AI Notetaker Adoption in Financial Services: What the Numbers Say and What's Next

During a webinar on AI meeting notes and regulatory compliance, Global Relay polled its own audience of compliance and financial-services attendees on a simple question: what is your firm's current policy on AI meeting tools? Open use, restricted use, or an outright ban?

The results, read live by Ben Hall, Head of Partner Marketing at Global Relay and the session's moderator:

"A whopping 68% are having restricted use. Open use is 12%, and a fifth of respondents are still dealing with outright bans at the moment."

Two-thirds restricted, one-fifth still banning the category outright, and only 12% with genuinely open use. Hall's framing of that result: "when it comes to our meeting tools, it's still something obviously that is met with, with that caution."

The rest of the conversation, between Aydin Mirzaee, CEO and co-founder of Fellow, and Ryan Sheridan, Director of Regulatory Intelligence at Global Relay, was about what that caution actually looks like in practice and how a firm moves through it.

What "restricted use" is replacing

Mirzaee's response to the poll result focused on how restricted-use policies have traditionally been enforced, and why that method is now changing:

"Typically when you have restricted use, the way that that works is someone writes a policy. You know, it's sometimes a long policy depending on how detailed it is. And the idea is that everyone's going to read that and then fully internalize it and then never make any errors, right?"

The problem with that model is obvious once stated: a written policy depends entirely on individual compliance, with no enforcement mechanism behind it. Mirzaee's point was that this is no longer the only option:

"One of the nice things is what technology now allows you to do is actually create rules around these things. So you can actually say rules if it's a meeting type is like this, or if it's that part of my organization. And so, and in these cases, there's no transcript. In those cases, there is a transcript. And so I think one of the things that I would encourage everyone to really think about is from a restriction of use, how much broader can something get so you can get more business application out of it if you are able to, using technology, actually enforce the policies that you want."

His summary of the shift: "that's the other thing that has really changed over the last year is that now you can really enforce a lot of the things that previously you would just put on paper and hope that people can implement." That is the practical difference between a policy that exists on paper and one that is enforced automatically, by meeting type, by department, or by participant, without relying on every employee remembering the rule.

Adoption is real and accelerating, even where caution is highest

Asked early in the call what stance firms are actually taking, Mirzaee described a shift from blanket prohibition to active adoption, even inside financial services specifically:

"I think the good news at the end of all of this is that while I think even late last year, most organizations were just straight out banning AI note-taking, it's changed significantly. More and more are adopting it, and if they're not adopting it, they're actively considering it. They're considering how they should bring it into place because the business really needs it."

Sheridan's read on why financial services specifically has moved more cautiously than other sectors, while still moving:

"For financial services, they have traditionally led the way when it comes to regulation, given its relationship to the capital markets and the investing public. However, that said, when companies look across the risk within their business units, we are seeing it in other verticals like life sciences, corporates, and governments... The bottom line is people in compliance, they want to meet the needs of the business, but they also have to meet the regulations. And in the same right, they're there to protect the franchise."

Later in the call, Sheridan put a sharper point on the tension compliance teams are navigating: "compliance cannot just be a no function. They have to be adaptive and figure out a way to make this work within each one of their franchises." That is the operating reality behind the 68% restricted-use figure: not resistance to the technology, but the absence, until recently, of a way to adopt it without losing control of it.

The practitioner route out of an outright ban

For the one-fifth of firms still in the ban category, Sheridan laid out a specific, sequenced process when asked directly what the first step looks like:

"First and foremost, you have to get a policy stood up and then have the discussions with your front office personnel, your legal department, your HR department, and come up with a framework or fold it into an existing framework to monitor these types of communications.

I think then second of all is obviously reach out and do a scan across the industry to what types of technology service providers offer this type of service. Do your vendor due diligence like you would for any other types of platform that's business critical within your organization. And then begin the onboarding process through demos, POCs, and in your typical sales cycle."

He was explicit about why the process needs to be documented at every stage, not just completed: "It's a methodical step-by-step approach that should be documented along the way to just ensure that, you know, if and when a regulator asks you how you made certain decisions, you're pretty well buttoned up and can provide that information."

Mirzaee's addition was about sequencing the rollout itself once a vendor is selected, rather than attempting an all-at-once switch:

"I would let the business needs kind of dictate that. Where would this be the most impactful? So if we wanted to do a proof of concept to start in one part of the organization, again, judge it by impact and then start from there and then keep layering it on and adding more and more parts of the business and use cases."

He also pointed to peer precedent as a practical shortcut for firms starting from zero, using private equity as the example: "not all private equity firms are obviously the same, but a lot of the meeting types are the same. And so this is something that you can especially work with your peer firms. And if they've already stood up policies like this, it becomes easier to start from a place because there is more similarity."

Who needs to be in the room

Asked to close out the session with what groundwork firms should be laying today, both speakers converged on the same answer: this is not a compliance-department decision made in isolation. Sheridan's list:

"The reality is it's all stakeholders. When you think about it, it's your HR partners, it's your inside counsel. You may even want to get outside counsel just to, you know, use them as a sounding board. I think it's going to be your front office personnel, your compliance personnel.

You're going to want to roll this all up into your operational risk assessment. So again, I don't, although it's a newer technology, folks don't necessarily have to reinvent the wheel. You look at this as if the same way you transact business through other areas of your organization, you leverage your key internal stakeholders and you build a robust policy around it. And just make sure that everybody has a seat at the table and a voice."

Mirzaee's closing point reinforced the same sequencing logic he raised earlier, treating this as a staged rollout rather than a single decision: "If you're looking for a step-by-step approach, or, you know, a tiered approach, I would let the business needs kind of dictate that... judge it by impact and then start from there and then keep layering it on."

What the poll shows

Read plainly, the 68/20/12 split is not a story about resistance to AI meeting tools. Every part of this conversation, from Mirzaee's read on adoption accelerating, to Sheridan's description of compliance as necessarily "adaptive," to both speakers' shared roadmap for moving off a ban, points to caution driven by the absence of enforceable controls, not caution about the technology itself.

Firms that can put rules-based enforcement behind a written policy, rather than relying on a document nobody reliably reads, have the clearest path out of the 68% and the 20% and into the 12% doing this well, deliberately, and defensibly.

How Fellow and Global Relay close the gap

Fellow lets firms set a configurable delay, for example a 7-day window, between when an AI meeting summary is generated and when it is archived to Global Relay, so a human reviewer can check the record before it becomes part of the firm's permanent surveillance file. As Mirzaee described it: "a 7-day delay or something like this, so that you have the opportunity to have a human reviewer in cases where it makes sense." That delay is a direct answer to the tone problem raised earlier in this conversation: a raw transcript can flatten sarcasm or context into something that reads as incriminating, and a review window is what catches that before it is locked into a system of record. Once reviewed, the record moves into Global Relay's archive, where it becomes part of the firm's existing surveillance and sentiment-analysis program rather than sitting outside it.