Where AI Belongs in Workplace Investigations

This transcript has been generated by AI and may contain slight discrepancies from the audio or video recording.

Nichol: [00:00:00] Workplace investigations can be one of HR's most demanding responsibilities, whether it's responding to a harassment complaint, investigating employee misconduct, or reviewing allegations of discrimination or retaliation. HR teams are responsible for gathering facts, documenting evidence, and making fair, well-supported decisions.

AI has the potential to reduce much of the administrative burden, but as the technology becomes more sophisticated, the line between automation and human judgment becomes less obvious. So where can AI help streamline documentation and investigations while empowering HR professionals to focus on the judgment and critical thinking only humans can provide?

Joining us today is Jackie Schafer, founder and CEO of Clearbrief, an AI [00:01:00] platform that brings legal-grade accuracy to HR and legal investigations directly inside Microsoft Word and one of the startup companies in SHRM's Workplace Tech Accelerator cohort this year. Jackie is also a former litigation attorney and has spent her career protecting organizations from legal risk.

Jackie, welcome to The AI+HI Project. 

Jackie Schafer: Thank you so much. Thanks for having me. 

Nichol: Yeah. We're looking forward to an exciting conversation right after this message. 

Ad: Hey, since you're tuning into this episode, you already know shifting priorities like AI are transforming the workplace. But is your team building the right skills to stay ahead?

At SHRM Skills Accelerator this November, you'll dive into high-impact upskilling seminars and specialty credentials in the topics that matter most to organizations right now. Think everything [00:02:00] from AI, analytics, and investigations to talent, inclusive workforce strategy, and even SHRM certification prep. 

You'll walk away with deep expertise in the most in-demand, highest-risk areas of HR practice. So join SHRM Skills Accelerator, part of All Things Work twenty twenty-six. Learn more and register your team at shrm.org/atw-skills. We'll see you there 

Nichol: Okay, Jackie, I'd like to start with a Clearbrief customer case study from the legal world.

I recently read that the law firm Cozen O'Connor began using Clearbrief to help their team review legal briefs and strengthen the accuracy and verification process behind their work. Can you share what was the challenge that they were trying to solve, and what role AI served in supporting that process?

Jackie Schafer: Absolutely. So legal has actually been sort of like the canary in the coal mine when it [00:03:00] comes to identifying the issues with using generative AI to draft really critical, important documents. Um, just like HR, you know, in legal that comes out as briefs or motions. Um, in HR, you know, we'll be talking more about, you know, investigation reports and documentation.

Um, in the legal context, these mistakes actually do get found out and discovered, um, routinely because of the litigation is an adversarial process, right? Where in your legal writing, every single sentence has to be supported by a citation, and the other side is going to be scrutinizing those citations to see, okay, does the case really say that?

Is that fact really true in evidence? And so there's now to date been over 1,700 documented sanctions cases where attorneys and their firms have been sanctioned, and their clients too, for, um, using generative AI in ways that can, you know, [00:04:00] put the firm, the firm's reputation at risk. Um, and it's, it's very tricky because generative AI can sort of create these citations that seem and look real.

Um, in legal, it's sort of like a string of numbers and text. It's kind of tricky to actually go and verify and look up the source when you have something like 500 citations or more in a document. So that's, that's the challenge, and firms every single day are filing, you know, dozens of, of pleadings every day.

And Cozen O'Connor was a great example of a firm that wanted to really look, um, for an operational solution to this problem versus the way that we're, you know, in legal normally taught to, you know, catch mistakes is manually, right? But in this new era with AI generating so much text and there's so many different hands that touch a brief or a motion before it gets filed, it just isn't tenable really to do it all manually anymore. 

And so, um, they're one of hundreds of, you know, [00:05:00] large global firms across the country that use Clearbrief for checking systematically their work product. And it's in Microsoft Word. It lets them-- automatically it displays the source, uh, both references to the law. We have integrations with, you know, the most trusted legal databases, as well as references to factual evidence.

So that was, that was how we, we helped them implement our site check report, which is, you know, a sort of deterministic AI workflow that helps them check over systematically. It flags mistakes for the lawyer, so nothing gets filed with a potential issue in it. And then it also creates a report documenting, "Yes, I looked at this.

This is okay for you to sign, law firm partner." Um, and it gives them the ability to also see side by side that cited evidence. Mm-hmm. Um, so you know, it's, it's a trust issue and a trust challenge to have so many people using generative [00:06:00] AI. I think what the HR space can really take from this is the mistakes that, you know, might be hidden sort of in our documentation might not get found out immediately like they do in, in litigation where someone is-- it's, it's everyone's job to actually scrutinize your opponent's writing.

After you-- after it's filed. In HR, those mistakes might not become evident until months or years later when someone actually goes to see what were these conclusions based on, and they realize it was hallucinated factual evidence and it, it didn't exist. 

Nichol: The moment a hallucination is discovered, I imagine the entire case is in question.

Jackie Schafer: Absolutely. And literally, the attorneys are called before the court, they're sanctioned. We're seeing sanctions for these hallucinations going up to $100,000, um, and more. And, you know, penalties like the attorneys are being, um, removed from the case. The, [00:07:00] the legal arguments are being-- Like, the actual case is dismissed.

There's a lot of, you know, more and more severe penalties that we're seeing, um, inclu- you know, attorneys not being allowed admission to another jurisdiction because they had a hallucinations case in, you know, the last year, things like that. So it's really a reckoning right now that the legal profession is, wow, we've had all these fun tools for, you know, for legal research to help us create text, but we need to also have a systematic plan and, and a set of tools that help us verify and check over the work product.

Nichol: Mm-hmm. And then for the HR leaders, what are the lessons that they can take away from, from that example when they're thinking about using AI in their own workflows? 'Cause we've heard a bit about what happens to the lawyers, but, you know, what should the HR leaders be taking into consideration, and what should they learn from the Cozen example? 

Jackie Schafer: Sure. So, uh, what I would say is, you know, [00:08:00] we've been, um, working really closely with all of these, you know, law firms, many of whom focus on employment law, employment litigation, and they're sort of dealing with the aftermath of mistakes and issues that happened, um, in HR. And, you know, the way to sort of get ahead of that is if you are using generative AI for your HR work, make sure that the tools that you are using have those safeguards built in around verification, validation, and make sure that you have a set process so that leadership, you know, is-- has some visibility into, you know, what the supposed evidence or what, you know, what the underlying, um, documentation actually says.

Mm-hmm. I would say that's the biggest thing because it's so easy now to just generate, generate. Um, and I'll give you an example. So Clearbrief's tools, we are specifically designed for, um, investigation reports, as well as we have a workflow for documenting, um, performance improvement plans, [00:09:00] PIPs. Um, we also have tools that help you create your witness interview outline.

And the first step of this process is you bring in your factual evidence into Clearbrief. It's i- right in Microsoft Word, so it's sort of splitting the screen in Word. And then we have a guided workflow. So that's one of the biggest differences, um, versus, uh, using a chatbot-type experience, where a lot can go wrong basically, or there can be a lot of inconsistencies behind the scenes if you're just sort of chatting live with, with, um, an-

large language model. Um, and what we've done is actually packaged it into a workflow so that individual HR folks don't have to be prompting experts. And by the way, this was the same approach we've built for legal because it's the same problem. Nobody in legal are like, "Look, I've been spending all day. I, I'm an expert at what I do.

I'm not an expert at prompting." [00:10:00] And so by handling the, um, normalization essentially of, of the data behind the scenes and making sure the sort of pipeline is, is set up properly, we can dramatically reduce the risk of a hallucination in that output. 

Speaker 4: At the start of the year, SHRM identified seven trends set to shape the world of work in 2026.

SHRM tracked the signals so you don't have to guess how each trend has evolved, what the data reveals, or where your organization needs to go next. Get the 2026 workplace trends. Check in and stay ahead of the curve with SHRM. Visit www.shrm.org/2026predictions. 

Nichol: SHRM For me, um, I have three Cs when people ask me when to, when not to use AI. 

And, uh, my three Cs are, um, you don't use it, um, uh, you don't [00:11:00] use it for conflict You know, to make a decision around conflict, um, or to communicate around conflict. You don't use it for, um, commitment. So that's the, you know, actually deciding who to hire, like the actual decision. And, uh, you don't use it around a crisis.

Um, so crisis, conflict, and commitment. 

Jackie Schafer: Bogs us down that, that I hear all the time from HR leaders, like, I'm spending so much time just summarizing the emails and just, you know, organizing what dates this happened when, you know, and what are-- what's the, the order of operations here and, and what we can prove with that documentation we have, either from notes, from emails, from, you know, other recordings or Zoom transcripts, things like that.

Um, that's the piece that really bogs people down, and it, it often doesn't leave a lot of time actually for the, the part of the human judgment because you have so [00:12:00] many things to get through in your day. 

Nichol: So a, a quick question. Um, when you look at HR documentation today, um, around workplace investigated- investigations, what do you see as the most common mistake that organizations are making?

Jackie Schafer: So one of the most common mistakes I see is that they haven't listed out and documented the universe of evidence/documentation that they have reviewed as part of the investigation. So what that does is it makes it hard later to go back. If, let's say the conclusion was there was no harassment, we did not feel this conduct, you know, rose to the level of harassment.

That's fine. That's perfectly fine as a conclusion. But if, let's say there turns out later there were a whole set of emails that anybody would agree constituted harassment, t- it would be very helpful to know, hey, tho- those emails were not part of the set of data that we had access to [00:13:00] when, when we did this investigation.

Nichol: Mm.  

Jackie Schafer: So if you're not documenting, look, here is the box essentially of all of the, the facts that I had at the time I conducted this investigation. And if you're not making it easy to go back later and see and access that, then you're exposing yourself personally and the organization. So that's one of the also the tools that we built into Clearbrief so that you have this neat package that you just save to the file that shows, look, this is exactly what I reviewed as part of my investigation.

'Cause there's gonna be stuff that, look, you can't-- you didn't have access to or, um, you know, maybe it didn't even exist until after the investigation was over. But that can get hard to prove later. 

Nichol: Yeah. So it sets-- What I'm hearing is that it sets up a discovery timeline, um, that provides protection, uh, for the, the HR leader and also helps ensure that the, that the case that they're building is You know, is [00:14:00] properly verified.

Um, that's good to know. And so a-according to our research from SHRM's report, we have a report called the State of AI in HR 2026, we see that 39% of HR professionals said that they've adopted AI in their HR function, and 7% intend to launch AI in the near future. Organizations may want to do it to streamline functions like investigations and compliance, but might not have the governance in place.

What do you recommend as the very first guardrails every organization should establish when they're starting this journey? 

Jackie Schafer: Yeah. I think having someone very knowledgeable on your InfoSec team who understands the sensitive, you know, nature of the data that you're using in HR. Um, but specifically, um, there's a lot of tools and, and products out there that don't have a contractual relationship with the large [00:15:00] language model provider that would exempt that data from being used in training the model.

So I'll give you an example of like, you know, OpenAI, um, Anthropic, these model providers, many times their, their default-- typically their-- currently their default terms are when you use, um, even the enterprise, you know, contracts, um, when you use those large language, language models, they can hold onto that data for 30 days at least, and use it for abuse monitoring, which means they can check whether it violates any of their terms of service and that sort of thing.

Um, Clearbrief, because we have, um, actually Microsoft is one of our customers, um, Microsoft's legal team. Um, we have a relationship with Microsoft Azure's OpenAI API. So that's how Clearbrief's generative AI tools, it goes through Microsoft, and we have that contractual relationship that says, "We will not [00:16:00] retain any data that, that you use with these OpenAI models, even for one day." 

It's zero day retention, so it just can't be used for training a model. So that's one thing that an InfoSec, you know, leader might be just sort of, you know, thinking if, "Oh, well this, um, this, you know, startup or this, this company's AI company is saying they, um, you know, have an enterprise contract with the model providers."

You need to be looking at how long is that data retained, what are the specifics around if this data can be used for training the models, because that's just-- We, we don't want HR data being used in that way. It's creates a lot of risk. 

Nichol: Okay. Well, uh, my next question to you is going to be about what are the questions that HR leaders should ask vendors, and it sounds like that's definitely one.

Uh, how long do you retain data? Can you, you know, quickly list two or three more questions that, that let's [00:17:00] say I, as an HR leader who's looking to bring in, um, something to help me with workplace investigations, what else should I ask? 

Jackie Schafer: Yeah. I would be asking, you know, "How are you-- Are you-- How are you using my interactions with the product?

Are you training on... You know, it's separate from training on the data. Are you training on my interactions with the product?" And just, you know, make sure you understand what, what that looks like, um, from a, you know, a data lens. Um, another thing I would ask about is, "How does your product ensure that- there's human review of the output.

And I would say, you know, really scrutinize it 'cause sometimes it'll say, "Oh, well, we give you citations or references." But in my experience, that's actually-- it's very challenging. They'll just give it to you, and it's not hyperlinked. It's not giving you the exact page or the exact spot of the document.

And so in a long investigation report, it would take hours to manually go through and click on each one and scroll and [00:18:00] find where that source supports that, and people are gonna skip that. So I would say really think about does the tool provide a way to verify easily, um, exactly where this information came from? 

Because the work we do in HR is just too important to delegate entirely to a machine, even when we're busy. We have to make sure there's, um, a component of that product that aids us in, in verifying. Um, yeah, and I would just say, um- You know, really that going back to what we talked about earlier, you know, how much of the, this product is reliant on the user being good at prompting.

And, you know, h- or has the tool been thought through already to create sort of safeguards that are behind the scenes so that you're always gonna get a consistent work product. 

Nichol: Well, that sort of leads to how do I, as an HR leader, partner with legal [00:19:00] when I wanna bring something like Clearbrief in? What are the, what are the ways that you've seen it be really successful?

Do we jointly review Clearbrief? What happens? 

Jackie Schafer: Yeah. It's, it's interesting. I've seen it happen all different ways, where sometimes the legal department's the one that will see this and be like, "Oh my gosh, we need this for our work, and I think HR would also benefit from it, too." Again, they're like, "Please, like we need help with this relationship," and both getting on the same page about, you know, the documentation piece.

And, um, and then other times we see HR like, "Look, this is like just what I'm spending so much time on day in and day out. And, you know, our legal team, I don't kind of don't..." You know, I, I hear this kind of more where they're sort of like, "Look, I want them to split the, the budget with us, but this is our tool and we're gonna really own this, this tool."

Um, so I don't know. I see all different things. It sort of just depends on, you know, what the dynamic is. And I [00:20:00] think the one thing I'll say is like we're really aware of that dynamic and, um, I think I, I'm really proud of how we can sort of support that better communication between legal and HR, where it is, it's just going to make each of your lives so much easier to, to be able to have it be a very dispassionate sort of discussion.

Like, this is what the documentation says. Like, you know, this is my recommendation. Here's the documentation. And each, you know, each party can really, um, get a, a very clear understanding of, yeah, this is what a court would look at, too. If we have all the evidence laid out here, it sort of brings us all aligned on the same team. 

Nichol: Our audience is fully HR leaders. And so, um, if you're coaching an HR team tomorrow, what are the three documentation workflows that you would say start with? 

Jackie Schafer: So the number one I think would be that investigation report. It's just a delight for anybody who's had to do these [00:21:00] manually. Um, it-- like I said, it, you know, it, it provides a summary, an executive summary, but every single sentence of that, of the findings, they're all linked in Word, so you can see and double-check that the generative AI did what it was supposed to do and neutrally summarized that evidence.

Um, but the part that's left for you to fill in is the, my conclusions are, you know, it does not try to fill in what your ultimate outcome- Mm-hmm ... should be on this investigation. Um, two I would say is our, our PIP, um, documentation, which is very similar actually in a way to the investigation, where when you're creating a PIP, what we often miss uh, in HR is to actually document all of the sort of receipts that we have sort of explaining why this person is on a PIP.

Not that we're gonna provide them actually, you know, we don't need to like overwhelm the, the recipient of the PIP with all of the backup we have, but we need to know we have that [00:22:00] backup. So that's what the workflow does. It, every sentence of the PIP, it backs it up with, you know, emails or the, you know, the time cards and whatever this, the issue is about.

And you can click a button, then you can make a version that doesn't have all the, the references if you, if you wanna, you know, just give the person something that doesn't, doesn't show all the documentation you've gathered. However, we need that behind the scenes. So we, we built this tool in partnership with HR leaders who have to do a lot of PIPs.

It really helps you, you know, make sure you're buttoned up behind the scenes. Um, and it gives them a really nice, you know, layout, and it, it calculates all the dates and things of when you're gonna check in and- 

Nichol: Okay ... 

Jackie Schafer: very helpful.  

Nichol: And three? 

Jackie Schafer: The third one I would say is witness interview outlines. So, uh, this is probably one of the biggest ways your investigation can sort of veer off course, um, is again, it goes back to that consistency where you asked one [00:23:00] witness a set of questions and then you lost, you know, your notes on that or whatever, you know, you were sort of winging it for that.

You had to do five more interviews and the questions were different, you know, for each person. And so at the end of the day, you don't have a very clear one-to-one comparison of what each witness said about the situation. Clearbrief has a tool where you can actually dump in, again, those-- any documents you have.

It will... And you explain a little bit of the context of what you're investigating or what you're meeting with them about. It will create the witness interview outlines with notes for you about, "Hey, by the way, this document says this." So it's all-- And it's linked. So you can sort of test credibility of, you know, "Oh, so, you know, what did you, you know, what were you doing on that day?"

And, and you actually have the email at your fingertips to just, that documents what, exactly what they were doing that day. But, um, it just helps you sort of come very prepared to those, to those [00:24:00] interviews and also just make sure that every, um, discussion is consistent. So you, you ask the same questions to each person.

Nichol: So, uh, we're in the last, uh, couple of minutes, and so we like to close out with a little bit of fun and a game that the audience can join in. So, um, AI can be a powerful tool, but we know it's not perfect, and most of us have experienced instances where it's hallucinated or made up facts. So this segment we call Two Truths and a Hallucination, where, uh, I will share three incidents of AI error in organizations.

Two of them are true and one is not. And, um, I'm gonna ask you to identify which one is the hallucination. 

Jackie Schafer: Okay.  

Nichol: Um, so I'll read all three, um, and then at the end you tell me, oh, you know, A, B, or C is the hallucination. So the first one, uh, Google AI once suggested adding [00:25:00] non-toxic glue to keep cheese on a pizza.

That's number one. Number two, a company's AI meeting summarization tool created a transcript featuring a fictional employee. That's number two. And number three, a newspaper once published an AI-generated reading list of books that didn't exist. Which one was the hallucination? The pizza, the fictional employee, or the, uh, or the nonexistent book?

Jackie Schafer: So which one of those is not true? 

Nichol: Yeah. 

Jackie Schafer: Okay. Um, I think the book. The other two I think are maybe... Yeah, I could definitely see those being true. 

Nichol: Yeah. The, um, in 2025, the Chicago Sun-Times recommended 15 titles, each with real authors, but nine of the books did not exist. Oh, 

Jackie Schafer: wow. Okay. Yeah. [00:26:00] 

Nichol: So Jackie, thank you so much for joining us today. 

We learned so much. And thanks everyone. We'll see you next time. 

Jackie Schafer: Thanks for having me. 

Outro: Did you know our podcast is only one part of the AI+HI project? We also host an annual event featuring visionaries from the worlds of business, tech, and HR. Plus, we publish a weekly newsletter that includes topical articles and the latest research from SHRM, all to help you stay ahead of the trends.

To sign up, just make your way to shrm.org/aihi. And you can also follow SHRM on social media for even more clips and stories. Like, share, and tell us how you're innovating towards a brighter future.