who decides
• Corrections needed per week
• People still going back to paper. If this does not fall, the app is still harder than paper.

When not to use this method

• Work that must not fail, such as machine safety systems or systems that need certification before use. Design them in detail first.
• Replacing a core system across the organization, such as ERP, still needs the full textbook blueprint and plan; use the loop for the work around it.
• Organizations whose management will not let paper go. The loop never closes, because step 5 cannot happen.

Start this week

1. Pick one production line and spend a day there with the floor observation sheet.
2. Add up the minutes lost, take the biggest row, and write it as one decision sentence.
3. Let AI draft the worksheet, then review it with the job owner for an hour.

Do these three and you have a first loop ready to build, and the first box on your blueprint comes from the real floor.

Where the ideas come from

Floor-First is my own adaptation. It builds on the textbook approach to enterprise architecture, the Design & Track worksheet and enterprise blueprint layout by ดนัยรัฐ ธนบดีธรรมจารี, and the TOGAF standard. The idea of going to the floor (gemba) comes from lean manufacturing.

ดนัยรัฐ ธนบดีธรรมจารี and วรัญญา ปุณณวัฒน์. (2022). Enterprise architecture design [in Thai]. In Strategic Digital Technology Management. Sukhothai Thammathirat Open University Press. ISBN 978-616-16-2818-5

The Open Group. The TOGAF Standard. https://www.opengroup.org/togaf

The drawings are new; the floor observation sheet and blueprint examples are illustrative, not company data.