Leveraging the Potential of AI and ML

Jason Lee: Well, thank you all for joining us today. We hope to make this educational and fun and informative. Joined by Jay Venkat, who's our chief digital and information officer. There's also a QR code. So, if you don't mind, please use that to submit questions in advance. We should leave enough time at the end of this session for some Q&A, and if not, we will follow up after.

So, let's just kick this off. This is going to be a back and forth between Jay and I. I'll ask some questions. He'll ask some questions. We'll certainly try to address the major topics that we think are at play in this. And, Jay, I think where we'd like to start—there's certainly a lot of hype about AI and ML in the news today. How should we really think about that as a small business owner?

Jay Venkat: Fabulous. Thank you, Jason, and thank you all for coming. Let me ask you, show of hands, how many are in HR? Okay. Just one lady here. Okay. Well, let me ask you differently, how many of you care about your employees?

All right. I hope many more hands went up. That's good. So Jason asked an important question, which is—there's a lot of hype around the AI and ML. How should a small business think about it?

Well, first of all, I do genuinely believe that we have crossed sort of the hype territory. Yes, there's a lot of hype as it relates to sort of ChatGPT and what happened in the last six months, but many of the tech companies have been working on AI and ML research and development for almost 10 years now.

And this technology has been maturing slowly. It wasn't just ChatGPT came on board and everything turned upside down. In fact, if you talk to Microsoft, Google and Amazon, they've been working on AI and ML for quite some time. The technology is mature. Use cases are starting to get mature and applications are quite relevant. And if I think about a small and medium business, I would first start with employees. What does this really mean for your employees? And I do think it's as great a shift in productivity as happened with the industrial revolution many moons ago or with the advent of the internet not so many moons ago.

But it is a huge productivity game changer for employees. In every part of the value chain for a small and medium company, if you're in sales, you can target your prospects better. If you're in marketing, you can qualify and score your leads much better with this technology. If you're in customer support, your life just became significantly easier, because even if you are a very new customer support agent, you can capitalize on knowledge in a way that you could not do before.

And if you're in technology and engineering, you can use the new technology, especially around generative AI, to be able to leapfrog productivity from an engineering development perspective. So no matter where you are in the organization in a small medium business, clearly people have multiple roles in the company, especially for small companies. And this is a game changer from a productivity perspective. So, that's number one reason why you should care about AI and ML. The second is your customers, and your customers are all looking for different ways in which you can add value to your customer, and AI and ML gives you opportunities to add value differently.

And if you don't do it, your competitors are going to do it. So how do you actually make sure that you're providing the most value to your customers and utilizing AI and ML to be able to do that? Either by embedding the technology into your own offerings or by partnering with somebody who can actually provide that for you, and your employees and your customers can actually harness it and you and this becomes a place for competitive differentiation, if you will.

So, short answer to your question, Jason, is I do genuinely believe that small and medium businesses should really care about this, should invest in learning about it, should think about use cases for their employees, for their customers and how they can use the technology to leapfrog competitors. And finally, the game changer in the last five years or so has been threefold as it relates to this technology.

Firstly, data. There's a lot more data available that's tagged, consolidated and stored in a way that it can be utilized. The second is the funding for these technologies has increased significantly. And the third is the computing power that is required to make good use of these technologies has also expanded significantly.

So it's now mainstream, it's now real, and it's another reason why, as a small and medium business owner, employee, stakeholder, you should really care about it, which is a segue into my question to you, Jason. You run and lead payroll and payroll related products for TriNet. Obviously, for small and medium businesses, getting paid, getting your employees paid is a critical operational activity. But out of that operational activity hang many, many other important operational activities, whether it be your benefits, your voluntary benefits, your deductions for retirement plans and so many other things. How is AI and ML affecting the way of the world of payroll?

Jason:
Got it. Well, thank you for that. I think it's a perfect place for us to start. We think that the payroll and workforce management and tax space is one of the most ripe for this type of technology, and I'll go through several use cases that, we think, point us in that direction. Clearly, as you touched on customer service, the advent of chat bots, natural language processors that simulate a human answering questions or providing information to the workforce, it improves customer satisfaction, it improves response times, it reduces the reliance on individual people being available to answer the phone and having the knowledge to do so. There's a host of use cases there and later, you know, perhaps we can discuss some of what we've seen in our customer base that have taken advantage of that, the impacts that's had on their on their workforce.

Another huge area of opportunity for us is really digitizing the compliance service delivery that we offer today. So, TriNet, you know, obviously is a leader when it comes to maintaining compliance with all the legislative obligations of employers in the United States to the extent that we can actually build systems that will continuously audit the data, will continuously learn and make predictions about where there might be problems, where there might be fraud, where there might be errors, but also where there might be potential compliance gaps and recognizing those early—that's a huge opportunity space for us as well.

We see, you know, again, we collect somewhere around 1.8, 1.9 million service requests each year. And those are people generally asking us to do something on their behalf, potentially asking for research on something, but in large part, asking us to do something that's too complicated to do on your own, to the extent that our natural language processing systems can learn from those, learn from the years of history we have. We process roughly 12 million pay slips a year. We've done that over 15 or 20 years now, to the extent that we can mine that data to actually be proactive and intelligent about potential issues rather than capturing those after the fact.

We want to do that as well. We're also looking at a lot of predictive algorithms that can kind of define, you know, best, worst, most likely circumstance, and we're looking at applications of those for compensation plan set up, for developing job descriptions, for identifying the best benefits. Again, we have 300 million health enrollment records.

So, really, that's what we're looking at is where can we take advantage of our scale to really train these models and where can we make our customers' lives easier, their employees' lives easier, while maintaining kind of best-in-class service that TriNet is known for.

Jay:
That's great. Thank you, Jason.

Jason:
Yes, sir.

Jay:
And as you were talking, I couldn't help but look at the question in the poll, which says, "Are you currently using AI in your business?" And it's split exactly 50-50. 50% yes and 50% no. Well, I would argue that you are all already using AI in your business, even when you type an email that say, uses Microsoft Outlook and it suggests the completion of a sentence for you, that's utilizing AI.

Jason talked about payroll. When you pay your employees, there are banks involved in the process and banks for the last decade have been using AI for fraud detection to make sure that this wire transfer that's coming from you and going somewhere else is actually a legitimate wire transfer. So I like that. This is real time. The more I say, maybe I should say a few more things and we'll go up to 100%. So, I do think that it's, that's why I meant that it's actually quite mainstream, which sort of leads us into more of a conversation about TriNet and what we could do there.

Jason:
Yeah, I mean, I think that's great. So if I were a leader of an SMB, if I were a CEO how would I get started into this? I identify it. I know it. Maybe I'm one of the folks that Michael referred to earlier that hasn't yet uttered the words ChatGPT. How would I, like, get my feet wet in this?

Jay:
Yeah. And as with many of these new technologies or new world technologies, the first steps can be daunting. You're always wondering, "do we need to make a lot of investment? Is this going to be a significant expense? How do we get the right talent?" So those are all valid questions, but all need to be tackled with some simple baby steps. I think the first and foremost piece is to understand what are the real practical applications for your small and medium business, for your employees and for your customers. And I think most small and medium companies will be able to articulate that. Where do we need extra intelligence so that our employees and our customers can do their jobs better? Where can we actually have productivity gains by using AI and ML?

So job number one is identify what those use cases are. Where can you add the most value with AI and ML? Job number two is actually related to data. Very often, efforts related to this technology fail because the underlying data doesn't exist. It's the garbage in, garbage out. Now, the good news about this technology is that you don't have to spend years organizing the data. You only need to organize it to some extent. And then the technology can harvest what you have and be able to get insights out of it. But, as a small and medium business, sometimes all your data doesn't even sit with you. It sits with many other partners. It sits with third parties that you do business with. It sits with your customers.

So how do you make sure you have a decent handle on the data that you have that will actually help you with these practical use cases? The third piece is who do you find to be able to make some baby steps here or even go further? As a small and medium business, it's hard for you to have an engineering department at scale. You're not going to be able to develop these technologies much further out based using your own staff. The good news is there's already been significant investment in this area from the same companies that are going to be providing you cloud services, from the same companies that provide you HR services.

So, leveraging your partners and their capabilities in this area is a smart way to go lower expense, shorter time to market, as far as using these technologies are concerned. The last piece is related to talent and again, for a small and medium business coming up with talent that understands the space, is an expert in the space, and can leverage the space. It's not easy.

So be targeted in, again, utilizing the capabilities that your partners have, because they have thousands of people who are skilled in the science, in the technology. How do we use their capabilities in order to leapfrog? And then folks like TriNet who are partners for you in your day to day operations—we are essentially helping you with your HR, your benefits, your compliance, your payroll. And we have made investments, which we'll talk about today and are planning to make investments on a go forward basis in this technology. How do you use us, our services, our products, that we will be embedding AI and ML into and use that for leapfrogging in this space as well?

So those are some of the areas where, and I think this poll will be interesting, I'm very curious as to what the results of this poll are so that we understand what's on your minds as well. I might take our conversation in a slightly different direction, which is obviously—everyone wants to add value to employees, to customers. There's a lot of hype. We're going to use partners, we'll get some in house talent. A lot of the conversation in the media has also been about the ethics of AI. And even though that is a broader, abstract question, there's still a lot of relevance for that question as it relates to the smaller medium business space as well.

So Jason, my observations, considerations, as you think about the ethics of AI and how should our small and medium businesses think about it?

Jason:
That's a great point, Jay. As we've talked with our customers and we've talked with experts in the field, we tend to narrow the ethical considerations regarding AI into kind of three large buckets. So first, there's the bias that may be in the training models. I think that's where leaning on the expertise of partners is very important, so that you have someone who is an expert in the field, they can identify any inherent bias in those models. And what you don't want, what I mean by that is you don't want the algorithms to draw the wrong conclusions because it was looking at too narrow a set of data for the indicators.

And so you want a broad data set that more accurately represents the picture of the domain you're looking for. The other aspect of that, in addition to the expertise of folks really studying those models and making sure those are trained in a non-biased way, is explainability. What we see oftentimes is that because in some of the less mature AI models, the answer comes out, but you have no idea how that answer was developed.

Again, you may not see that bias and you may not be able to adapt the algorithm to respond to it. So having very strong explainability in terms of the specific data points that were evaluated to generate the conclusion is very important to us. The second area that we really see this is really all about data privacy. You know, data privacy is something that is becoming increasingly global and local with just an enormous explosion of jurisdictions and laws relating to the collection and use and the authorization of your employees to allow you or your partner to authorize and use that information for the purposes of AI.

So this is a very tricky issue. This is where it's great to work with partners, people that specialize in the field. And if you are going to do something yourself, you absolutely want privacy attorneys looking at this and ensuring that you're not running afoul of any regulations. And then I think the third main category that we see is really about how do you protect, especially when you talk about generative AI, copyright and trademark rules?

And so, you know, again, I think the way that we're approaching that, it's very much an emerging field. I don't think that the industry or the market has figured that one out yet. I just saw Microsoft has been really on the forefront, I think, in terms of establishing and holding certain liability with themselves. But from our purposes, we're really looking at generative AI as an input to decision making, as an input to a process, but not as the final product. And that's your basic safety guard to make sure you're not going to run afoul of those things. So those are the things that we're looking to bake into the product, Jay, to make sure that we're doing this in a very ethical and transparent way that is in accordance with our values as a business.

Jay:
Excellent.

Jason:
Yeah.

Jay:
That's a good segue into what is TriNet doing more broadly as we think about AI and ML and how are we playing in the space? Clearly, we're not a small and medium business. We're a midsize company. We have about call it somewhere close to 4,000 employees, we're a $6 billion public company. So how do we think about AI and ML and where are we investing as it relates to the space? And I'll, again, start by dividing it into what are we doing for our employees and what are we doing for our customers, i.e. many of you?

Let me start with the customer aspect. And as Jason mentioned, if we think about our business as really helping everybody with HR, benefits, compliance, and payroll. In each of those domains, we're thinking about how can we add more value to our customers and help them grow their business?

How can we help our customers be successful in those areas? Jason already talked about payroll. What are some of the technology investments we're making there, whether it be fraud detection and continuous auditing to self-service payroll, many areas that are going to be enormously useful for payroll administrators going forward?

If we think about benefits—benefits is an extremely confusing area. Decision making for employees and even for administrators is difficult. There's a plethora of plan choices. What do I pick? When do I pick it? Do we do a high deductible health plan or not? Some of these decisions can actually be done better by using decision science powered by AI and ML.

And TriNet is investing in those technologies to be able to help our employees, our worksite employees, i.e. employees of our customers, really make better decisions as it relates to their benefits. When it comes to HR, there's a whole spectrum of activities. Everything from finding the right candidate for a new job, all the way to how do you engage in helping employees actually learn better?

How do you engage with employees and do performance management better? These are all, again, for a small business, these are activities that hopefully should take less time and yet be very effective and efficient. And that's where TriNet is also investing to make sure that our suite of tools is designed in a way that uses the power of AI and ML to help make better decisions in HR.

And as Jason mentioned about compliance is an area that is very ripe for investing in this technology. And we're going to be doing quite a bit of that there as well. And as it relates to our internal colleagues, we want to help make their jobs better. We want to have them do less manual activities, be able to go to higher order tasks, give them intelligence—we call it amplified intelligence—which is, you're given intelligence, but it's not intelligence that makes your job obsolete. Rather, it's intelligence that helps you do your job better. And that's a big difference. We're not trying to eliminate the human being. We're helping the human being do their job much better.

So that's how TriNet is thinking about investing. We think about it as a core element of our product roadmap. As new products come out every year, you'll see sprinklings of AI and ML into all of those and we don't think about AI and ML as something that is just done by a special set of people sitting in a corner, separated from the company; we're embedding AI and ML into our normal operations.

We're asking all our employees to embrace and be part of how we deliver it into our product suite. So Jason, as we talked about AI and ML and productivity and how people are thinking about embedding it into workflow and what does it do to the intelligence that a human being has in performing their jobs, a question that keeps coming up and is very relevant for our small and medium businesses who have fewer employees is what could this do to the jobs landscape? Is this going to eliminate a bunch of jobs? Are we going to be in a territory where a small and medium business needs half the people it once needed? Again, there's a lot of hype around this. There's a lot of theory. What's your perspective?

Jason:
Yeah, great question. And, so I certainly watch the news. I read the reports. I see the hype as well. I can really only relay from my experience talking with customers who have adopted some of these technologies. What we found from our customer base is that they're able to transition a portion of their workforce into more strategic jobs than they had before, which does a couple of things.

It actually improves—we have several documented results now where it's improved—retention of those employees because you get less churn. People are actually having the opportunities to grow and do different things in their career. There's less of a ramp time because you're not having the same churn of employees and that same onboarding time every six months, taking 30 days to ramp someone back up.

So we're seeing an elevated productivity from that respect. And we're also just seeing higher customer sentiment, higher customer satisfaction because they're able to spend more time with people and they feel like the people they're spending time with are more knowledgeable in the subject area. So while I think it would be a bit naive to speculate that there won't be any efficiency as a result of these, we're certainly not seeing as a one-to-one correlation; we're actually seeing what that does is open up tons of growth opportunities for the existing upscaling of talent and population.

I also, just before we go to the next one, I wanted to jump to the last point you just made about all the different technologies applications. One of the really interesting ones that we're finding as we talk to partners is there's a lot of investment right now in wellness and employee wellness and employee engagement. And we're seeing a ton of research and innovation around detecting patterns of when we may need to encourage an employee to take some time off, when we may need to look to take an employee into a different job. Again, all with the lens of improving the experience for the employee, but ultimately reducing cost for the employer as well, because you're reducing your attrition and turnover.

You're probably reducing health care costs as a result of that as well. So, you know, it's very much this technology can be applied very much to improve the lives of the employees, not just the bottom line of the business. And so that's what I wanted to add that last conversation.

Jay:
Yeah, and that's a really good point around improving lives, upskilling, being able to expand the capabilities. You take your own capabilities and skill sets and the applications. Now it opens up an entirely new universe.

A complete sidebar by one of our colleagues at TriNet. She has a kid who is in his third year of college. Three years ago, when the kid was choosing college majors, he decided to choose linguistics. And the mom was very upset because she thought that it was a completely unemployable undergraduate degree and now kid is a junior and when ChatGPT burst out on the scene, he's already locked in internships, job offers. They come in all the time for training large language models and being able to apply linguistics principles to data science is a booming field. So in the same way, it's not that the jobs go away. It's just a shift to a different type of job, and it all behooves all of us to invest in ourselves so that we're continuing to be relevant.

Jason:
Yeah, that's great.

Jay:
Are there any questions that popped up? Please.

Audience member #1:
How do you foresee dealing with liability when AI is wrong? You brought up a great example of, "Hey, it should tell us to take employees off." And I've been using it a lot and as amazing as it is, it also lies and is stupid often. And so there's this weird gradient where, who's responsible if there's a mistake there, but it wasn't done by a human, but a human gets hurt because of it. Like, how does that work?

Jason:
Yeah, cause I think it's a great point. I think at the state of where this technology is today, this is very much a suggestion that a human judgment needs to take action on. So to the extent that we are not releasing, we're not allowing the AI to make decisions that would actually influence and affect people. We're allowing it to provide with strong traceability and explainability, why it's suggesting X or Y. I think that we're gonna have to train these models to be much more robust but ultimately, in the same way that software vendors have long kind of held the liability for how you use the software versus the software's integrity that I'm providing. That's probably gonna be the separation that we see in the market as well, is that the providers of the technology will provide it simply as a technology.

In order for consumers of that to feel really comfortable, we're gonna have to see a lot of robust protections put in place. So I think like, that's kind of where we are today. It's hard to predict like 10 or 12, you know, 15 years in the future what it would look like, but that would be my best guess.

Audience member #1:
I mean, I assume there'd be far less mistakes in the future.

Jason:
Right. Yeah. It's a great question though. Thank you for that.

Jay:
And the way we think about the same piece in our business is that, let's say we have support agents who are answering a question related to payroll taxes. We have to get it right because otherwise we're liable for giving you a wrong answer. So when we think about application of this technology, we're first starting with historical data and to see if when you apply the same question to historical data and ask the machine through an algorithm—what's the answer that it gives and how correlated is it to an answer that was given by an expert support agent in the last five years and how does it correlate to documents that exist on the topic? So those steps have to be cleared before we can confidently say that the algorithm is going to answer a question about payroll taxes accurately or not.

Jason:
Great point.

Audience member #2:
What data that you're aware of that leads to showing how AI is impacting retention of employees. As I manage multiple businesses, I understand that as you move people in the direction of their dreams, you tend to retain them for much longer. And it does help with your culture, right? I guess my question is—is there any data or information that has been calculated around on how AI might be helping companies retaining their employees?

Jason:
Yeah, absolutely. So, the question, if I can paraphrase, is what data, if any, is available that would support the hypothesis that AI is actually improving retention and having outcomes that show that? So, we've certainly discussed with many of our partners the data from their research. We've also looked at secondary research. I don't think we've commissioned primary research ourself. But I'd be happy to share what I have found with you after the… Yeah, okay.

Jay:
Yeah. And one of the pieces we're seeing… it's still anecdotal; we don't have hard evidence for it. But as we help employees make better decisions on benefits, healthcare choices, vision, dental, medical choices, giving them the confidence with these tools that they're actually making the right choices for their families—that goes a long way in terms of the employee experience and them feeling that they are valued member of the company and we help them curate the right choices.

It's the same with learning. A lot of employees these days say that they hold their employer in higher regard if the employer helps them invest in themselves and learn, and this technology applied the right way can offer bite sized learning opportunities for employees that they didn't historically have, which again can go in the direction of retention.

There's a question back there.

Audience member #3:
Yeah, what about AI being used regarding stress and mental health in the workplace? There's a lot of stress, a lot of mental health going on. That's a big thing right now in, you know, the world basically. Are you doing anything regarding HRs for, you know, using NLP to determine how you can help your employees in that matter?

Jay:
So, if I heard it correctly, it's about reducing stress in the work?

Audience member #3:
Mental health, you know.

Jason:
Mental health, stress, yeah.

Jay:
Yeah, mental health and stress. I mean, it's sort of connected a little bit to the employee retention question. I don't know that we have quite figured out exactly how to apply the technology to improve mental health. At the same time, there are areas that are related to employee wellness where the technology is already being used a little bit. So, for example, there are many technologies, TriNet is also introducing some of them around health advocacy. And health advocacy makes it easier for you to think about your own health, your family's health.

You quickly, through answering a few questions, recognize, first of all, whether you have a problem from a stress perspective. Is this something that you need to go to the doctor for? And building that muscle and routine. There's also apps that, again, through our portal, we will over time curate, which is around reminder based better behaviors, whether it be working out, whether it be going to a physician, and those reminder based behaviors, and helping you take better control of those behaviors is one way in which these technologies can help with mental health.

Audience member #3:
Yeah, because I think the key too is to be able to live a balanced life, right? Work life balance, so I think there's opportunities there using the technology, I would think.

Jay:
Absolutely. There's a question back there.

Audience member #4:
Yeah so I'm Kiana, I'm a product manager at Red Hat, IBM owned, for those that don't know. So my question is a serious question, but I'll try my best trying to make as easy as possible. But related to HR, since you guys are involved in your day-to-day operations and helping small and medium-size businesses kind of outsource some of the HR decisions. As a woman of color in technology, I'm often looking at the bias piece, which is one of the buckets that you touched upon, but being directly in the crux of helping influence hiring decisions, like what data is there on there on knowing, like, if these hiring decisions are being done ethically in a way that it's actually inclusive? The datasets are too narrow. Then it's not going to be inclusive of all that the job application pool to its fullness and inclusivity isn't yet built into these models, so I'm just curious of like—is there a way that at least thinking about it, because I know and to the mental health question, we have no data on this yet.

We are guinea pigs. We are currently in a situation where we're messing around and finding out after. But at least with the mental health piece, the automation does help and help buy back time to spend with your friends and times and family. Just giving my answer, a little piece to that. But to the original question of the hiring decisions how can one be responsible when those decisions are brought to them to act on behalf of those companies to make sure that, you know, they're making the right decisions and they're doing so with having the human oversight that is needed? That was a lot.

Jay:
No, no, I think it's a great question. I don't know if everybody heard it, but is there again around—how do you not introduce unconscious bias into your decision making? And this is still very much in its infancy. And I would at least right now, the problem is being tackled in two ways. One is how do we make sure that our training data sets are actually more inclusive? Because the algorithm essentially feeds on the data that's given to it. So if the data is not inclusive, the outcomes are not going to be inclusive. So there's a lot of work being done about tests that will certify whether a particular data set is an inclusive data set. And if so, can it be used for training purposes or not? And if it passes the inclusive test, then it can be used for training.

And there's companies that are coming up whose whole job is to create these certified inclusive data sets. That's sort of one aspect, but that will take time. And the stamp of sort of authenticity in the training data will take some time. In the meanwhile, the notion is very much around how do we add, don't just rely on the algorithm and the data for decision making, but use it as input that helps you make better decisions, but not necessarily the only input to making the decision?

I'm also seeing a hand up around time. How much time do we have? Can we take one more question? Okay, we'll take, there's a couple of questions. Why don't you both shout out your questions?

Jason:
Not at the same time.

Audience member #5:
So, you mentioned about investing in education for the employee and I think that most employers are quite apt to invest as long as they can get an ROI from that that is trackable in a short amount of time, of course. Is there any use of predictive analytics to help surface the kinds of education that employees can have with the shortest ROI or time to ROI that TriNet's working on?

Jay:
Maybe a slightly different angle to it, which is we're trying to figure out how to make learning management systems more effective. So whether you take a two-hour course or a 30-minute course on the same subject, how do you make that content better aligned with your interests and how you learn? That's an early area of research. I wouldn't say that we're necessarily doing it, but there are those who are experimenting with it, which hopefully we can benefit from.

Audience member #6:
So, you previously mentioned about the linguistics in terms of like, a useless degree and…

Jay:
Well, I won't quite say that.

Audience member #6:
But you said you were able to implement that using AI. So, my question is how is that gonna be useful to communication and trying to understand in the workplace for employees and consumers. Or like, how can we use linguistics or what can we do, in order to get more conscious in communication with AI or like understanding data as such?

Jay:
Yeah, the real application right now is around unstructured data because still a lot of our historic data is still unstructured, whether it be voice, whether it be chats, whether it be conference calls, scribbles on a piece of paper. And at the end of the day, most of the AI technologies really are effective when you are able to get to the structures behind the language that is there. And that's what essentially the large language model feeds on. So, the application of linguistics is to be able to help us take that unstructured data and be able to make it structured enough for the algorithm to utilize it. That's simplifying it a lot.

Jay: Very good. Thanks for your time.