The problem
Identifying a spare part shouldn't take a full workday.
In every manufacturing company the process repeats the same way: a broken component, a faded marking, hours on the phone with the vendor, three people digging through the ERP. Meanwhile, the line is down.
Paper documentation
Catalogues as PDFs, scanned technical drawings, photocopied spec sheets. Information everywhere and nowhere.
Dirty master data
Duplicate codes, inconsistent descriptions, missing fields. 15–20% of codes in an industrial ERP are duplicates or misclassified.
Tacit knowledge
Only John knows where that part is. When John retires, 30 years of experience walk out of the company.
No tools in the field
The technician walks back to the office, boots the PC, opens SAP, searches, doesn't find, calls a colleague. Downtime keeps ticking.
Inefficient procurement
You re-buy what you already have. You pay the OEM premium when equivalents exist. Stock grows, cash is locked.
How you search a part
Six search modalities. One answer: the right code in your ERP.
The operator has the part in hand, a fragment of a code, a vague description, an old invoice. OpSight Maintain combines every piece of information available — even partial, even contradictory — and maps it to the right record in your ERP.
Visual search
A photo from a smartphone. Vision models recognise the component even with poor lighting, dirt and awkward angles.
Labels and printed codes
Industrial OCR on markings, nameplates, abrasive labels. Extracts serial numbers, models and batch codes even from degraded text.
QR, barcode, data matrix
1D/2D code scanning from the camera. Direct integration with OEM part numbers.
Natural-language semantic search
"Radial ball bearing 25 mm inner diameter". The engine understands synonyms, abbreviations, typos and multiple languages — even when the part is written in a different language in the catalogue.
Bill of materials and manuals
Extraction of part codes from BoMs, maintenance manuals, technical drawings. The part is always in the context of its plant.
Photo quality precheck
An automatic check evaluates the image before consuming AI resources: if the photo is blurry, too dark or does not frame the part, the app immediately asks for a retake.
How it works
A proprietary end-to-end method.
Subject to a patent application.
OpSight Maintain is not a search engine: it's a closed loop that goes from the photo of a part to the code in the ERP, through data enrichment, quality checks and operator feedback. The process is measurable, versioned and auditable — every step leaves a trace.
STAGE 01
Capture
The operator photographs the part at the machine with the mobile app — even offline. The quality precheck evaluates the image immediately; industrial OCR extracts markings, codes and serials even from degraded labels.
STAGE 02
Recognise
The AI pipeline cross-references the evidence with the ERP catalogue: exact match on the manufacturer code, multi-tier fuzzy match on text and family, visual confirmation. The component is classified into the merchandise hierarchy — 46 families, 300+ groups — with a confidence level for every proposal.
STAGE 03
Validate and integrate
The proposal reaches the back-office console, where the supervisor confirms or corrects it. The validated code flows back to the ERP via SAP export templates, and every correction feeds the continuous-learning loop.
The technical detail of the method is described in the filed patent application.
Why OpSight Maintain
Built for real industry,
not for a sales slide.
In production, not in demo
The platform is live in multi-plant enterprise environments, with real operators on the line in Europe and the United States. The flows you see are the ones running every day.
Offline-first mobile app
iOS and Android, distributed via corporate MDM (Intune, Workspace ONE, SOTI). The app works offline and syncs when back online: remote sites, warehouses without coverage.
Back-office validation console
Worklists for supervisors and specialists, shared help requests, AI-vs-operator divergence analysis: coding quality control is part of the product.
Native SAP integration
Direct match on the material catalogue, export templates for bulk loading, manufacturer-code mapping. Documented REST APIs for custom systems.
Continuous learning
Every operator feedback improves the system: corrections feed the knowledge graph and the AI prompts. The platform grows with your company, every day.
Enterprise-grade security
SSO with Entra ID (OIDC/SAML), tenant isolation, audit log, at-rest and in-transit encryption. Autoscaled AI pipeline on Azure, data in EU regions. GDPR compliant.
Traction
In production at a global tissue manufacturing leader — multi-plant, including sites in the United States — with apps published on enterprise stores and components processed every day.
Who it's for
One tool, five business functions.
Maintenance & MRO
Rapid spare-part identification at the machine. Less downtime, higher first-time-fix.
Warehouse & Logistics
Automatic inbound acceptance, PO reconciliation, expediting, duplicate control.
Procurement
Reduction of duplicate codes, cross-OEM equivalent identification, sourcing decisions based on clean data.
After-sales & Service
Customer portal to identify parts autonomously. Fewer support calls, more parts sold.
Operations & Leadership
Dashboards on identification times, master-data quality, warehouse KPIs. Evidence for continuous improvement.
Integrations
Speaks natively with the systems you already use.
OpSight Maintain does not ask you to drop your ERP: it plugs on top of it. Ready-made connectors, two-way sync, writing of codes into the right master-data fields.
Implementation
Your data in.
We do the rest.
Up and running in a few weeks, without disrupting your existing processes. We start from a subset of your catalogue, validate the model, then extend.
- Week 1 Catalogue analysis, pilot scope definition, environment setup
- Week 2–3 ERP data ingestion, model calibration, reference photo collection
- Week 4–6 Operational pilot with a small team, active feedback loop
- Week 6+ Progressive rollout, user training, measured benefit KPIs
FAQ
Frequently asked questions
Do we need to train the model on our spare parts before starting? +
No. OpSight Maintain' VLM models are pre-trained on the industrial domain and immediately recognise common families (bearings, belts, motors, sensors, valves). Fine-tuning on your specific parts happens continuously through operator feedback, without dedicated training cycles.
How do you handle sensitive data? Do you send it to OpenAI? +
It depends on the client. For those with strict constraints we offer on-premises deployment with local models (no data leaves the perimeter). For cloud-accepting clients, we use enterprise providers with zero data retention and GDPR-compliant DPA contracts.
What happens if the photo is bad? +
An automatic pre-check evaluates image quality before spending AI resources. If the photo is out of focus, too dark or doesn't frame the part, the app asks the operator to re-shoot. Fewer costs, fewer false positives.
Does it work offline? +
Yes. The mobile app is offline-first: photo, OCR and basic classification run locally on the device. ERP matching happens at the first reconnection. Built for remote sites, ships and plants without coverage.
How long to see a ROI? +
First measurable results arrive during the pilot (weeks 4–6): average identification time, first-hit match rate, duplicate codes identified. Full ROI depends on catalogue size and usage frequency, but is typically under 12 months.
Can I integrate it with my custom systems? +
Yes. Beyond the native SAP export flow, OpSight Maintain exposes documented REST APIs to integrate the platform with existing ERPs and workflows.