The Employee Voice Layer: Using AI to Understand What Employees Are Really Saying

Hello, everybody! Welcome on this beautiful spring Tuesday. Thanks, everybody, for joining in for a session on how to use AI to understand what employees are really saying. So, you know, we want to surface what matters, identify issues early, and understand the employee experience better than we can with surveys alone. And we’re going to talk about what that means from an AI perspective and in general. We’re going to talk about some prompts and cool stuff like that.

But while people are joining in, you know, we are here live all together, all 42 and counting. So love to see in the chat where people are calling in from, what industry are you in.

And we’re going to do some polls along the way, and if you have any questions or comments, please chime into the chat. We have people on call ready to bubble up your questions to me, or answer them on the spot in the chat, or just to share comments and your experience.

In fact, if you want to share, I think we’re doing a survey on surveys, so we’ll get to that in a second, but love to see where people are calling in from. I’m here, again, as you may know, in Seattle, as I normally am. I actually have my home office. I took a picture of my home office right here, and I actually use it as a virtual background. Like last February, I was in Spain, and I did a couple webinars from Spain, and I just used my virtual background, so it’s kind of fun to always be in your home office, so to speak.

But awesome seeing some people coming in from Boston, Toronto, Charlotte, I love it. Minneapolis, what’s up? Let’s go. Love all the people connecting and coming in from all over the country. And if you want to share what industry you’re in, I love to always talk about specific industries in the conversation. Denver, beautiful. We had our team gathering there last time in Denver, a lot of fun.

California, of course, Bay Area, that’s where we got our start at our company Recognize. So yeah, it’s a good time to just jump into it.

So, you know, who am I? I’m Alex Grande, I’m the CEO and co-founder of Recognize. I was in San Francisco, and I said to my colleague at the time, I want companies to have a culture of appreciation. I want to motivate the workplace beyond just the simple sticks and carrots. I want to be able to provide a badging interface to categorize company values, to incentivize positive behaviors, and to create a record of that.

And what’s been interesting about using AI now is that when you create those recognitions, this structured data, you can then utilize it so much more using AI. So I’m going to talk about that a little bit today, but that’s what I’m passionate about: company culture and organizational psychology. I have a background in psychology, and I’d love to connect. So if you’re on LinkedIn, find me on LinkedIn. You can scan the QR code, we’ll give you a chance at the end, and my email is on the screen.

So before we dive into the deep end on HR data and things like that, I want to demystify some of the AI buzzwords that have been floating around. You know, what does it mean, LLM? What is Copilot? Some of this may be review. We’re going to do two slides on AI terminology, and then we’ll get into practical things we can do.

So an LLM, you can think of it as the brain, right? And something like Copilot is the assistant sitting on top of that. It’s an assistant that allows you to connect to that. When you chat, you’re able to connect with this large language model, as it’s called, LLM.

And we don’t really need to know much about that, and honestly, I don’t think the people who are creating it know much about what it is or how it works. But that Copilot, that ChatGPT, is your interface, the UI to connect to that LLM.

Agents, on the other hand, are given a goal and can run on their own accord to complete a multi-step task. So agents are something that might live on your computer, have access to a folder, sort or summarize documents, or maybe live inside a browser tab.

I have a friend who actually used an agent, I think the Complex one, to actually negotiate with a large TV and telecom provider for a lower rate, and the agent did it on his behalf. It first talked to the company’s agent, and then it actually started talking to a real person over chat and negotiated a lower rate. So this is happening now.

Now you also want to know about MCP. That’s Model Context Protocol, unless you want to sound smart in a meeting. Think of MCP as basically a USB cable for AI. Instead of engineers writing custom code to connect AI to Workday, or ServiceNow, or my app Recognize, MCP provides one standard connection so your AI can securely talk to other data sources.

So if you’ve used tools like Cloud Connects and connected to sales data, you can ask it to summarize the last 10 sales calls, and it can do that. This is not too hard anymore, it’s click and go.

You’ve probably also heard about OpenClaw. It’s an open-source AI agent that has absolutely blown up, possibly one of the fastest growing GitHub projects in history. People use it from WhatsApp or Slack to run tasks, summarize emails, even control phones by mirroring them on a computer.

But OpenClaw isn’t the only player. If you want browser automation, you have tools like Thunderbit or Firecrawl to crawl webpages. There’s also NanoClaw, like OpenClaw but smaller and more secure. A lot of this ties into security as well.

There are many tools out there. Lindy AI is another interesting one—just Google it. Very interesting capabilities. Companies have a lot of concerns around security with these tools, and we’ll talk more about that.

You can even get a $500 MacBook now, install something like Lindy, and have it summarize your text messages or tell you when you’re having dinner with your parents. We’re inundated with messages and email, and AI can help us manage it.

You may not be using Cursor or Microsoft Code tools, but I encourage you to try them. It’s easier than ever to write code with AI. You don’t have to use them, but it’s good to know what’s out there.

We have a poll now, and I may have made the polls in the wrong order, but let’s go ahead anyway. Are you utilizing your recognition data for your employee recognition program, or do you not have a program at all?

We’ll come back to recognition later. It ties into survey fatigue.

The results show very few people are using structured recognition data. Some are, and we’ll talk later about how to start doing that without buying more software.

Let’s talk about survey fatigue. People are burned out on surveys. When we send annual surveys and only get 30–40% response rates, we have to ask whose voices we’re actually hearing.

Research shows 47% of employees feel pressured to hold back honest opinions. Only 27% believe HR will take meaningful action on feedback. When feedback goes into a black hole, trust drops and participation falls.

We need to supplement surveys with passive listening.

Passive data is the digital exhaust we leave behind: Teams messages, Slack messages, calendar overlaps, recognition data. It requires no effort from employees.

Surprisingly, 72% of employees are comfortable with employers using passive system data, as long as it’s transparent and used at a cohort level, not to punish individuals.

Surveys still tell us why. Passive data tells us what is happening and gives leading indicators.

If your company uses Microsoft 365, you’re sitting on a goldmine. Viva Insights aggregates collaboration data like chats, emails, and calendars while protecting privacy.

Then you can layer Copilot on top. Instead of waiting a month for analysis, HR or comms teams can ask Copilot questions in plain English and get real-time insights.

You can even spot burnout before it happens. Microsoft found 87% of employees think they are productive, but only 12% of leaders agree. That’s a trust gap.

We can detect signals like late-night messaging, withdrawal from channels, or changes in communication patterns.

AI can also summarize thousands of employee comments instantly in tools like Viva Glint or Recognize. It extracts themes, sentiment, and outliers based on prompts.

You can even refine prompts over time and store them in a knowledge base like SharePoint, Confluence, or Notion so teams standardize analysis.

This removes administrative burden and frees time for actual problem solving.

Just try it. I once asked my CTO if I could use AI for a task, and he said, “Just try it.” It worked. It was incredible.

If you’re here to understand employee data, there has never been a better time.

Now we look beyond the org chart. Organizational network analysis, or ONA, maps how work actually gets done. It reveals informal leaders and influence networks.

These are the people everyone goes to for advice. They may not be managers, but they are critical nodes in the organization. ONA helps identify and support them, and prevent burnout by redistributing responsibility.

Now we move to another poll about AI privacy concerns.

Results show people are generally somewhat to extremely worried about AI privacy. That’s expected.

It’s important to use tools that do not train on your data, like enterprise LLM setups such as Copilot or Gemini within secure environments.

There’s also a cautionary tale of companies gamifying AI usage and spending huge amounts on tokens due to misuse and incentives. So we must be careful what we reward.

Recognition systems, like the one in Recognize, create structured data from peer recognition tied to company values. This can be analyzed with AI.

For example, I once pulled four years of recognition data for an employee’s anniversary and surfaced meaningful contributions people had forgotten about.

Recognition becomes a strategic asset when combined with passive listening and collaboration data.

You can connect this to Copilot via Microsoft Graph and ask questions like who is collaborating across teams or who is under-recognized.

English becomes the programming language. Everyone can now query organizational data.

We also often ignore IT support ticket data, which is actually a frustration log. It shows onboarding friction and process breakdowns.

AI can analyze these trends and identify where internal systems are slowing down.

We can also use Copilot to draft empathetic communications and action plans based on survey insights, but always with human review.

We should also be mindful of environmental impact. Not every AI query needs a large model. Use AI for meaningful work, not trivial calculations.

Tools like Gamma can even turn outlines into presentations. I used it to build this presentation, and it helped someone turn their dissertation into a strong presentation.

The goal is not hyperproductivity, but higher-quality work.

Microsoft also offers Copilot impact survey templates in Viva tools so you can measure ROI of AI usage itself.

We must ensure AI is actually reducing workload and improving quality, not just adding noise.

Privacy is critical. Enterprise tools like Azure AI Foundry include PII detection and masking so sensitive data is protected while still enabling insights.

We also need user-delegated access so employees only see data appropriate to them.

If building agents, we must test for hallucinations and accuracy. Copilot Studio includes evaluation tools for this.

The future HR team becomes an intelligence function, not just administrative. Less compliance, more insight. More continuous listening across channels.

We close with an invitation to upcoming webinars and live demos of Recognize, including recognition-focused sessions and training.

Thank you everyone for joining, and I hope to see you next time.