Storytelling About ishield With Kesavan Kanchi Kandadai

William Tincup: [00:00:00] This is William Tincup and you are listening to the Use Case podcast. Today we have Keevon he’s on, his company is called Ice Shield, and we’re gonna learn about the use case of the business case for Ice Shield Keevon, would you do us a favor and introduce yourself and I Shield?

Kesavan Kanchi Kandadai: Thank you William.

I’m super excited to be here. I’m k Shevin co-founder and CEO of I Shield. I Shield is a generative AI platform that helps HR teams create talent communications 10 x [00:01:00] faster and at scale and bias free. So that’s what we do. Yeah.

William Tincup: Yeah. So talent communications give us some examples of that.

Kesavan Kanchi Kandadai: The simplest form of talent communication is a job description, right? And we have recruiters and sourcer who are creating a job description and they need to publish the job description. Typically, it’s not enough that you create a job description, but you also create multiple versions of it for internal job posting.

Then you have a social media post that you want to promote the job description. Then you want to probably send it out in a job alert email campaign. Then you will do a another version for getting referrals. So a recruiter and sourcer does all of these content that they’re creating. And so we’ve created an AI platform that can help companies create that content using the ai.

So it’s faster. [00:02:00] And on brand and consistent all the time.

William Tincup: So are we using chat, G p T at all? I love the part of the AI because it gets smarter, right? So the more you use it, in theory, but in practice, the more you use it, especially with job descriptions, the smarter it gets with kind of the nuances of your company or your hiring managers or your recruiters, et cetera.

But Yeah. So the question is are we interacting with chat G P T or large language models in general? Yeah.

Kesavan Kanchi Kandadai: So chat, G P T gives you a average job description, right? And it’s not customized to the brand. It’s not free of biases or errors. Okay? It’s not personalized. So what you have is like a classic case of a good English speaking intern writer piece of content.

But the amount of time you take to make it precise for your brand is probably the same as you creating it manually. That’s the state of affairs today. I’m sure things will improve. We’ve created our own AI models. We [00:03:00] filed patents for some of these. And we do three things. Of course, we use a large language model that speaks good English or that gives us good English content, but we trained our models on millions of job descriptions.

So our our millions of all talent communications, I would say job descriptions. One example, this could be an assessment. This could be a candidate nurturing campaign. This could be a social media post or an employer branding blog. It could be all of that content. So we may trained our models on millions of such data points, so we understand the architecture and the content.

Then we also our AI understands the brand. So when we deploy for a a company X, then it would learn about all the brand personalization that is relevant for that company X. And most importantly we have built models that detects biases in AI recommended content. And this biases could be towards [00:04:00] gender, race, religion, language, whatever that is.

Our AI detects that there is a recommendation or a piece of content that has bias, so it auto eliminates and replaces that with bias free content. Essentially, in a nutshell, we make the AI really work for you as a brand. So I gave you a long answer. Yes, we do use the, we do use the large language model, but you’ve built our own.

We built our own for an enterprise. That’s not enough. So

William Tincup: first of all, this is fascinating. Biases one person’s bias might not be another person’s bias. In fact, I asked someone recently, what’s the difference between preference and bias? And they said, preference is how you justify bias, which I found fascinating, right?

So some folks, This is also true in the sourcing world, so I know you know this, but like somebody, people have a mandate, like we need to recruit more veterans or more people with disabilities. And maybe even more specifically with people with disabilities, et cetera. So what if [00:05:00] one of your customers actually wants to be biased or maybe prefer women for, the, an engineering position, et cetera, which would normally come out as bias.

But can they tailor it? Is there a way to modify it for that particular post?

Kesavan Kanchi Kandadai: Of course. That is a preference as you rightly put it. And and, it forms broadly under the realm of, accompanies taking a conscious action. Ah okay. So I want to, we want to differentiate between what is unconscious bias and what is conscious bias.

So unconscious bias, Is something discriminating or having a piece of language that discriminates women or stereotypes women is a unconscious bias. But what the example that you have given is a customization for the brand where they, so in our platform, they can customize and say, this is the theme that we are.

Creating this content for if you’re writing a women’s day post, it is going to be about women. So yes, so that’s part of the customization. It’s conscious. [00:06:00] And so if it is conscious, then our filters let that through because that’s the choice that the customer has made. That’s first point. The second one is we don’t have any auto correction modules for our bias.

They’re all assistive in nature. That is, we make a recommendation and if we still leave it to the human to decide whether we want, they want to take that recommendation or let it go so they have a discretion. And that is how they can align it to their own goals and say, yeah, yes, I understand that this is, this could be biased, but this is aligned with the program that we are running, so I will let it go.

So that’s okay. So our platform has no. We have not built any restrictive capabilities yet.

William Tincup: I love that. So other than job descriptions, cuz you’re broadening out with talent communications, what else are your customers using ICE shield for? Yeah,

Kesavan Kanchi Kandadai: We are, so if you look at the overall talent lifecycle, the first part is the whole employer branding.

So companies [00:07:00] use us for creation of employer branding, content carrier website copy, blogs, articles. And they also use us to give them a audit. They give, use them to do an audit of their all their existing content to see how aligned and how branded that is. And how bias free that is. So they get, they use it for branding purposes.

The second bit is recruitment. So recruiters and sourcer use our con, use our AI to create job descriptions, candidate campaigns, nurturing campaigns, candidate emails. Campaign content and the and, emails and all the recruiter work, so to say, wherever they’re creating content.

So this is how companies are using our product

William Tincup: today. It’s almost if you’re gonna use Microsoft Word to open up and create a document, whatever that document might be, if it’s a social post or whatever, then you don’t need to do that. Correct.

Kesavan Kanchi Kandadai: Yeah the a different way I put it is it’s like [00:08:00] having a full-fledged content marketing agency work with you.

It’s just that it is an AI.

William Tincup: So what’s the workflow for them if let’s say we’re working with a company and they’re going through 500 jobs in a year they’re gonna hire fire 500 people. Is it recruiters and employer branding and sourcer? Like what? And then what I’m really trying to figure out is what’s it connected to or is it connected to anything else within their kind of, their talent their technology stack.

Kesavan Kanchi Kandadai: That’s where we differentiate ourselves from a lot of other tools people may want to use. We have built the integrations with the most popular applicant tracking systems and talent, c r M systems. So let us say that the job post. Or the communication originates in greenhouse. Then we have a plugin for Greenhouse.

So you get the content recommendations right in the Greenhouse platform. So this could be Greenhouse, this could be Lever, this could be a, or you could be using Phenom [00:09:00] people, SmashFly. So we’ve integrated with all these platforms, so you are not navigating away from the system where you are currently working and you’re getting the content recommendations right inside the platform.

So that’s the use. Fantastic. Yeah. Yeah. But if customers do want greater account management and team management, then we do have a web app they can use. But our mo most of our customers use our plugins in their HR tech ecosystem. Yeah.

William Tincup: Right, almost as an overlay. So they’re doing it.

And if it needs to be published or it needs to be housed again in the ATS and on the marketing side, it would be also, like you said, ature. It’d be in some of the CRMs. I think you mentioned is Smash Fly, so that would happen there as well. Yeah. Smash IP

Kesavan Kanchi Kandadai: on people. Yeah. Yeah. Have

William Tincup: you integrated with, or you have, you, have you been asked to integrate it with some of the programmatic, the Pando Logic app cast symphony talent, some of those folks.

Have you been asked to integrate with [00:10:00] the programmatic job at, it’s kinda omnichannel.

Kesavan Kanchi Kandadai: Job. That is correct. Yeah. Yeah. We’ve evaluated symphony talent. Integration. One of the use cases there is production of variations of the add postings Oh. Interest at a scale.

So you

William Tincup: can do you can do AB testing.

Kesavan Kanchi Kandadai: We can do AB testing. Yes, exactly. Oh, cool. And you can do, and you can do variations of the same, content and you can keep refreshing, so to say over a period of time. Of the ad service. So yes, we have that’s in beta right?

We have, we are not live yet. It is something that we’ve been asked we have a product for that in beta.

William Tincup: Love that. Okay, so let’s do some buy side questions because the podcast is trying to help all of us, but practitioners learn how to buy software. And this is a new, this is a new category.

This is, I, and I wouldn’t say, investors hate it when I say new category, but if they haven’t bought, if they haven’t bought something like eye shield before, what [00:11:00] should they be asking? What types of questions, if not specific questions, should they be asking you and your team?

Kesavan Kanchi Kandadai: So the, there are four questions that customers ask us repeatedly.

Because this is an ai, everybody is concerned about data privacy and data security. Because if we are learning about a brand, then people want to know where is that right? Where is that intelligence reciting? So that’s a question that people ask us. The second question is how we generate branded content.

So they want to understand how unique it is going to be compared to let’s say generating the content for some other brand. Because quite honestly, and this is again, if two teams use chat, g p t, they both will get almost similar content recommendation for the same query.

But here companies are concerned for us in terms of how unique is the content recommendation. The third question that companies ask us is, Intelligence of people and persona [00:12:00] and audience is locked up in HR systems that already exist in the company, right? That, so they ha they know who their candidate is, what the candidate pool is, what their mix is.

All that information or the applicant data and the information is locked up in ATS, C R M. Various systems. So their question, the third question they ask us a lot is around personalization. In order for you to personalize, you need integration and you need to know these insights for you to generate personalized content, right?

So people ask us a lot about, are you integrated with, and they list a few software they use and that’s what they expect. And the fourth question really is around people. Communications, talent, communications needs layers of de-biasing. So they are very curious what types of biases does it, at what level does it work?

Can it work at a state level? Can it work at a company, organization level? What about different geographies? So there are a lot of questions around [00:13:00] unconscious biases. There are a lot of questions around. Compliance. There are questions around state level compliances and they really will have a checklist they want to go through to say, are, do you meet these requirements or not?

So we keep com customers ask us these four questions, security, data security, and privacy. Personalization through integrations and insights and bias.

William Tincup: I love all of those questions, especially the personalization which you talked about at the very beginning. It’s yeah, large language models that are out there, that’s great, but if it’s not personalized you’re still gonna have to do a lot of work.

I used chat G p t this weekend, and I had it right by bio. So basically said a thousand words or less. Write William Tincup bio. Yeah. And then I flipped it around and I said, okay. Thousand words or less. Write William. Tincup obituary. Yeah. Oh, okay. Which is a little dark, which I understand.

But it was the obituary [00:14:00] was, 90 per 95%. It was pretty spot on. And yeah, I found, I was found, and again, that’s generalized, which people don’t understand that, that it is wonderful. It really is a kind of a fun thing to do, to like really play with the large language models. And again, to your point earlier, it can get you generic content if that’s what you need.

Kesavan Kanchi Kandadai: That is correct. It gets you started, it gets you started, it gives you a decent draft. But if you tried the regeneration process three or four times, you start getting more and more garbage. So it doesn’t become more precise it all upon how you query it. That’s right. So there is a barrier for an enterprise to adopt it at an enterprise scale.

And what we are doing is giving that last mile. Problem solution to companies to actually make it work for interface. That’s what is probably the thing that companies need before they say, okay, I’m using generative ai.

William Tincup: Right. And does, do you find that it [00:15:00] helps with prospects if they’ve played with or, ticking around with open AI chat, G P T or any of the others that are out there, does it help if they’ve, they know how to run queries?

Kesavan Kanchi Kandadai: I think everybody has already tried it in some shape or form. So I think that question, people have crossed that barrier already. I think, I also think that a lot of companies have thought about integrating it, but are also scared. We saw what happened with Samsung, where people are, yeah. Yeah. I think that I was at a conference where we asked, I asked 72.

CHROs were in a room, they were discussing a topic and only 10% were ready to use it at an enterprise level. But almost everyone was using it personally. This is the so they’ve learned the chat gpt and it helps that they know AI can do a lot of good stuff, right? But there are barriers to crosses how I look at it.

William Tincup: Do you do you feel like again, running queries even in, in Europe, because [00:16:00] it’s more specific to the company and to recruiting in general? Do we need to teach them queries? Do we need to teach ’em how to write more specific, like a job description for a software engineer? Like I could write the query myself, but do we need to teach ’em that

Kesavan Kanchi Kandadai: to use our product?

No. Okay. Okay. When we deploy, we have abstracted all of that learning into the ai, and that’s how we deploy it for a company. And so they still use Simple Plain in English to say, can you generate a job description for the following post or for the following role and give some basic information or data points.

And that’s it, that’s all what they need to do. Gotcha. No query training or any is required.

William Tincup: Thank you. That’s perfect. Alright. Your favorite part of the Ice Shield demo. When you get to show somebody that’s never seen it before, maybe not even understand exactly what you do, what’s your favorite part of the demo?[00:17:00]

Kesavan Kanchi Kandadai: Before chat g p t became popular. We already had the generative ai and when I used to demo that, the AI creates the content people did go Wow. But there was also very healthy skepticism, right? But today that wow factor has gone away. But people do say, wow when they see. Unconscious biases being highlighted by the tool, right?

And you make manual edits or, AI is generating the content, all these biases are highlighted and you so that people do see, okay, wow. Now I understand why AI requires these additional filters. I can see the value add. So that’s my one of my favorite parts.

Of the overall IS SHIELD demo.

William Tincup: Love it. Listen, this has been absolutely wonderful. I love what you’re doing. How long you been doing, how long is I forgot to ask you at the beginning, how long has is SHIELD been available to folks?

Kesavan Kanchi Kandadai: Two years actually. So we started very early[00:18:00] when it was still in the realm of possibility, the general generative AI things.

We started really early. And we started working on creating our own appropriate models. And yeah, here we are now two years down the line, still working at it.

William Tincup: Congratulations, you’ve built something absolutely beautiful. And thanks for coming on the podcast. Thank

Kesavan Kanchi Kandadai: you very much.

This has been great. Thank you for chatting and giving me this opportunity.

William Tincup: A hundred percent. And thanks for everyone listening to the Use Case podcast. Until next time.