Yield and Rework Analysis
Most optics shops collect scrap and rework data but rarely analyse it across jobs. An AI layer over the ERP can answer questions like which materials and tolerance classes lose the most pieces, which operation produces the most late scrap, and whether yield changed after a process or supplier change. The value is speed: the analyst asks a question in plain language and sees the query and the result, rather than waiting for a custom report.
The data must be coded well, because AI cannot infer reasons that nobody recorded.
RFQ Triage
Optics shops receive requests that range from catalog-like windows to unproven aspheres. AI can read the request, extract the key specs and flag fit: materials you do not process, tolerances beyond your history, export-control indicators or unusually short lead times. It can also surface similar past jobs and the quoted-versus-actual result.
A person decides whether to quote, but the first pass takes minutes and the estimator spends time on the bids worth pursuing.
- Fit check against processes, materials and equipment
- Tolerance check against historical capability
- Export-control and customer-restriction flags
- Similar past jobs and their actual margin
Coating and Metrology Queue Questions
Planners spend time answering questions such as which jobs are ready for the next coating run, what is waiting on the interferometer and which orders will miss their dates. An assistant connected to open orders and routings can answer these with the underlying records. It does not replace the planner, but it removes a lot of manual lookup and spreadsheet reconciliation.
Try it here, free
Manufacturing AI Readiness Assessment
Score your manufacturing operation's readiness for AI across data, systems, people, and governance, and get a prioritized roadmap for closing the gaps.
Keeping It Private and Controlled
Optics data includes customer prints, process recipes, coating designs and sometimes export-controlled technical data. Running open-weight models on your own GPUs, with retrieval over ERP data and documents, keeps this inside your boundary. Netray builds this pattern with read-only ERP access, shown queries and human approval on any write-back.
The customer owns its compliance program, and the architecture should support it with role-based access and logging.
When It Is Not Worth Doing
If your data is mostly on paper, if scrap is not recorded or if you make a handful of repeat parts, AI will not fix the underlying gaps. Start by cleaning up reason codes, travelers and job history. A simpler dashboard or a monthly review may deliver more value than a model, and the vendor's built-in reporting is often enough for basic questions.
Frequently asked questions
How can AI be used in optics manufacturing?
Practical uses include extracting specs from RFQs and drawings, drafting quotes from similar past jobs, analysing yield and rework by material and tolerance, answering planner questions about queues and assembling inspection records. These work best on your own ERP and quality data rather than generic models.
Does AI replace optical engineers or estimators?
No. It reduces searching and retyping, but decisions about feasibility, price, process and conformance remain with qualified people. The best results come when estimators and engineers review a grounded draft rather than start from a blank page.
What data do I need before starting?
Reasonably clean job history with operations, hours and scrap reasons, coded part attributes such as material and tolerance class, and accessible documents like prints and inspection files. If these are missing, fix recording first, since AI cannot recover information that was never captured.
If you want to see what your own yield and quoting data can answer, ask Netray for a scoped, on-prem pilot over your ERP and a sample of inspection files.