Many of your people are already
using AI with customer data.
Most $20M to $300M manufacturers we meet are in the same place: staff on personal AI accounts, an ERP reimplementation underway, the one person who understands the data about to retire, and a customer asking how their data is protected. An assessment tells you what is exposed, what to automate first, and what it is worth, in three weeks for a fixed fee.
The plant produces data all day. Almost none of it reaches a decision.
Production and inventory live in the ERP. Quotes start on paper. A large customer sends identical-format PDF orders that one person re-keys every morning. Quality is a spreadsheet. The Monday report is somebody's export, out of date by the time it lands. And in the meantime, a dozen or more employees have quietly started using AI chat tools on their own accounts to get through the day, with whatever data was in front of them.
None of that is a software problem, and no AI model fixes it. It is that nothing connects and nobody has written down what is allowed. We fix those two things first, then build the automation and reporting that pay for themselves.
- A written inventory of every system and who depends on it, before the retirement
- An AI use policy that fits the customer contracts and NDAs you signed
- Order entry, quoting, and three-way match automated where the hours actually go
- One source of truth across ERP, quality, and scheduling, feeding your dashboards
How Manufacturers Work With Us
From a recent discovery at a manufacturer, anonymized.
A family-owned manufacturer mid-way through an ERP reimplementation. These are the findings that set the roadmap.
People using AI on personal plans
HR, finance, and email. One staffer spent four hours find-and-replacing customer names in a file to feel safe uploading it. No written policy, and NDAs signed by sales that nobody downstream could see.
To key one batch of prints
Roughly 300 prints per estimate, entered by hand. Quoting started on paper and was stamped by hand. The automation case wrote itself once the hours were counted.
Of revenue through one person's keyboard
One customer, half the business, identical-format PDF orders re-keyed into the ERP every day. First on the roadmap, and the pilot that pays for the rest.
Until the one person who knows the data retires
The IT manager who understood the database was leaving within the year. The inventory became the handover document, and the warehouse became the way the knowledge stopped living in one chair.
Built on the manufacturer's own servers,
run by the manufacturer's own team.
A mid-size manufacturer
Went from ERP exports and Monday-morning spreadsheets to a governed warehouse and dashboards, with one definition per metric and loads that run on their own. It sits on the client's own servers, and the client's team runs it.
Read the case study →What we build underneath the dashboards
- Automated extraction from your ERP and the systems around it
- A governed staging layer: one definition of scrap, one of on-time
- Dashboards through your own gateway, maintained by your own staff
- Private LLM deployment for data that cannot leave the building
What owners and presidents ask us
about AI on the floor and in the office.
Our ERP vendor already quoted us automation. Why not just buy that?
Sometimes you should, and we will say so. Vendor automation modules are priced per data source and tend to automate what the ERP can see. The assessment looks at the whole flow, including the quotes on paper, the PDFs one person re-keys, and the spreadsheets outside the ERP, and tells you which pieces the vendor tool covers and which it does not.
We are in the middle of an ERP reimplementation. Is this the wrong time?
It is usually the right time. A reimplementation is when data gets cleaned, definitions get decided, and twenty years of history gets migrated or dropped. The assessment makes sure those decisions are made with AI and reporting in mind rather than discovered afterward.
Can AI run on our data without it leaving the building?
Yes. We deploy private large language models inside a client's own environment when customer contracts, NDAs, or plain preference require it. We also tell you honestly when a hosted service with the right contract is the better answer. We do not resell either, so the recommendation is not tied to a commission.
What do our customer contracts have to do with AI?
If staff are pasting customer drawings or order data into personal AI accounts, that is a data handling practice your customers' security requirements and NDAs almost certainly cover. The governance section of the assessment maps your current AI use against the contracts you have signed, so there are no surprises at the next customer audit.
The person who knows our data is retiring. Where do we start?
With the inventory. The first deliverable of the assessment is a written map of every system, who owns it, and what depends on it. That document is worth the fee on its own in a plant where the answer currently lives in one person's head.
Who does the work?
The people who scope the work are the people who deliver it. You will not be handed off after the sales call.