Harmony AI
← Back to Resources
Field guide

The Manufacturing Paper Audit

Walk your own floor and count how much of it still starts as handwriting, lives in one person's spreadsheet, or never leaves a machine. Five things to measure, what each one means, and why it decides whether AI can do anything for you yet.

Read this first · How to use this

This is a walk you run yourself, station by station, in an afternoon. A clipboard or your phone is enough. It measures one thing: how much of your plant already exists as data a system can read, and how much still lives on paper, in a spreadsheet only one person maintains, or locked inside a machine that talks to nothing.

Nothing here is a number we invented. Where a figure appears it is a named public source with a link. Every other number on this page is one you produce about your own plant, because the average across other people's factories would not tell you what to fix on yours.

1. What this audit measures

Five numbers, each chosen because you can check it against your own floor in under a minute and know immediately whether it rings true. Counted together they tell you where your plant actually sits: how much of the week is data a system holds, and how much is still handwriting waiting for someone to re-type it.

They are the same five questions inside the AI Readiness Assessment. This page is the long version you walk in person. The assessment is the two-minute version that scores it for you.

What counts as a capture point

A capture point is any place a person writes something down that later has to become data: a batch sheet, a quality check, a traveler, a rounds log, a downtime note, a shift handoff. Count each one where it is first written, not everywhere it later gets copied. The count matters more than the page count, because each capture point is a separate decision about whether that record ever reaches a system at all.

A few rules so your own count stays honest

2. Why this matters before you buy AI

Manufacturing software is sold as though the data already exists. The pitch assumes a plant with records in systems, machines that report, and numbers that are current. The capture points on most floors tell a different story: much of the week still starts as handwriting at a station, and the software layer meets it only after somebody re-types it.

That gap is what this audit measures, and it matters commercially, because an agent that cannot read the record cannot act on the record. A copilot answering from a system that holds a third of the week gives confident answers about a third of the plant. So the honest first question is not which AI to buy. It is how much of your floor an AI could see today.

We are not going to quote you an industry-wide percentage here, because the only number that changes your decision is the one you are about to count on your own floor.

The five things to count

Each card is one metric: what to count, what the number is telling you, and a line to write yours. The printable worksheet at the bottom puts all five on one page you can carry onto the floor.

Metric 01 · Paper capture points

Count every distinct handwritten record in active use on the floor: batch sheets, quality checks, travelers, rounds, downtime logs, handoffs.

What it tells you: how many separate places your week begins as paper. Each one is invisible to any system until a person types it in.

Your number:  

Metric 02 · Data latency

For the numbers you rely on, sort them by how long until someone outside the station can see them: live, same shift, next day, month end.

What it tells you: the ceiling on how proactive you can be. You cannot correct a run you only hear about at month end.

Your number:  

Metric 03 · Machine data escape rate

Asset by asset, the share of your machines whose data leaves the controller at all: counts, states, faults, cycle times reaching anything beyond the panel.

What it tells you: how much of what your machines already know is actually usable. Most already measure. Few let the number out.

Your number:  

Metric 04 · Re-keying hours per week

Add up staff hours spent typing numbers that already exist somewhere else: the supervisor rebuilding the daily report, the planner re-entering the sequence, quality transcribing a check sheet, the office keying a PO.

What it tells you: the most legible cost on the list, because it converts straight into a budget conversation.

Your number:  

Metric 05 · Trace time

Time one real one-lot trace, from question to evidence, on an actual lot rather than what the SOP says.

What it tells you: the single most honest test of whether your records are connected, because a trace has to cross receiving, production, quality and shipping in one pass.

Your number:  

Each of these five is a question in the AI Readiness Assessment. If you would rather have it scored for you than walk it by hand, take the two-minute version first and come back for the walk.

3. Paper by the numbers

Metric 01 summary · Paper capture points

Count it yourself
How to count it
Walk the floor and tally each active handwritten record at the station where it is written. One tally per capture point, not per page.
What a low number looks like
Most records already start in a system or off a machine. Paper survives only where a glove or a wash-down demands it.
What a high number looks like
The week begins on paper at most stations, and your systems hold mostly what someone found time to key in later.

A capture point is any place where a person writes something down that later has to become data. The count matters more than the page count, because each capture point is a separate decision about whether that record ever reaches a system.

The record families that survive longest on paper tend to survive for a practical reason rather than a cultural one: at the point of use, paper is faster. A clipboard needs no login, works in a glove, survives a wash-down, and never waits on a network. Any digitization that ignores that loses to the clipboard. This is why the work is done at the station rather than designed in an office.

4. The latency tax

Metric 02 summary · Data latency

Count it yourself
How to count it
For each record you act on, note how long until someone outside the station could look it up. Sort into live, same shift, next day, month end.
What good looks like
The numbers you run the plant on are live or same shift. The morning meeting is about now, not about yesterday.
What bad looks like
Most numbers are next day or month end, so every meeting is a report on something already finished.

Latency is the distance between something happening on the floor and anyone outside the station being able to see that it happened. It sets a hard ceiling on how proactive a plant can be, because you cannot correct a run you only learn about at month end. A plant on next-day numbers runs a morning meeting about yesterday, which is a report, not a control loop.

The cost of that gap is arithmetic a plant can run on its own inputs rather than on ours. The Downtime Cost Calculator prices the stops, and the full set is at ROI Calculators & Tools. Those are your numbers, produced on your own floor.

5. Machine data: measured but stranded

Metric 03 summary · Machine data escape rate

Count it yourself
How to count it
Asset by asset, mark whether any data leaves the controller: counts, states, faults, cycle times reaching anything beyond the panel. Count the machines, then the share.
What good looks like
Most assets already push their data somewhere a person or a system can read without standing at the machine.
What bad looks like
The machines know their numbers and always have, but the numbers stop at the panel where nothing else can read them.

This is the one that surprises people. Most machines already know. They know their counts, their states, their faults, and their cycle times, and they have known them for years. What is missing is not instrumentation. It is the path off the machine.

The controller brands we encounter most are Rockwell/Allen-Bradley, Siemens, Omron, and Mitsubishi, and on all of them the data is reachable at the PLC over OPC UA. So the common case is not a plant that cannot measure. It is a plant whose measurements stop at the panel, where nothing else can read them.

6. The re-keying line item

Metric 04 summary · Re-keying hours per week

Count it yourself
How to count it
Ask each role how many hours a week they spend typing numbers that already exist somewhere else, then watch a few of those tasks to sanity-check the answer.
Turn it into a year
Once the weekly hours are real, the annual figure follows directly, and that is the version you take upstairs. Do not multiply a guess by fifty-two, because that just produces a more confident guess.

Re-keying is the most legible number in the audit, because it converts directly into a budget conversation. It is staff hours spent typing numbers that already exist somewhere else: the supervisor rebuilding the daily report, the planner re-entering the sequence, the quality tech transcribing a check sheet, the office keying a PO.

Once the weekly figure is measured, the annual figure follows from it directly, and that is the version a plant manager can take upstairs. We are not publishing an annualized number before the weekly one is counted, because multiplying a guess by fifty-two just produces a more confident guess.

7. Trace time under pressure

Metric 05 summary · Trace time

Count it yourself
How to count it
Pick one real lot and trace it end to end, from question to evidence, on the clock. Do not rehearse it. The first honest attempt is the number.
What good looks like
Minutes, from a screen, without leaving your chair.
What bad looks like
Hours or days, spent on phone calls and in binders, with the answer only as good as the handwriting.

Trace time is the most honest single test of whether records are connected, because a trace has to cross receiving, production, quality, and shipping in one pass. Anything disconnected shows up immediately as a phone call or a binder.

What a trace is now expected to take is not a matter of opinion. Under the FDA Food Traceability Rule, which implements section 204 of the Food Safety Modernization Act, covered firms must make required traceability records available to an authorized FDA representative within 24 hours of a request, and during an outbreak, recall, or other public health threat must provide the required information in an electronic sortable spreadsheet within that same window.

Public research  Source: 21 CFR 1.1455, paragraphs (c)(1) and (c)(3)(ii), FDA Food Traceability Rule.

Read that against a plant whose lot history lives in binders. The rule does not care how the record was captured, only how fast it can be produced and whether it arrives sortable. A paper trace can meet neither condition reliably, and customers increasingly ask for the same thing on a shorter clock than the regulator does.

8. Where the paper usually hides, by industry

The five numbers are the same everywhere, but the record that stays on paper longest tends to differ by what you make. These are patterns to check on your own floor, not statistics:

The worksheet

Get the printable audit worksheet.

Everything you need to run the walk is on this page. The worksheet just puts it on paper you can carry onto the floor, sent to your work email.

Work email only. We use it to send the worksheet and nothing else you did not ask for. Unsubscribe anytime.

Got it. The worksheet is on its way to your inbox. If you checked the box, a Harmony engineer will reach out to walk the audit with you.

9. What the plants furthest along did first

The plants that clear this audit did not get there by buying a better tool. They did the same unglamorous thing in the same order: digitize, connect, unify, before any AI. That is Phase 1, and it is the pattern the audit exists to make visible.

The two arcs

Arc 1 · The digital transformation

clipboards + Excel + ERP/software + PLCs/machines ↓ DIGITIZE, CONNECT, UNIFY ONE LIVE DATA LAYER (every record, system and machine)

Built on-site, in person, by forward-deployed engineers. This is the arc the audit measures, and the arc a plant has to finish before the second one is real.

Arc 2 · The AI transformation

the live data layer is the foundation, and on top of it sit · AI SEARCH (answers with sources cited) · AI AGENTS (propose, a person approves) · AI PRODUCTION SCHEDULING (the plan proposes, the planner approves) · BACK-OFFICE AI AUTOMATIONS (finance, sales, procurement, logistics)

Everything in this arc reads from the layer built in Arc 1. Skip the foundation and each item above is a demo, not a tool.

The three phases

Phase 1

Lay the Data Foundation · Digitization

Every pen-and-paper record digitized at the station, every software system connected, and all of the data unified into one live layer. The digital transformation starts here.

Phase 2

Production & Operations Scale

Factory operations turn proactive: live sensors and machine data, the AI scheduling board, predictive maintenance before failure.

Phase 3

AI-Native Operations

Agents across the floor and the back office act on the live layer: quality signals, reports, copilots. Humans approve.

A high paper score means a plant is at the start of Phase 1, whatever it has been sold. That is not a criticism of the plant. It is a description of the sequence.

10. How to run the audit

The whole walk, step by step. It takes an afternoon, not a week:

  1. Walk the floor. Go station by station with a line lead and list every capture point in active use during a normal production week.
  2. Classify each one. Paper, single-owner spreadsheet, shared system, or machine-generated. A record that lives in more than one place gets counted at its origin.
  3. Time the latency. For each record, how long until someone outside that station could look it up. Check it against a real example, not the process document.
  4. Check what leaves the machines. Asset by asset, whether counts, states, faults, or cycle times reach anything past the controller.
  5. Count the re-keying and run one real trace. Hours per week by role, and one actual lot traced end to end on the clock.
  6. Score it. Five numbers and a record inventory for your own plant. That is the whole deliverable, and it is yours whether or not you ever talk to us.

The printable worksheet above lays all of this out as one sheet with a row per station. Ask for it and it comes to your inbox.

Or have someone walk it with you

The same walk is the first week of a Harmony pilot, on-site, with forward-deployed engineers doing the counting alongside your team. Published at $15,000 to $20,000 one time, over 4 to 6 weeks, with working software running by the end of the pilot. The audit is a byproduct of work that was happening anyway, which is why it is a floor walk and not a questionnaire. Check the box on the worksheet form above, or get in touch directly.

Want to see your plant's number without doing the walk yourself?

A Harmony pilot starts with exactly this count, on-site, with engineers in your plant. $15,000 to $20,000 one time, 4 to 6 weeks, working software by the end of the pilot. Phase 1 first, because that is the order it has to happen in. See what the live layer looks like.

Book My Demo →
← Back to Resources