Manufacturing
Machine monitoring platforms: who owns the data from your machines?
By SourceX Editorial · Reviewed by Noah Loul ·
Short answer
Machine monitoring data ownership is set by your platform subscription terms, not by who owns the machine. Contracts range from vendor owns, to customer owns with a license back to the vendor, to customer owns outright. Rule of thumb: unless the terms call raw machine data yours and guarantee export at termination, assume your rights are limited.
Key takeaways
- The machine belongs to the plant, but the platform contract decides who may use the signals it streams.
- Read four defined terms first: customer data, usage data, aggregated data and derived data.
- A vendor license for product improvement is normal; a license to train models offered to others deserves scrutiny.
- Operator-entered downtime reasons are often more useful than raw spindle signals, and they carry more personal detail.
- Negotiate ownership, raw export rights and post-termination deletion at purchase or renewal, when you have leverage.
What data does a machine monitoring platform actually collect?#
A machine monitoring platform collects two kinds of records: signals read from the machine control and context typed in by people. The signals include machine state, cycle start and stop, spindle load, feed and speed overrides, alarms and part counts, usually pulled through an adapter, an edge device or protocols such as MTConnect or OPC UA.
The context layer is where most of the meaning lives. Operators pick downtime reason codes, supervisors annotate shifts, and the platform computes OEE, utilization and cycle time against standard. For an AI developer, the pairing of a signal with the human explanation for it is what makes a plant record useful.
| Data layer | Examples | Typical source | Rights question to ask |
|---|---|---|---|
| Raw controller signals | State, alarms, spindle load, part counts | Control via adapter or edge device | Is it defined as customer data or as service data? |
| Operator input | Downtime reasons, scrap reasons, shift notes | Tablet or terminal at the machine | Does it carry operator names or badge IDs? |
| Calculated metrics | OEE, utilization, cycle time against standard | Platform analytics | Does the vendor claim derived data? |
| Benchmarks | Comparisons with other plants | The vendor's pooled dataset | Is your history inside someone else's benchmark? |
| Integration data | Job numbers, part numbers, schedules | ERP or MES connector | Do part numbers identify your customers? |
Who owns the data? The clause spectrum#
Ownership of machine monitoring data sits on a spectrum, and the subscription agreement places you on it. Few contracts use the same words, so read the definitions section before the license grants: a clause can say you own your data while defining that data narrowly enough to leave out raw signals.
Many mid-market plants land in the middle of the spectrum. The customer owns customer data, and the vendor receives a license to use it, often in aggregated or de-identified form, to run and improve the service. Whether that license reaches model training, resale or benchmarks sold to other customers depends on the exact wording.
| Position | Typical wording, paraphrased | What it means for licensing |
|---|---|---|
| Vendor owns | All data generated by the service belongs to the provider | You may use reports internally but may need permission to license or move raw history. |
| Customer owns, broad license back | Customer owns its data; provider may use it for any lawful purpose | You hold title, but the vendor may already be using the same history commercially. |
| Customer owns, limited license back | Provider may use aggregated, de-identified data to improve its services | Licensing rights are usually intact; confirm whether improve covers model training. |
| Customer owns outright | Provider processes customer data only to deliver the services | Strongest position, still subject to customer and employee obligations. |
| Silent or unclear | No definition of machine data at all | Treat as unresolved and ask the vendor in writing before relying on it. |
Which clauses should you read first?#
The clauses to read first are the definitions, the license grants, the restrictions on you, and export and termination. Those four tell you more than any product page about who controls the history in the system.
Check the machine side as well. Controls suppliers and machine builders, including Fanuc, Siemens, Rockwell Automation and Haas, offer connectivity, analytics or remote service software, and that software typically comes with its own license terms. Equipment bought with a service agreement may also send diagnostic data to the builder separately from your monitoring platform. Read those terms rather than assuming either way, and log each one alongside the platform agreement.
- Definitions: how the contract defines customer data, machine data, usage data, aggregated data and derived data.
- License to the vendor: the purposes it covers, such as running the service, product improvement, benchmarking or training machine learning models.
- Restrictions on you: limits on bulk export, automated extraction, reverse engineering or sharing outputs with third parties.
- Export and termination: the export format, how far back exports reach, and how long the vendor keeps data after you leave.
- Confidentiality: whether your production data counts as your confidential information and how the vendor must protect it.
- Incorporated documents: order forms and online policies referenced by the agreement, which can change what the master terms appear to say.
Why machine data rights matter for AI licensing#
Machine data rights matter because AI developers building models for physical operations need real production histories, and their diligence asks the supplier to show it can license what it delivers. Expect questions about who generated the data, under which terms it was collected and what permitted use the supplier can grant.
The strongest packages pair machine signals with the plant's own decision records: the job, the downtime reason, the maintenance work order and the quality outcome. If the platform terms restrict export or claim derived metrics, you may still license the operator notes and ERP side of the story, but the package loses its timeline of what happened on the floor.
Customer obligations sit on top of vendor terms. Job names, part numbers and program files can identify whose parts you were cutting, and some customer terms treat production information about their parts as their confidential information.
Illustrative: a contract machining shop reads its monitoring contract#
Illustrative: a fictional family-owned contract machining company runs horizontal machining centers and lathes connected to a cloud monitoring platform through edge devices. Operators enter downtime reasons on tablets, and the platform links each cycle to the job number pulled from the shop's ERP.
When the owner considers licensing several years of production history, the IT director reads the subscription agreement. Customer data is defined to include operator input but not raw machine signals, which the vendor classes as service data it may use to improve and train its products. The export function covers summary reports, not the raw event stream.
The owner decides to negotiate at renewal instead of relying on the current wording, asking for raw event exports and a clause stating that machine data is customer data. In the meantime, the licensable scope is set to operator-entered reasons, maintenance work orders and ERP job records, with customer part numbers replaced by internal codes.
What to add to the contract at purchase or renewal#
The best time to settle machine data ownership is at purchase or renewal, when the vendor wants the signature. Mid-market plants often have more leverage than they assume, especially on multi-site rollouts or when switching from a competing platform. Ask for the positions below in the signed agreement or order form, not in a sales email.
| Ask | Suggested position | Why it helps |
|---|---|---|
| Ownership | Machine data, operator input and derived metrics are customer data | Removes doubt about what you can license or move |
| Vendor license | Limited to delivering the service and aggregated, de-identified improvement | Keeps your history out of third-party products without notice |
| Model training | Written opt-in required for models offered to other customers | Makes AI training a decision you take, not a default |
| Benchmarks | Your data enters pooled benchmarks only aggregated and de-identified, with an opt-out | Stops your history appearing in comparisons sold to other plants |
| Export | Raw event history in a documented, machine-readable format on request | Lets you build a licensing package or switch vendors |
| Termination | A defined export window, then certified deletion | Protects your history and closes the vendor's copy |
| Subprocessors | Notice of hosting and analytics providers | Shows where your production data actually sits |
| Changes to terms | Data use permissions cannot widen through an online policy update without notice and a right to object | Keeps the deal you negotiated from shifting after signature |
How SourceX approaches machine monitoring data#
SourceX treats machine monitoring data as one input to a plant's licensable record set, not a standalone asset. In the Rights step of the SourceX five-step transaction (Supply, Rights, Preparation, Approval, Delivery), the platform agreement, any controls or machine builder software terms and relevant customer contracts are reviewed before anything is prepared.
The fit check runs on metadata only: which platform, which machines, how much history and which export routes exist. If a package proceeds, the SourceX Evidence Packet records provenance, licensing rights, permitted use, the privacy record and release authorization, and large signal histories stay in the plant's own storage or ship on encrypted drives.
Frequently asked questions
Does buying the machine give me rights to the data the control generates?
Not automatically. You own the machine and control what happens in your plant, but software on the control, connectivity options and monitoring platforms come with their own license terms. Those terms decide what you may export, share or license, so read them alongside the equipment purchase agreement.
Can a monitoring vendor use my data to train its own AI models?
It depends on the license you granted. Many agreements let the vendor use aggregated or de-identified data to improve the service; whether that includes training models sold to other customers turns on the exact wording. If the terms are unclear, ask for written confirmation and negotiate an opt-in at renewal.
Are downtime reason codes personal data?
Reason codes on their own describe the machine, not a person. They can become personal when tied to an operator name, badge ID or shift record that identifies an individual. Before licensing, replace operator identifiers with roles or pseudonymous IDs. If you have California staff, note that the CCPA's temporary exemption for employee personal information expired on January 1, 2023, so ask counsel whether it or other state privacy laws may apply.
What happens to our history if we switch monitoring platforms?
Request a full raw export before the old contract ends, and confirm the format, the date range and what happens to the vendor's copy. Summary dashboards rarely carry the event-level history that a new platform or a licensing buyer needs, and access usually closes when the subscription lapses.
Do we need monitoring data at all to license production records?
No. ERP job records, maintenance work orders, NCRs and scheduling decisions can form a strong package on their own. Machine signals add a timeline of what happened on the floor, so they raise value when you hold clear rights, but they are not a prerequisite.
Sources
- The California legislature did not extend the CCPA employee and business-to-business personal information exemptions, so they expired on January 1, 2023. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.