Searchable Institutional Memory Gives Comms Teams a Durable Advantage Over Those Still Just Publishing

Enterprise communications teams produce an enormous volume of material every week, from executive broadcasts and video calls to Slack threads and internal memos. Most of it is created, distributed, and forgotten. All of it sits somewhere in the organization's systems, unsearchable, untagged, and functionally lost. AI's most valuable role in comms is making everything the organization has already said findable, verifiable, and safe.

Scott Grizzle is a Unified Communications and Digital Workplace Executive who has spent more than 25 years in enterprise video, streaming, and collaboration infrastructure. He most recently led UC strategy for 300,000 employees at IBM, overseeing a $100M budget across Microsoft Teams, Slack, Webex, and AI collaboration tools. His career managing global teams across languages, time zones, and platform boundaries convinced him that the hardest problem in enterprise AI is making sure the system understands what it's hearing.

"You have to teach AI about context, what is being spoken about. That's one of the big problems today," Grizzle says. Without that context layer, AI systems processing enterprise communications can't distinguish between terminology that sounds identical but means entirely different things, and the consequences can range from wasted engineering cycles to reputational damage.

Context failures are not hypothetical

Grizzle's sharpest example comes from selling AI to the U.S. Navy. A speech-to-text system mistranscribed an admiral's off-the-record remarks so severely that the output bore no resemblance to what was actually said. The Navy's conclusion was immediate: AI would not be permitted for any communications function where a hallucinated phrase could become an official statement.

The risk scales down in severity but not in frequency for global corporations. At IBM's Weather Company, Grizzle's development team in Budapest spoke English with accents that created persistent ambiguity for speech-to-text systems. Without context training, the system can't resolve the ambiguity, and downstream outputs start from a flawed foundation. "Biases are going to pop in. You're going to have hallucinations because the system heard something but doesn't know which path to take," Grizzle explains. "When your development team keys off one mistranslated word, you build the wrong thing entirely."

Institutional memory as a strategic asset

While context failures represent the defensive case, Grizzle sees an equally compelling offensive one for turning communication history into a searchable knowledge system. His personal practice during his sales career was never to delete anything. When a prospect resurfaced years later, he could pull up the original conversation and re-engage with precision that looked like extraordinary recall. "My emails went back ten years. I'd go back and say, 'Remember five years ago?' They'd say, 'You have a great memory.' No, I don't. I have the conversation right here."

The same principle applies organizationally. Enable closed captioning on every video call to generate a text transcript, and that transcript becomes searchable metadata. Suddenly, years of recorded meetings and executive communications become a queryable knowledge base. "Let's say you remember the conversation but not who it was with," he says. "Turn on speak-to-text, generate the written document, and now you can do a keyword search across everything."

Security determines whether any of this works

Every benefit of searchable comms infrastructure carries a corresponding risk. The same systems that make knowledge retrievable also make it exposable. Grizzle is emphatic that context, permissions, and access controls must be designed into the architecture from the start, not layered on after the system is built. He points to a security firm client that set a video stream to private but failed to restrict sharing permissions. The private conversation became public because participants could redistribute it. "Your private conversation is no longer private. It's out to the world. And this was a security company that forgot to put security features on," he shares. 

In a 300,000-person organization, the question extends beyond who can see what. Not everyone needs the same AI-enabled tools. Manufacturing teams don't need premium collaboration features. Sales and legal teams do. Determining who needs what access, where the boundaries sit, and how to prevent sensitive information from leaking through convenience features is now a core comms leadership responsibility. Context-aware retrieval, searchable institutional memory, and faster decision-making all depend on getting this layer right first. "Where is the data being held? Is it open where everybody sees it, or is it closed and only internal groups can access it? That's the decision you have to make before you build the system," Grizzle says.