Written for Operations, maintenance, quality, safety and technology leaders in manufacturing plants, and the sponsors who have to fund a first case.
What's in this playbook
Manufacturers are not short of AI ideas. They are short of first cases that can travel: a named process, a signed baseline, an operator who is not only the original builder, a person who can stand behind a work order, a reject, a line-stop or a parameter change, and a way to stop. This playbook is the single manufacturing document behind the manufacturing industry pages. It is complementary to the manufacturing design playbook already in the Resource Library — ten tests, five patterns — with predictive maintenance and asset health in depth, on purpose. It is not a near-duplicate of that design playbook, not a survey, and not a certification that a system is safe to run on a line.
It maps six families of work. Each family has an honest first case and a case that is too wide. A sense-predict-recommend assist that reads, and does not write, the historian the plant already runs — a score, the evidence it used, and a proposed work order for a named planner or technician to issue in the CMMS — is an honest first case in predictive maintenance, once the operator role has been raised. A model that automatically opens a work order, stops a line, or changes a process parameter is not. The same distinction is drawn for quality inspection (a person on every reject), production planning (a recommendation, not an unsupervised setpoint), spares and MRO, document and SOP support, and finance and administration: AI as a recommend layer first; an unsupervised shop-floor agent is not a first case this playbook will recommend.
How the tests and the method are meant to be used
Fourteen design tests are applied to every case. The answers will differ. The standard does not. A missing safety answer, an undefined "review," or an operator who is only the original builder, is a reason to pause, not permission to proceed. The safety function and counsel decide what a safety obligation or a related rule requires, and when. This playbook will not make those determinations, will not certify that a system is safe to run on a line, and will not claim a certified connector to a named industrial vendor. Hosting is decided before a case is funded. Shop-floor data may not be able to leave. Where they cannot, AsymiLink AI runs on your own hardware, with an optional air gap. Built to SOC 2 standards is the exact wording we use; it is not a safety certification.
The illustrative organization is the same hypothetical regional manufacturer as in the manufacturing design playbook, the readiness framework, and the 2026 diagnostic: overall 22, Developing, Process Automation a strength at 4, talent and governance the constraints, and a next-step target of 25, still Developing — a next step, not a timeline, and not a promise. Its honest first work is to raise the operator role, reuse a pattern the finance automation already proved, and only then to consider a historian-read-only maintenance assist — not a shop-floor agent, and not a model that opens a work order or stops a line unsupervised. The financial method is The Complete Guide to AI Automation ROI. This playbook will not invent a manufacturing-sector result. The 3.8x average first-year ROI across client engagements is an average, not a forecast or a commitment for any individual project. ROI is measured across client engagements against a baseline agreed before work starts. It is an average, not a forecast for any manufacturing project.
Six families of work
each with an honest first case, a case that is too wide, and a defined gate. Each item below is the shape of the map; the full wording, honest first cases and “too wide” warnings are in the PDF.
F1
Predictive maintenance and asset health
F2
Quality inspection
F3
Production planning and process support
F4
Spares, materials and MRO
F5
Documents, procedures and technician support
F6
Finance, administration and a proved pattern
The PDF gives, for each family, an honest first case, a case that is too wide, and a defined gate. The summaries above are the shape of the map, not a substitute for that wording.
Fourteen design tests (excerpt)
Fourteen design tests are applied to every case. The answers will differ. The standard does not. The full wording is in the PDF and the companion workbook.
T1–early
inventory and a named OT and IT boundary; proportionate consequence; a person on any work order, reject, line-stop or parameter change, with review defined; the safety function and counsel asked in time; a named source of truth, explanations and a frozen evaluation set.
Mid-tests
hosting decided up front, including where shop-floor data cannot leave; a write to a CMMS, MES, historian or parameter as a different design from a read; an audit trail that can reconstruct a day; a named operator, including on the floor, and a switch-off; training and reuse written down.
Later tests
anything that could put a person in harm's way designed with the safety function, or refused; honest notice to the people who work with the system; a bounded first case that reuses a proved pattern and raises the path; a signed baseline on the ROI method, and monitoring after launch that the organization that owns the system writes down.
A completed test sheet is not a certificate, not a compliance score, and not a finding that a system is safe or aligned with a named regime.
Get the guide and workbook
Enter your name, work email and company. The files appear on this page immediately. A backup download link is also emailed when email delivery is available.