Just finished reviewing a contract for a new marketing automation platform, and I found something that gave me pause. Buried in the data processing addendum was a clause granting them the right to use "anonymized and aggregated" customer data to train and improve their machine learning models. On the surface, it sounds harmless, but the definitions were incredibly broad.
Here’s what made me push back:
* "Anonymized" was defined per their "reasonable efforts" and industry standards, which can be a moving target.
* "Improve their services" could mean using insights from *our* campaign patterns to build features they then sell to our competitors.
* There was no explicit list of what model training this covered or an opt-out mechanism for this specific use.
I've seen similar clauses before, but this one felt particularly sweeping. It wasn't in the main service agreement but tucked away in a DPA, which is easy to miss.
Has anyone else negotiated this lately? What did you ask for? I'm thinking of requesting:
* A clear, detailed description of what "anonymization" entails.
* A restriction that our data cannot be used to train general-purpose AI/models offered to other clients.
* An explicit opt-out right without it affecting our core service terms.
Would love to compare notes—this seems to be popping up everywhere now. Are we being overly cautious, or is this a legitimate data ownership concern?
~Jen
Always testing the next best thing.
Oh, absolutely. That "improve their services" line is the real kicker, isn't it? It's basically a blank check. I had a similar tussle last quarter, and I found the pushback had to be super specific.
You're right to target the opt-out and the restriction on general-purpose models. What also worked for us was adding a use-case annex. We made them list the *specific* model purposes, like "sentiment analysis on support ticket text" or "email send-time optimization," and got it attached as an exhibit. It stops the creep.
Also, check if the clause references "derived insights" or "aggregated learnings." That's often the loophole where they package up the intelligence from your data without the raw data itself. Ask for that to be excluded or owned by you. Good catch spotting it in the DPA, most people glaze over those.
If it's not measurable, it's not marketing.