Short answer
A franchise network can supply detailed, standardized service procedures across many locations, but the rights are split. The franchisor typically controls brand standards and training content; each franchisee controls its premises, staff and local recordings. Licensing video from a network usually needs the franchisor's agreement plus a clear path to each franchisee's permission for anything recorded on a franchisee's site.
The buyer's problem
Multi-site operators look attractive because they standardize work. A restaurant brand, a hotel chain or a cleaning franchise writes one way to do a task and trains hundreds of sites to follow it. For a buyer who wants demonstrations of the same task across locations, that standardization gives repeatable procedures and natural variation in layout, staff and equipment.
The difficulty is that "the company" rarely owns everything you want. A franchisor may hold the copyright in training films and manuals, while the footage that cameras at franchised locations record belongs to, or is controlled by, the independent franchisee. Corporate-owned locations are different again. Before comparing prices, map who can grant what. This page covers that rights pattern; for the general case of training archives, see evaluating corporate training video archives.
Three kinds of material a network may hold
| Material | Typically controlled by | What it can support | Limits |
|---|---|---|---|
| Brand training videos and manuals | Franchisor, often produced by a vendor | Task structure, step order, language | Staged or edited; may include third-party music and talent |
| Standards documents and checklists | Franchisor | Success criteria, constraints | Documents, not motion; see what SOPs can and cannot teach |
| In-store or on-site camera footage | Franchisee or corporate store operator | Real operations at a site | Cameras placed for security, not learning; staff and customers visible |
| Newly commissioned recordings | Whoever signs the collection agreement | Designed viewpoints and tasks | Needs site access and consent at each location |
The first two are the franchisor's to license, subject to any third-party content in them. The third is where the split bites. The last can be designed with consent from the start, which is one reason a network program may end up as commissioned work; compare the routes in existing assets, new collection or specialist providers.
Why multi-site data is worth the complexity
A network gives you a controlled test of generalization. When the same procedure is performed in 20 locations by different staff in different layouts, you can hold out whole sites and measure whether a model transfers, rather than hoping a single-site result generalizes. The same structure helps planning models, where the written standard supplies the step vocabulary; see workplace video for task planning.
Standardization is also a risk. Staff who follow the brand playbook closely produce little variation in method, and a model trained on them may not cope with a site that departs from it. Ask whether recorded variants exist and whether deviations are noted. Multi-site collection standardization covers how to keep camera and protocol settings aligned when the tasks are recorded on purpose.
Rights questions to settle
These are diligence questions, not conclusions. This is not legal advice.
- Who owns the copyright in training content? The U.S. Copyright Office's report on generative AI training discusses where copying can implicate rights during development and the licensing options available (U.S. Copyright Office, 2025). Whether vendor-produced training films were assigned to the franchisor depends on the production contract.
- Does the franchise agreement address data? Some agreements give the franchisor rights over operational data or require system use; others say nothing about recordings. A franchisor cannot necessarily license what its agreement does not reach.
- Who is the controller of on-site camera footage? European guidance on video devices treats the entity that decides purposes and means of processing as responsible for notice and lawful basis (EDPB, 2020). Suppliers here are US businesses, so state-law and contract questions apply; the EDPB text is a useful checklist of questions about purpose, notice and retention, not a statement of US requirements.
- What were workers and customers told? Security cameras installed for loss prevention carry notices that may not extend to AI training. See worker consent and identifiable people.
- Is brand-confidential material visible? Menus in development, equipment layouts, back-office screens and supplier labels may need cropping or redaction. Redaction affects what the data can teach; see redaction in demonstration video. It does not settle permission.
- Can the franchisor bind franchisees? If not, each site needs a signed agreement, which affects both cost and how many sites you can realistically obtain.
Illustrative rights map
# Illustrative only - synthetic entities
network: brand_X (240 locations; 60 corporate-owned, 180 franchised)
assets:
- id: A1
type: brand training films
licensor: franchisor
open_questions: [vendor assignment, music licenses, talent releases]
- id: A2
type: standards checklists (PDF)
licensor: franchisor
open_questions: [separate license from video?]
- id: A3
type: store security camera archive
licensor: store operator (corporate sites) | franchisee (franchised sites)
open_questions: [notice wording, retention, camera angles, third-party customers]
- id: A4
type: commissioned task recordings
licensor: per collection agreement
needed: franchisor consent + franchisee site agreement + participant consent
Designing a pilot across a network
A pilot across a network should answer two questions: can the network's sites be recorded consistently, and does the data differ enough between sites to matter. Start with a small, deliberately mixed sample, for example two or three corporate-owned sites and two or three franchised ones, covering at least two layouts of the same task. Use one capture protocol, one task script and one label scheme everywhere, and record the site, operator and equipment identifiers with every clip so that you can later split by site.
Expect the franchised sites to take longer to bring on board, because each owner decides for themselves and may want something in return, such as a say over what is shown. Plan the schedule around signed site agreements rather than camera availability. Keep the corporate-owned sites as the baseline, since their permissions are simpler, and treat franchised sites as the test of whether the network's cooperation is real.
For evaluation, decide in advance which sites are held out. The value of a network is that you can keep entire sites out of training, so write the split into the delivery terms; see demonstration dataset splits. Finally, agree how brand-confidential frames are handled before recording, not afterward. A franchisor that reviews footage only at the end tends to ask for large cuts, which damages the very continuity you were buying.
Questions to ask a franchisor or multi-site operator
- Which training and standards materials did you produce, and which were made by vendors?
- Do franchise agreements allow you to license operational recordings, or only your own content?
- How many sites are corporate-owned versus franchised, and could the recorded sample include both?
- What do notices to staff and customers say about recording and later uses?
- Which tasks follow the standard, and where do sites commonly vary?
- What brand-confidential items could appear on camera?
- Can a single agreement cover participating franchisees, or is each site contracted separately?
- Who signs for the network, and who can confirm they have the authority?
What network data cannot support
A network's material, however well licensed, has the same technical limits as any workplace footage. Training films and standards documents give task structure, step order and success criteria, but they contain no motion, force or perception data. Security camera archives show real operations, but from ceiling angles chosen for loss prevention, with hands small, occluded or out of frame, and without the robot joint states, actions or forces a humanoid controller needs. None of it, alone, can train a humanoid's physical control; it can support task understanding, planning, progress estimation and evaluation, and turning it into control data needs retargeting, inferred actions or co-training with robot data.
Network-specific limits add to these:
- Uniform method. Close adherence to one playbook narrows the variation in how a task is done, even when layouts differ.
- Uneven coverage. Sites that agree to take part may not be representative; franchisees who decline can leave whole regions or formats missing.
- Happy-path bias. Brand films show the standard done right; mistakes and recoveries are rarely kept.
- Redaction cost. Cropping brand-confidential items and customers removes frames and context, which reduces training value.
- Changing standards. Procedures are revised over time, so footage from different years may show different versions of the same task; ask for the standard's version date per clip.
Discuss your requirements
If you need standardized service tasks recorded across several locations, describe the task, site mix and permitted uses to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and agrees terms only where a supplier does. A request does not guarantee a match. See the workplace demonstration hub and the humanoid training data guide for the wider map, or the licensing terms for robotics training data for clause topics.
Explore sourcing and licensing options with SourceX
Sources
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — Copyright Office report on generative AI training and licensing options.
- Guidelines 3/2019 on processing of personal data through video devices (Version 2.0) — EDPB guidelines on personal data processing through video devices.