The Smart Factory Bottleneck Is the Interfaces, Not the Model
Plant AI stalls where the MES, the historian, the PLCs, the ERP and the floor disagree about what a part or a batch is, so owning that shared vocabulary matters more than a smarter model.
Manufacturing
Standard work for your knowledge work.
A lot of what limits a shop happens away from the machines: the RFQ waiting on the one person who can price it, the print revision checked from memory, the setup only one operator knows. That’s document work, it repeats every day, and most of it never shows up in the ERP.
- Experience
- West Michigan manufacturers
- Our deepest work so far, as of fall 2026
- Your ERP
- Stays the system of record
- Ownership
- Your accounts and your files
- First conversation
- Always free
Standard work
Your floor already runs on standard work.
On the floor, standard work is the current best way to do a job. It’s written down, everyone follows it, and it changes on purpose when someone finds a better way. Plenty of shops have that for the machines and still run the office on memory, so the quote, the print check and the setup notes depend on who happens to be in that day.
An Operating Standard is the same discipline for the knowledge work. It’s a written record of how your quotes, prints, instructions and reports get handled, kept in plain files the company owns, and your people and your AI tools both work from it.
Holden’s father was a hands-on manufacturing operations consultant who moved machines, changed processes and standardized work. He re-engineered the physical operation, and we re-engineer the information operation.
The sheet in words
- Step 10: Check print rev against BOM rev. Key point: Revs must match. Reason: A stale rev is scrap. Source: Print; BOM. Last checked 2026-09-14.
- Step 20: Confirm material cert on file. Key point: Cert meets spec. Reason: The spec requires it. Source: Cert file. Last checked 2026-09-14.
- Step 30: Attach current work instruction. Key point: Latest rev only. Reason: Same setup each time. Source: Doc control. Last checked 2026-09-02.
- Step 40: Flag what no rule covers. Key point: Ask; never guess. Reason: Judgment stays human. Source: Owner rule. Last checked 2026-08-28.
The work
The paper around the machine time
The system reads, checks and drafts. A person approves anything that goes to a customer or out to the floor.
01
Quoting and RFQs
An RFQ comes in as an email, a PDF or a print. Someone decides whether it’s worth quoting, finds a comparable past job, checks material and capacity, prices it and writes it up, and often all of that waits on one person.
- The systemReads the RFQ and the prints, pulls comparable past jobs and current costs, flags customer terms that need a closer look, and drafts the quote with its reasoning written down.
- People approveThe estimator sets price and lead time and approves every quote before it goes out.
02
Prints, revisions and BOM checks
Someone has to check each new revision against the bill of materials, catch the callout that changed between revs, and confirm the traveler matches the current print. It’s careful work, easy to rush at the end of a shift, and expensive when something gets missed.
- The systemReads the print and the BOM side by side, runs the exact comparisons in code, and lists every difference along with where it found it.
- People approveEngineering works the exception list, and nothing unresolved gets released as final.
03
Quality and customer paperwork
PPAP packages, first-article reports, material certs and supplier trackers are the paperwork a customer or an auditor asks for, and they usually live in folders and spreadsheets that drift out of date.
- The systemGathers what a package needs, checks each document against the requirement and the current revision, and lists what’s missing or expired.
- People approveYour quality lead signs off on every package before it goes to a customer.
04
Work instructions, captured near the work
Some setups only one person knows how to run, and some travelers say less than the operator knows. Writing it down has always been the right idea, and it has always lost to the next hot job.
- The systemTurns what an operator shows and explains at the machine into a draft instruction or traveler in your format, tied to the part and the current revision.
- People approveThe operator checks it, and engineering and quality approve it before it goes on the floor.
05
The knowledge in your senior people’s heads
Customer quirks, supplier workarounds and the reason a job runs the way it does tend to sit with a few experienced people, and they leave when those people do.
- The systemYour Operating Standard keeps it in plain files the company owns, each line with its source and the date it was checked, so every person and every AI tool starts from the same facts.
- People approveThe people who know it confirm it, and when the record and the shop disagree, the record gets fixed the same day.
06
Reports and applications the plant owns
The ERP holds the numbers, but the view people need gets rebuilt by hand in a spreadsheet every week or every month-end.
- The systemBuilds the report from your ERP exports with its checks written in code, or a small internal app when several people need the same view, running on machines and accounts the company owns.
- People approveThe person who owns the number compares the first runs against their own work and accepts it before anyone relies on it.
Quoting
One RFQ, from the inbox to a reviewed quote
Every shop quotes a little differently, so treat this as the general shape. The model reads and drafts, code handles the lookups and the math, and your estimator approves every quote.
Step by step
- An RFQ arrives by email with prints attached.
- The system checks whether it’s worth quoting against the rules you’ve written down and flags anything that doesn’t fit. A person decides what gets declined.
- It finds the closest past jobs and shows what was quoted then and why.
- It checks material and capacity against current data, and it won’t draft from costs too old to trust.
- It writes a draft quote, and every number traces back to where it came from.
- The estimator reviews the draft, sets price and lead time, and approves it.
- The approved quote goes out, and the estimator’s corrections are kept so the next draft starts closer.
What it’s worth
How we judge the return in a shop
We agree with you up front on how to measure the process and record how it runs today, so no improvement gets claimed without a before-measurement. Then we judge the result three ways, and we don’t count the same gain twice.
01
Capacity
Whether the people you already have can take on more quotes and releases without one person setting the pace.
02
Output
Whether you get useful work you weren’t getting before, like the RFQ that used to go unquoted, the work instruction that never got written, or the report nobody had time to build.
03
Quality and resilience
Whether a quote carries its reasoning and a revision change gets caught the same way every time, and whether the work keeps moving through a vacation or a retirement because less of it depends on what one person remembers.
When there are results worth publishing, they’ll appear here with their source.
How we work
How an engagement runs
Every engagement is scoped independently, so there’s no fixed package and no published price. The order of the steps stays the same. Conversations are always free.
Every step in detailOP 010
A conversation
You getWhether it’s worth doing
Walk us through the process that keeps stalling: what comes in, who touches it, what goes out and how often. If it isn’t worth your money, we’ll say so.
OP 020
Working sessions at the desks and on the floor
You getAn agreed first process
When the first conversation shows a real path, this is usually the next step, and it’s typically free too. We walk the job with the people who own it, at the whiteboard, at their desks and out on the floor, and agree on what a better result would look like. If you’d rather begin with an Opportunity Assessment, which is half a day to a full day on site and a written plan you keep either way, we can start there instead.
OP 030
A scoped first project
You getA written scope you approve
We write the first project down in plain language: the process, the result we’re after, what’s left out, and how we’ll test that it’s done. We also agree with you on how to measure the process as it runs today, and you approve the scope before any work starts.
OP 040
Feasibility, when the outcome is uncertain
You getGo, change course or stop
When it isn’t clear that your documents can support a reliable result, we test that first on your real prints and jobs in a small paid step. It ends with a written go, change course or stop, and stop is an acceptable answer.
OP 050
Foundation
You getA foundation the company owns
We set up your Operating Standard in your company’s accounts under your logins, with passwords kept in your password manager. Your IT provider is in the loop from the start, and two of your own people can reach everything without us.
OP 060
One process at a time
You getA working process, measured against the old way
We build it at the desks of the people who do the work, on real jobs. They decide where a person checks the result, and those checkpoints stay. The new way runs next to the old one until the results match, and you get a clear view of progress plus a short written update on a regular schedule, reviewed by Holden before it’s sent.
OP 070
Training
You getPeople who can run it
We train your team inside the company, in the room first and then at each person’s station. Your Champions, the people who will own the system, learn to change it themselves.
OP 080
Handoff
You getThe code, instructions, settings and accounts
We’re done when your people have shown they can run it, fix it and change it without us. Ongoing support, which we call Cadence, is optional and month to month, and you can cancel any month.
Your systems
The ERP stays the ERP.
Nobody needs a second system of record. What we build sits next to the ERP you already run. It reads from approved views and exports, writes back only through paths your team has approved and tested, and leaves the ERP in charge of its own records.
When there’s a better way in, like an interface the ERP vendor supports, we try it on your data before we build on it.
Field notes
What we’ve learned inside shops
Our deepest work so far has been inside West Michigan manufacturers. These are patterns from that work, written generally so no client can be recognized.
01
A quote is two jobs wearing one name.
Part of it is judgment about whether to chase the job, what the risk is and what the customer will accept. The rest is mechanics, like reading the print, building the parts list and pulling current prices. The mix differs from shop to shop, and either way we keep the judgment with the estimator and build the mechanics around them.
02
Build on what your people already made.
When someone on your team has already built a tool that works, we don’t start over. We make sure the company owns it, meaning the source, the accounts and the instructions to rebuild it, and the old version keeps running until its new home is proven.
03
The ERP has the data but not always the view.
A report someone rebuilds by hand every week or every month-end is often a good first project. The numbers already exist, the person who builds it knows what right looks like, and the work never touches the ERP’s records.
04
Bring IT in before the first build.
Your IT provider, your email and file systems, and your customers’ confidentiality terms decide what an AI tool may read. We settle that early, and customer drawings get a decision before any AI reads them.
05
A no-go is a real outcome.
When nobody knows yet whether your documents can support a reliable result, a small paid test on real jobs answers the question. Learning that early beats learning it halfway through a build, and when the answer is stop, we say so.
06
The people who know the job have to check it.
Some of what matters lives with the people doing the work and never made it into a document. They check that the system has understood it before anyone relies on it, and their corrections go into the Operating Standard.
We don’t publish a client name, a logo or a result without the client’s sign-off, and every number we publish has a source.
Who you’ll work with
Holden Richardson
Founder and principal
He’s the person you work with, from the first conversation through handoff, and the working sessions happen on site with the people who do the work.
The honest list
What we’ll tell you not to build
We’d rather say it early, before you spend anything.
- A customer-facing chatbot as the first project. It’s visible, it does little for the work behind the website, and when it fails it fails in front of customers.
- Anything that depends on data nobody maintains. If the source is wrong, the system will be wrong with confidence.
- A process that runs twice a year. The setup will never pay back.
- Letting an AI tool set a price or promise a lead time. Those are commitments to a customer, and a person makes them.
- Replacing the ERP to make room for AI. Keep the system of record and build next to it.
- Checking that needs a person’s judgment every single time. Some checking is judgment, and it stays with a person.
Further reading
Reading for manufacturers
Longer pieces on the same problems, each written to be useful whether or not you ever call us.
All articlesYou Are the Evaluation Body: Testing AI Against a Physical Process
Nobody outside your plant is testing AI against your production line, so you build the acceptance test yourself, with a frozen baseline and a number written down before you look.
How to Pick Your First AI Project
Score candidate projects on business value, workflow fit, evidence, measurable outcomes, risk and ownership.
What an AI Opportunity Assessment Should Actually Produce
An assessment should leave you with measured processes, ranked opportunities, a list of what not to build, and a brief for the first project.
Questions
Questions shops ask
Do we have to replace our ERP?
No. The ERP stays the system of record. What we build reads from it through approved views and exports and writes back only through paths your team has approved and tested.
How is this different from the AI features in our ERP or office software?
Those features work inside one vendor’s product, and some of them are worth using. The work that stalls usually crosses systems and people: the RFQ in email, the print on a shared drive, the costs in the ERP and the judgment in an estimator’s head. We build around how your company does that work, write it into an Operating Standard you own, and leave the ERP in charge of its records. If a tool you already pay for covers the job, we’ll tell you to use it.
Who from our side needs to be involved?
Someone who can make decisions, the people who do the work we’re looking at, one or two people who will own it afterward (we call them Champions), and your IT provider. The people who do the work matter most, because the system is built around how they do it.
Can an AI tool read our customers’ drawings?
Only once you’ve decided it can. Customer drawings often come with confidentiality terms, so we work out with you and your IT provider which documents an AI service may read, and where, before anything reads them. If a customer agreement rules something out, it stays out.
What does a first project usually look like?
One process with a clear start and end, like an RFQ going from the inbox to a reviewed draft, or a recurring report rebuilt from ERP exports with its checks written in code. It’s scoped in writing before work starts, and when the outcome is uncertain it starts with a small paid feasibility step.
How do you judge whether it worked?
Any result we claim is compared with how the process ran before, measured the way we agreed with you at the start. We look at the capacity it gives back, the quality of the work and the review and correction it takes, and every number we report has a source.
What does it cost?
Every engagement is scoped independently and quoted directly, and you approve the scope before work starts. Conversations are always free, and when the first conversation shows a real path, the working session after it is typically free too.
Conversations are always free.
Tell us which job in the office waits longest on one person, and we’ll tell you whether AI can help.