The News: Recent updates to Zoom’s terms of service combined with the June release of Zoom IQ, a smart companion that is enabled through generative AI, have amplified attention on what customer data is being used to train Zoom’s AI features and if customers have opt-out options presented to them.
Several news outlets, including Gizmodo and CNBC, recently began to question subtle changes in the terms of service made in March 2023, prompting Zoom to update the terms once again to clarify what data is being used. Chief Product Officer Smita Hashim also created a blog post to clarify opt-out options presented to customers.
See Zoom’s current terms of service and the blog post by Smita Hashim for current information from Zoom.
Zoom’s Changes Cause Confusion on When Data Is Used To Train AI
Analyst Take: In March 2023, Zoom updated its terms of service “to be more transparent about how we use and who owns the various forms of content across our platform,” according to Hashim’s August 7 blog post.
The dubious updates to the terms of service deal primarily with customer content and occur in section 10 of Zoom Terms of Service. Zoom defines customer content as the combination of “customer input” (data, content, files, documents, or other materials that the customer has provided, uploaded, or originated in the service) and certain derivatives, transcripts, analytics, outputs, visual displays, or data sets resulting from the customer input.
Hashim clarifies in her post that Zoom has permission to use this customer content to provide value-added services based on the content, but the customers continue to own and control their content.
The Devil Is in the Details
However, the news outlets mentioned above are questioning the addition of the language that outlines Zoom’s rights to use this content for “the purpose of product and service development, marketing, analytics, quality assurance, machine learning, artificial intelligence…” This is where the confusion lies.
In explaining this wording in her blog post, Hashim references an example of machine learning (ML) where Zoom may scan webinar invites to ensure Zoom is not part of a mass spam effort. The point being made is that Zoom employs ML and AI technology in its ongoing, everyday enablement of the platform and related services. In other words, this seems to refer to the application of an AI feature (e.g., identifying all spam emails from the webinar product) by comparing customer content against an “already-trained” AI feature.
But hold on for a moment – the topic does become cloudier as there is a second mention of data that is used for ML and AI. In addition to customer content, Zoom may use Service Generated Data, which the company defines as “telemetry data, product usage data, diagnostic data, and similar content or data,” and which Zoom retains ownership of. This data is also used for the same purposes as customer content but with the added note of “training and tuning of algorithms and models” related to ML or AI.
So, what gives?
Understanding Data for Training AI and Data for the Application of AI Features
I am willing to speculate that what is taking place here is the difference between what data is being used to train certain AI features and what data is being predicted or evaluated based on the already trained AI.
It appears to me (and still requires some validation) that the customer content discussed in the terms of service has the potential to be used as a target variable that Zoom wants to understand more clearly, for something such as, “Is this spam?” It is not being used to train Zoom’s AI.
In fact, Hashim explicitly calls out in her blog post that Zoom’s AI features do not use audio, video, or chat content for training their AI models without customer consent.
But there is an exception to this situation, and it occurs when the customer does give explicit consent for Zoom to use their customer content to train Zoom’s latest generative AI features.
Zoom IQ, a Generative AI Feature With Customer Consent, and the Ability to Opt Out
In June, Zoom released two new generative AI features under the umbrella branding of Zoom IQ. Available through free trials for customers in select plans, the features are Zoom Meeting Summary and Zoom Team Chat Compose.
Meeting Summary enables the meeting host to create a summary powered by Zoom’s own large language models (LLMs), while Team Chat Compose uses generative AI to draft messages based on the context of a Team Chat, change message tone and length, and rephrase responses to customized text recommendations.
But for a meeting host or a chat user to get access to these features, it requires the Zoom Administrator to proactively grant or deny consent for customer content to be used to enhance (i.e., train) the generative AI features in Zoom IQ.
Even more important, after the administrator enables the AI features, they can still disable data sharing at any time. Hashim does a great job of explaining this in her blog post and provides sample screenshots of the dialog boxes the Zoom administrator interacts with when enabling these features.
As you can see, the administrator is given a chance both during the setup and after the feature is enabled to provide or remove consent to use customer content.
End User Experience
The key idea so far is that the customers can opt out, but it is at the admin level.
It is slightly different for end users. An end user of Zoom does not have an explicit option to turn off the Zoom IQ feature that an administrator has enabled for their entire Zoom instance. But they are provided with ample notification that the call they are about to participate in may potentially be used to improve the Zoom IQ features. Below is a sample notification I captured from my own Zoom experience.
Zoom’s Terms of Service Compared to Similar SaaS Offerings Is Business as Usual
First, let’s talk about the terms of service update. Getting clearer on the different types of data and how they are used was a smart move for Zoom. From my perspective, this update is business-as-usual, especially when compared to most SaaS offerings in the market today.
Refining products takes data, and in most instances, the software provider retains rights to the data that can help to improve the product. Microsoft, Google, and Salesforce, to name a few, all retain data ownership on their equivalent of “service-generated data” and have or request some aspect of licensed access to content that is uploaded or created to assist with product improvements. (Degrees of access will vary, some data is anonymized, etc., but the model for improvements with customer content is not exclusive to Zoom.)
For those familiar with terms of services for SaaS offerings, the hubbub around this is just a big “nothing burger.”
Access To Training Data Will Become More Important to All Industries Adopting AI
But what is interesting is that this will not likely be the last time companies are questioned about how they are training their AI and ML models.
In Zoom’s workplace collaboration market, vendors have been using lab-created data sets for years to improve video production rules, audio intelligence, and in-meeting experience.
But Zoom is also a SaaS provider and, like other SaaS peers, can leverage even more powerful data that is real-time and takes place closer to the edge in the customer’s environment.
Generative AI has proven that these larger data sets enable the technology to respond to a much broader array of situations with more customized responses. With proper consent from customers, this is the path for solution providers to provide better experiences and outcomes for customers using their technology.
Key Conclusions
I fully expect to see more conversations between solution providers and end customers negotiating access to customer data that will greatly improve the training of AI. Most likely, this approach would start with anonymized data, but the potential to provide a bespoke experience with a company’s own data will continue to be a goal.
The vision of workplace collaboration, the one that has been discussed in industry conferences and executive presentations, depends on obtaining relevant training data. And if this is the case, the race to differentiate will be based on well-trained AI and will be the most valuable intellectual property a company can create.
Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.
Other insights from The Futurum Group:
Zoom Adds Generative AI-Powered Zoom IQ Features. Will They Resonate?
Zoom Announces Expansion of Zoom IQ, a Smart Companion, at Enterprise Connect 2023
Author Information
As Practice Lead - Workplace Collaboration, Craig focuses on developing research, publications and insights that clarify how the workforce, the workplace, and the workflows enable group collaboration and communication. He provides research and analysis related to market sizing and forecasts, product and service evaluations, market trends, and end-user and buyer expectations. In addition to following the technology, Craig also studies the human elements of work - organizing his findings into the workforce, the workplace, and the workflows – and charting how these variables influence technologies and business strategies.
Prior to joining Wainhouse, now a part of The Futurum Group, Craig brings twenty years of experience in leadership roles related to P&L management, product development, strategic planning, and business development of security, SaaS, and unified communication offerings. Craig's experience includes positions at Poly, Dell, Microsoft, and IBM.
Craig holds a Master of Business Administration from the Texas McCombs School of Business as well as a Bachelor of Science in Business Administration from Tulane University.
