Latest / NOXTalks / Data Rights, AI Training, & The Next Contract War

Transcript

Data Rights, AI Training, & The Next Contract War · NOXTalks

• 0:07
  SPEAKER_00: Welcome to Knox Talks, the AI-generated podcast exploring the groundbreaking ways technology shapes our built environment.
• 0:14
  SPEAKER_00: Today we're diving into a question that's getting weirdly competitive in smart buildings.
• 0:19
  SPEAKER_00: Who owns the building's brain?
• 0:21
  SPEAKER_00: Not the concrete, not the chillers.
• 0:24
  SPEAKER_00: The brain is the living history of how the place actually behaves.
• 0:28
  SPEAKER_00: Trend logs, alarms, overrides, schedules, set points, and all the labels that explain what each point really is.
• 0:36
  SPEAKER_00: Add in maintenance notes and work orders, and suddenly you've got something far more valuable than raw sensor readings.
• 0:43
  SPEAKER_00: You've got operational context, and that's exactly what modern analytics tools want, because that's what makes machine learning work.
• 0:51
  SPEAKER_00: Here's the scenario I keep hearing in different forms.
• 0:54
  SPEAKER_00: An owner or facilities team brings in an analytics platform to cut energy and improve comfort.
• 1:00
  SPEAKER_00: The vendor says, no problem, we just need access to your building management system data.
• 1:05
  SPEAKER_00: Everyone's happy for a year, maybe two.
• 1:07
  SPEAKER_00: Then the portfolio wants to add a second tool or switch providers, and that's when the fund starts.
• 1:12
  SPEAKER_00: The export is incomplete, the download is just a pile of time series numbers with no tags, no asset IDs, no equipment hierarchy, no, this point belongs to this air handler.
• 1:24
  SPEAKER_00: The automated data feed has limits nobody negotiated, and buried in the contract is a clause that says the vendor can use de-identified combined data to improve services.
• 1:35
  SPEAKER_00: Which sounds harmless until you realize it can include training models that benefit the vendor's product roadmap and not necessarily your building.
• 1:44
  SPEAKER_00: At that point, the fight isn't about dashboards, it's about leverage.
• 1:49
  SPEAKER_00: If you can't leave with your full data set in the context you paid to create, you're not owning intelligence.
• 1:55
  SPEAKER_00: You're renting it.
• 1:56
  SPEAKER_00: And this is where we own the data becomes a trap phrase.
• 2:01
  SPEAKER_00: Ownership sounds strong, but it doesn't help if you don't have clear, enforceable permissions and practical access.
• 2:08
  SPEAKER_00: When you strip it down, data rights are really six questions.
• 2:12
  SPEAKER_00: First, who owns the raw information?
• 2:15
  SPEAKER_00: The trends, alarms, and metadata?
• 2:17
  SPEAKER_00: Second, can you access it easily?
• 2:20
  SPEAKER_00: Not upon request, not we'll get back to you, but can you pull it automatically any day in real time?
• 2:28
  SPEAKER_00: Third, what is the vendor allowed to do with it?
• 2:30
  SPEAKER_00: Only deliver the service you're paying for?
• 2:33
  SPEAKER_00: Or also reuse it for product development and model training?
• 2:36
  SPEAKER_00: Fourth, who owns the derived work?
• 2:39
  SPEAKER_00: Things like fault rules, tagging maps, normalization layers, and performance metrics that were built using your building's history.
• 2:46
  SPEAKER_00: Fifth, how long is it kept and what does deletion actually mean?
• 2:50
  SPEAKER_00: Is it truly removed or does it live forever in backups?
• 2:54
  SPEAKER_00: And sixth, can you move it in a usable format with the context intact, without a six-month re-tagging project?
• 3:03
  SPEAKER_00: One fun fact from the trenches teams rarely fail because the algorithms are bad.
• 3:08
  SPEAKER_00: They struggle because the building data is messy, inconsistent naming, missing tags, no agreed asset registry.
• 3:16
  SPEAKER_00: The glamorous part is the insights.
• 3:19
  SPEAKER_00: The grind is making the data understandable.
• 3:21
  SPEAKER_00: So where do contract wars usually happen?
• 3:25
  SPEAKER_00: 1.
• 3:26
  SPEAKER_00: The contract says you own the data, but access is throttled, expensive, or missing key context.
• 3:33
  SPEAKER_00: 2.
• 3:33
  SPEAKER_00: The vendor claims the intelligence is theirs, even if you paid for tuning and configuration, and you leave with a raw export that's basically unusable.
• 3:44
  SPEAKER_00: 3.
• 3:45
  SPEAKER_00: Training rights are vague.
• 3:46
  SPEAKER_00: Improved services becomes a blank check unless you draw boundaries.
• 3:51
  SPEAKER_00: 4.
• 3:52
  SPEAKER_00: Retention after termination is fuzzy.
• 3:54
  SPEAKER_00: If you cancel, what happens to five years of history and enrichment?
• 3:59
  SPEAKER_00: 5.
• 4:00
  SPEAKER_00: Offboarding is slow.
• 4:02
  SPEAKER_00: Support ends on termination, but exports arrive weeks or months later, which can force you to keep paying just to transition.
• 4:10
  SPEAKER_00: Now, layer in privacy.
• 4:12
  SPEAKER_00: Building data isn't always just building data.
• 4:14
  SPEAKER_00: As soon as you can tie it to identifiable people, access logs, named occupancy records, certain types of video analytics, you've stepped into privacy law territory.
• 4:25
  SPEAKER_00: That changes the stakes.
• 4:26
  SPEAKER_00: Cross-border cloud storage, subcontractors, breach notification, incident cooperation, all of that belongs in the deal, not in hopeful assumptions.
• 4:37
  SPEAKER_00: So what should you ask vendors in plain English?
• 4:40
  SPEAKER_00: If we terminate, can we export everything, including tags, asset IDs, equipment hierarchies, and documentation?
• 4:48
  SPEAKER_00: And how fast?
• 4:50
  SPEAKER_00: Can we pull data automatically or do we need to request it manually?
• 4:54
  SPEAKER_00: Do you train models on our data?
• 4:56
  SPEAKER_00: If yes, exactly what data, for how long, and can we opt out?
• 5:01
  SPEAKER_00: What do we keep if we leave?
• 5:03
  SPEAKER_00: The fault rules, the tagging schema, the performance metrics configuration?
• 5:08
  SPEAKER_00: Where is the data stored and who else can access it?
• 5:11
  SPEAKER_00: And if we want two analytics tools at once, can both read the same feeds without restrictions?
• 5:17
  SPEAKER_00: Then do a simple test, a data fire drill.
• 5:20
  SPEAKER_00: Pick one critical system and request a real export, say the last 60 days, plus all metadata and structure.
• 5:28
  SPEAKER_00: Then try to use it somewhere else.
• 5:30
  SPEAKER_00: Whatever breaks is your lock-in.
• 5:33
  SPEAKER_00: Finally, what does good look like?
• 5:35
  SPEAKER_00: A one-page data charter you attach to every main ag.
• 5:39
  SPEAKER_00: Finally, what does good look like?
• 5:41
  SPEAKER_00: A one-page data charter you attach to every main agreement and statement of work.
• 5:46
  SPEAKER_00: Short, enforceable, boring, and that's the point.
• 5:51
  SPEAKER_00: It should say the vendor can use the data only to deliver the service.
• 5:55
  SPEAKER_00: No model training unless you explicitly opt in.
• 5:59
  SPEAKER_00: You get automated access, exports include full context, deletion happens on a defined timeline with written confirmation, and anything you pay to customize can be exported and reused.
• 6:11
  SPEAKER_00: Because the most expensive lock-in isn't hardware, it's the invisible brain you've built over time.