Is visual context the missing link in AI for CX? (with Zaccari Scalzi of Cobrowse.io)

About the show

Hosted by Nikola Mrkšić, Co-founder and CEO of PolyAI, the Deep Learning with PolyAI podcast is the window into AI for CX leaders. We cut through hype in customer experience, support, and contact center AI — helping decision-makers understand what really matters.

Summary

AI has become remarkably good at answering questions — but customer experience should go beyond answers. It’s about resolution. Too often, AI fails when customers can’t easily explain what they’re seeing on their screen.

In this episode of Deep Learning with PolyAI, Jenn Cunningham sits down with Zaccari Scalzi of Cobrowse.io to explore why visual context is becoming essential for effective customer service AI. Together, they discuss how combining conversational AI with real-time visual grounding helps agents — human and AI — understand complex, multi-step customer journeys.

They cover:

The takeaway is simple: AI works best when it has full context. And for many customer journeys, that context is visual.

Key Takeaways

Transcript

Jenn Cunningham
00:40 – 01:17
Thank you all for joining us, for another episode of deep learning with PolyAI. We’re here to help CX leaders get a window into the latest and greatest developments in AI.
I’m your guest host, Jenn Cunningham, subbing in for Nicola today, bringing you insights from our friends at Cobrowse, to discuss integrated agentic capabilities for customer service. Now, have to encourage you to, like, subscribe, you know, five star rating, on your podcast app of choice.
And now I will be introducing in, Zaccari Scalzi. Thanks so much for joining us, Zach.
How are you?

Zaccari Scalzi
01:17 – 01:19
Doing well, Jenn. Thanks for having me on.

Jenn Cunningham
01:19 – 01:39
Yeah. Excited excited to chat today.
I guess for a bit of background on this episode, I was just creeping on our different project Slack channels. Saw a really exciting demo with Cobrowsee.
io, and I said we have to talk about this on the podcast. So, Zach, would you mind introducing yourself and talking a little bit about, what Cobrowsee.
io is, what you do?

Zaccari Scalzi
01:39 – 02:46
Yeah. Of course.
So I’m director of sales over here at Cobrowsee. io.
We’ve been around for about eight years. The product that you saw today is very different than the legacy product that we’ve gone to market with.
So to give a little bit of context, Cobrowsee. io is collaborative browsing software.
So it’s basically a way for remote employees to see what’s happening on end user devices. Traditionally, we’ve been installed in contact centers for support.
So think about, you know, calling into your bank, your insurance provider, on your mobile app, or on your website. Our product allows those agents to see what’s happening on your screen, annotate, guide you through the resolution, and help troubleshoot with you.
Fast forward eight years. Now our product is focused on enhancing AI with the same level of visibility.
So, basically, taking that same window into the customer’s experience and grounding AI models with that knowledge that it’s not just a lookup bot. Right? We’re not just looking at knowledge base articles, taking that information, and responding back anymore.
What our product allows AI and virtual agents to do is actually understand what the customer is looking at, what error messages they’re facing, and actually helping uniquely guide that client, given the information that that understands, through the resolution.

Jenn Cunningham
02:46 – 03:13
Yeah. Which makes sense.
From our end, when we’re thinking about customer service conversations, we’re always thinking of how can we plug into new knowledge sources and different kind of sources of information in the background. I know everyone is kind of looking for one common brain, right, or knowledge base that they can maintain.
But I think you’re able to take a really interesting approach where, you know, users or agents can actually really see what the users are seeing in real time. So everyone’s aligned and on the same page.

Zaccari Scalzi
03:13 – 03:44
Exactly. Yeah.
Sometimes, you know, simplicity is key, and no matter how much data, small little mistakes, tend to kinda throw a wrench in the gears. So simple things like somebody forgetting to put the right format for an email address.
Right? That might be missed within the troubleshooting of a knowledge base lookup and transmission back to an end user. Simple visibility says, hey.
You’re missing an at symbol or you’re missing a period. Right? It’s a small piece that comes into play, but it makes a makes a really large impact when it comes to accurately troubleshooting and, you know, guiding through through that resolution.

Jenn Cunningham
03:44 – 03:59
Absolutely. So, you know, really, when there’s the gap between what the AI knows and what’s actually happening, that’s just that’s where you have issues.
And so AI consistently performs best when it has the most context possible. And so, really, I think your software is fantastic because you’re able to close that loop.

Zaccari Scalzi
03:59 – 04:17
Yeah. Even looking at human agents and virtual agents, no matter how great the human agent is, if the end user can’t accurately communicate what’s happening on a platform that they’re not really familiar with, that’s where that kind of process breaks down.
Providing that context is definitely really helpful.

Jenn Cunningham
04:17 – 05:06
A 100%. And so with PolyAI, we’re really specialized in customer service AI across phone and digital channels.
Cobrowsee. io are really specialized on having this kind of additional video capability or visual capability to your customer service.
Right? I think for us, we’ve really found because we’re specialized, we’re able to handle more complex use cases, because we know we aren’t just looking at one broad model for everything. We’re looking at what do people actually need to tackle for customer service and what good what does good look like in a customer service context, which is really what’s good or what’s best as opposed to, you know, something more broad based.
how do you feel about specialized solutions?

Zaccari Scalzi
05:06 – 05:54
Yeah. I mean, I think we’re in line with our go to market messaging there.
Complex use cases is where people get more value out of visibility. Right? If somebody is sitting there trying to reset a password, chances are that knowledge base lookup and that transmission back is pretty simplistic.
When you look at kind of one AI or one system to do it all, it’s the same messaging that people have dealt with in the support center. Right? If you have one CCaaS product that does everything, it’s not gonna be the best at anything.
When you have the ability to combine forces with PolyAI and Cobrowsee. io, right, the best in class solutions with what they’ve gone to market to do, that’s where that becomes really a powerful solution.
So I think combining all of that information, all of that functionality underneath one roof with a seamless integration is the best path versus just trying to find one AI that’s good enough to kind of get by.

Jenn Cunningham
05:54 – 06:48
Yeah. Right? Because I think of when I think about the limitations for AI technology so a couple weeks ago, my oven broke.
I called up the customer service number, but it took I was on hold for about forty five minutes. And so during that time, I was on ChatGPT, and I’m like, it’s making this weird noise.
It’s not heating stuff up. I don’t know what’s going on.
And ChatGPT gave me a list of potential issues, but only I knew from then going to the company’s website, seeing what my model was, right, seeing what that additional information is. I was able to say, okay.
I’m pretty sure this is the issue. So then by the time I spoke to the customer service agent after forty five minutes, I was able to say this is what’s going on.
I’m pretty sure this is what you need to send someone out for. Right, you just you need that complexity because I think.
the broad based solutions don’t always work. And also the models change.
Right? Which makes another issue.

Zaccari Scalzi
06:48 – 07:37
I I think as far as, like, the, you know, ChatGPT has obviously gone to market with some really strong solutions. It’s one solution that can do a lot.
But I think, again, when you start combining the needs and requirements of specific companies, it’s not enough to kinda cover the broad strokes. I mean, ChatGPD specifically has come out with some really cool technologies.
They’ve definitely blazed the trail for a lot of different companies, but I think one that comes to mind is their operator system. I think it launched in, like, January earlier this year, which, you know, has the ability to look at these sites and kind of tries to solve the same thing that Cobrowse has inherently done here.
But there’s a difference in styles of technology, which happy to jump into. But I think what it boils down to is broad strokes versus very specific tools for specific problems.

Jenn Cunningham
07:37 – 08:04
Yeah. Absolutely.
Especially for us. So we have our own model, Raven, that, you know, is specifically focused for customer service capabilities.
So when you’re thinking of more LLM wrapper companies, you know, the model changes, your logic may change, your solution may change. Since we have full control over our model and we have visibility into our deployments, we already know, you know, how are we optimizing every time to make sure things work and keep working, in production.

Zaccari Scalzi
08:04 – 08:05
Yeah.

Jenn Cunningham
08:05 – 08:15
But, so from a Cobrowsee perspective, can you walk us through a scenario where, you know, traditional bot guidance just falls apart? And what does that handoff actually look like in reality?

Zaccari Scalzi
08:15 – 09:39
Yeah. Great question.
So I think it’s really kind of where the context of the situation matters. Right? There’s a lot of simple probes where a forgotten password.
Right? You can just pretty accurately say, hey. Hit this button.
You get the email. Click the link.
Three steps to resolution doesn’t really add much complexity to it. Where our product has found a lot of value is not only the multi step, multimodal, you know, start on the website, go to the application.
There’s another domain. Right? There’s a lot of complexity that goes through specific journeys.
But I think as far as the improvement in troubleshooting and providing accurate resolutions becomes a more of a factor when there’s alternative recommendations based on configurations. Right? If this account is set up this way, you have to take this path.
If it’s set up another way, it’s a completely different resolution path. It’s at that point where there’s a deterministic, hey.
This is not the right resolution that you’re giving me, and you’re looking around for a button. Right? The other day, I’m trying to hide a gift I got for my wife from Amazon.
Right? I can know that there’s a way to archive these orders, but I can’t figure out how to do it. And I’m on my mobile device.
Digging through enough knowledge based articles, you go on and you’re saying, oh, you actually have to be on a desktop. Right? The difference right there makes a huge outcome.
The bot was. trying to tell me, like, there’s a way to do this.
I’m like, that button doesn’t exist. So it’s those scenarios, I think, where our product and that insight and that, you know, we’re coining the term, like, visual intelligence is really kind of a big player.

Jenn Cunningham
09:39 – 09:51
Yeah. No.
Absolutely. And when you’re thinking of that visual intelligence, the live context, what is the AI actually seeing? You know? How does it know what’s clickable or, where someone would be stuck?

Zaccari Scalzi
09:51 – 11:11
Great question. So backing up a little bit on the core product that Cobrowsee.
io has gone to market with. We have been with a lot of different Fortune 100 companies.
It’s enterprise grade. We’ve been battle tested.
How our product works is our SDK gets installed within the web application or a mobile native application. We support both channels.
When that SDK is installed, you’re basically predefining what the, you know, human agent or, in this situation, a virtual agent can and cannot see, what domains are visible, what pages are visible. We also have fine grain data redaction when it comes to, you know, billing information, any sort of PII that exists.
And how our product works is it’s not just like a video. Right? If I share my screen on a Zoom or Teams, it’s a visual transmission of what I’m looking at.
There are some solutions out there that throw a black box up on there, and they’re like, hey. Agents can’t see it.
Our product is what’s considered a DOM based approach. So we actually understand the page structure, behind the application.
So it’s not just a visual transmission. What that means is normally we would take that page structure and then rerender it for the agent, whether that’s a human or virtual agent.
With that fine grain redaction, we block data at the end user’s device. So that.
credit card information, that name, that address never gets transmitted through our servers. In the virtual agent, that means there’s very strict guidelines on what it can and cannot access, and all of that’s predefined during setup.

Jenn Cunningham
11:11 – 11:30
That’s fantastic. So when you’re thinking through the governance in different use cases, that’s really it’s tailored to how are you protecting PII.
I’m assuming also how are you handling different government regulations as well beyond just guiding PII, making sure you’re following kind of the latest and greatest, all that good stuff.

Zaccari Scalzi
11:30 – 12:22
Yeah. It’s always the first topic of conversation that comes up with our clients.
So at the first mark, we offer two different provided services. We have a hosted service, and we have a self hosted service.
On the hosted side, we have all of our abbreviations, you know, our SOC two, our ISO, our GDPR, our CCPA, our HIPAA, ECI, the list goes on. A lot of companies within financial services, insurance, health care, telecom really want to own all the data.
In that situation, we also do support self hosting on AWS, GCP, and Azure, which allows them to maintain all data ownership with Xero phone owned back to our servers. So between that and the data redaction that’s stopping the information actually at the end user’s device, It’s just a way of going to market that has more of a data minimization approach versus let’s just open up the fire hose and let the I AI drink as much as they can.

Jenn Cunningham
12:22 – 12:33
Absolutely. Especially because you never know from our side when we’re handling conversations, especially over the phone, people will say anything.
Right? And so you can’t just be prepared for,.

Zaccari Scalzi
12:33 – 12:33
Yeah.

Jenn Cunningham
12:33 – 12:48
oh, you know, these are the given use cases. I think you never know what someone’s gonna say.
You never know what’s gonna be on the other side of that screen. And so to make sure that you’re able to support even the most sensitive of conversations, you have to take that more holistic approach.

Zaccari Scalzi
12:48 – 13:24
Exactly. And so that’s where, you know, we were talking about ChatGPT a little bit with the operator approach.
There are other ways of sharing screens that don’t involve installing, you know, an SDK. But if you don’t do that, the other approach here is proxying traffic.
Right? We’re routing all of the end user’s device through the server. In that scenario, there isn’t the same level of guardrails.
If I go to my, you know, Bank of America account or if I go over somewhere I shouldn’t go, we’re off the rails. Everything within these, whether it’s a human or AI approach, is really on the rails of what has been predefined as okay from the enterprise, like, governance perspective.

Jenn Cunningham
13:24 – 13:34
That’s and that’s such a relief. So if I’m ever calling up customer service and we’re seeing an integrated and I’m working on an integrated, system with Cobrowsee.
io, I know my my data is in good hands.

Zaccari Scalzi
13:34 – 13:55
Yeah.

Jenn Cunningham
13:55 – 14:04
So if we take a step back and then think, okay, conversation is not successful or we can’t resolve the conversation. We need a hand off to an agent.

Zaccari Scalzi
14:04 – 14:05
Of course.

Jenn Cunningham
14:05 – 14:12
Where does the AI to human hand off happen? You know, what makes a good hand off in this sort of scenario?

Zaccari Scalzi
14:12 – 14:45
Great question. And I also wanna maybe reframe it a little bit too because it’s not just that this AI failed.
Right? People. think about containment as, like, this end all.
Hey. Let’s make sure that 95% of these conversations are contained within AI.
I think AI is incredibly powerful, but there’s always gonna be a number of cases or, you know, explanations that need to happen that need that, like, emotional connection to a human. So I think it’s really important to have this handover be as seamless as possible, whether it is, hey.
I wasn’t able to solve my problem, or maybe I just have additional questions that I need answered that, you know, I’m not able to get from this interaction.

Jenn Cunningham
14:45 – 14:45
Yeah.

Zaccari Scalzi
14:45 – 15:11
Within that scenario, I know PolyAI does, you know, great summarization of things that have happened, that plus the context of the screen now. And so they can say, hey.
This person knows on x y z domain. They’re struggling with login.
They’ve tried three times. There’s this error message that’s popped up.
All of that combined with that, so the human agent can pick up the phone. We get the quick, you know, twenty, thirty second download of what’s going on with that context, and they’re able to pick up right where that a o AI has left off.

Jenn Cunningham
15:11 – 15:48
No. Absolutely.
And you raised a good point. I it was on a webinar that we did a year ago, shameless plug, on the importance of resolution.
Right? But it’s not just, okay. We’ve contained the call.
How do we resolve the call? But with that as well, I think it’s how can we also set those agents up for success? Like, being a CSR is a really, really hard job. And so.
the more you can do to make your agents successful and make their job easier, the better you’re gonna see in terms of technology adoption, the happier your customers are gonna be, and just the better experiences you’re gonna be able to provide.

Zaccari Scalzi
15:48 – 15:56
Perfect.

Jenn Cunningham
15:56 – 16:15
You’ve mentioned accessibility a bit. So also how does visual grounding help when someone has, say, an accent or they can’t describe describe what they’re seeing.
We’re quite good with accents, but I think when you’re looking at accent and then you have additional context in terms of a screen or a visual aid, it’s it’s a different type of conversation. Right?

Zaccari Scalzi
16:15 – 16:52
Yeah. I mean, given our interaction, this is completely on the PolyAI side with the accents.
Your system has done incredibly well. We’ve done a lot of testing there.
I think what our product helps with is just the lack of ability to communicate. So it’s not even the accent that understands it.
It’s just somebody that doesn’t know what they’re looking at. They don’t know how to describe what the buttons are, what page they’re on.
Right? Whether that’s an older demographic or someone who’s just unfamiliar with the digital portal, that visual context of what is going on is incredibly helpful. So kudos to the PolyAI team on the accents,.

Jenn Cunningham
16:52 – 17:07
Absolutely. I mean, I think the main theme of this conversation is how can we provide the most context as possible to AI and what is kind of integrating AI solutions to enhance the customer experience using all of the touch points you have available.

Zaccari Scalzi
17:07 – 17:08
Absolutely.

Jenn Cunningham
17:08 – 17:29
So as we’re thinking about these kind of integrated solutions, if anyone’s kind of if their wheels are kind of turning, if they’re thinking about this, I mean, there are so many different applications for customer service AI. Where do you see the best possible, use cases or applications for this kind of joint technology?

Zaccari Scalzi
17:29 – 18:31
Good question. I would say it’s really the more complex use cases that are multistep, multimodal, and there’s optionality throughout the resolution path.
Right? If there’s one predetermined, even if it’s 12 steps, you do step one, two, three, four, and they go right through. Right? But we work with let’s talk about, like, a bank use case.
Right? You all are in the support center. If I call in and I’m trying to apply for a mortgage, based on what I select, it’s gonna change the path that I’m going down.
And so this step one, two, three, four, five, I’m no longer following. I’m over here in left field.
So I think the versatility of resolution path makes for a really fun solve for both Poly and Cobrowsee. io, and that’s really the wow moments.
Like, even our best human agents wouldn’t know if you select option a. There was three options hiding behind that.
But, again, because our product understands the page structure, we know what’s coming next because your product has the ability to, you know, talk through these complex paths. Now we’ve joined those solutions.
It’s like mirroring, like, your best agent possible.

Jenn Cunningham
18:31 – 18:58
Absolutely. That’s that’s our hope.
Right? We want the best agent. We want the best experience.
It’s always the goal. And products change.
Right? We’ve been we’ve rolled out omnichannel earlier this year as a more formal offering. How are you staying aligned from a Cobrowse perspective, to make sure that everything stays functional and everything’s staying consistent as you continue, to just enhance your capabilities.

Zaccari Scalzi
18:58 – 20:05
Yeah. I might take this question in a different direction, because my brain is sparked up.
So changing in products. We’ve talked with a lot of enterprise clients that are trying to understand, like, how can we feed the right information back into our AI.
Right? Cobrowse is obviously the best option, but the other option here is basically installing a number of different APIs within your product to feed that information out of the back end into the AI models that kind of understands what’s happening. As products change, right, you just roll out some new functionality, the UI changes, the UX changes, all of the knowledge base articles need to be updated.
If you took the a API routes, you have to update all those APIs. The beauty of the simplicity of Cobrowsee.
io SDK is every time we start a session, we’re looking at that page structure again. If something has changed, we’re feeding new and relevant information back to that AI model.
And so it’s not just this constant treadmill of trying to keep up with changes and making sure that your AI is aware of it. This SDK actually grounds it in reality of what is in front of it today versus what it is looking at in the knowledge base.
So I think, not exactly probably the path you wanna stick with that question, but pops into my head there.

Jenn Cunningham
20:05 – 20:38
No. I think I think it’s a good it’s a good way to think about it.
For us, we’re kind of as companies are, like, changing their knowledge basis and they’re changing their data, that’s a problem that we’re always trying to solve for how we’re making everything relevant. So it’s great that you’re just starting there.
Right? And it’s just what is right in front of you. So I know we talked a bit about resolution.
When we’re looking at these integrated solutions, what do you really see as success, or how do you define success, for this scenario? You know, what metrics actually matter when you’re measuring if something’s working?

Zaccari Scalzi
20:38 – 22:12
Yeah. You know, traditionally, you know, looking at the Cobrowse product, I think whether we’re talking about an AI, you know, human or just a straight up human, The question isn’t just what metrics, but it’s what business, you know, strategy we’re aligning with.
Traditionally, people were coming into us saying, hey. My call center metrics aren’t good.
I wanna improve them. Okay.
Right? We talk about handle time, resolution rate, satisfaction rates. The conversation has evolved both on the human and AI side, which is not just, can we resolve the problem quickly and the customer’s happy, but it’s, can we prevent that problem from coming back in in the future? So there’s this level of digital kind of adoption, user education that happens throughout these scenarios because our MCP plugged into PolyAI basically enables two different pieces of functionality, visibility, but also the kind of, like, hand holding through resolution.
So we’re highlighting on the screen. We’re guiding them through the resolution.
And so they’re the ones driving. Through that action, they’re teaching themselves how to use that digital portal in a more effective manner.
We have a lot of clients that see reduction in call volumes overall year over year. We see that there’s less of a, you know, resource dependency when it comes to, hey, I need to do this one thing every month.
My bill changed. Why did it change? Oh, I remember last month, you know, when I talked with their AI bot, it actually showed me how to do that.
So there’s. this new kind of long tail evolution of this visual intelligence, which isn’t just the call center metrics, but it’s actually, like, less dependency year over year on that team to, you know, constantly do those same tasks because they’re learning along the way.

Jenn Cunningham
22:12 – 22:22
Absolutely. You’re better able to educate customers.
Right? And so they’re better able to self serve in a new way, beyond just kind of pointing them to the right page on the website.

Zaccari Scalzi
22:22 – 22:25
Correct.

Jenn Cunningham
22:25 – 22:35
So what is one thing that you’ve learned from customers using this that really surprised you as you’re better teaching and training your customers’ customers?

Zaccari Scalzi
22:35 – 23:41
The complexity that exists within this space continues to amaze me. I think what is really interesting is the level of effort required behind the scenes to make these virtual agents accurate, and understand without this style of technology.
We came to market and we thought we had something that was really valuable. We thought it was really cool.
We spun up some great demos. We’ve launched over a few clients.
But it wasn’t until those conversations three to six months after launch that they were like, you have no idea how much time this has saved me. You have no idea how many different tasks that I don’t have to do anymore because of the information that you’re feeding these virtual agents.
Right? Knowledge bases are still important. Those are never gonna go away, but, you know, grounding it with imagery or specific instructions on how to do those different things are no longer responsible for it.
So I think for me, that’s been the biggest shocker is just how much this has actually impacted our clients and not just the end users. I always kinda thought like, oh, the end user would be like, wow.
This is really cool. Like, they’re on my screen.
They could show me how to click. But I really didn’t think about the operational overhead and the inefficiencies that existed without this technology.

Jenn Cunningham
23:41 – 24:47
Absolutely. It it relieves a major burden.
For us, something that I’ve loved to see from our customers, both thinking of kind of integrated capabilities and just stand alone, is being able to adjust for the customer themselves. Like, we do a lot of work with hearing aid companies, so they have customers who are deaf and hard of hearing.
And how are you navigating customer service conversations, especially over the phone in those scenarios? How are you, again, providing context? How are you meeting the customer where they are? And that just it requires additional nuance, that, again, I don’t think you get from a fully general out of the box solution. It needs to be tailored and specialized.
So for our CX leaders listening in, who may be thinking about agentic systems rather than, you know, let’s plug in one solution here or think about one solution here. What is the one thing you think that they really need to understand on how these technologies need to work together?

Zaccari Scalzi
24:47 – 24:58
I think the best thing that I can tell them is that the planning of this all is key. You know? I think when I talk with companies, they try to plan, but not in a directional manner.
Right? Right?

Jenn Cunningham
24:58 – 24:58
Yeah.

Zaccari Scalzi
24:58 – 25:52
They start talking about, like, all these things that they wanna achieve, and then there’s, like, individual takeaways. When you’re in that room of everybody that needs to be a part of the conversation, take a couple of, you know, sample user journeys.
Right? One ticket that’s consistently stumping your AI and say, why is this stumping it? How do we fix it? Where is it getting hung up? Using user journeys to test your, you know, assumptions around how these should work or what should be performing always makes for a much fruit more fruitful and actionable conversation. I think that’s why I see a lot of enterprise companies that have these broad AI strategies that just don’t move very fast.
There. aren’t those actual next steps.
It’s like, oh, let’s take it away and test it. We’ll put it out there.
You’ll see how it works. It’s like, you can do a lot of that internally with the correct prompting.
So, like, find those user journeys, take those key learnings, and make those your next steps, if I can provide you some general advice outside of just, like, Cobrowsee. io AI side.

Jenn Cunningham
25:52 – 26:37
No. Absolutely.
And I think it’s really making sure you have flexible solutions that you can integrate and you can have and create agentic systems. I think both of our solutions are quite easy to integrate with.
So you’re really able to have easy integrations, basically, plug into a current ace AI strategy with, MCP, which is great. Yeah.
Well, Zaccari, thank you so much. This has been so fun.
I know we have a fun surprise for our listeners, where we’re actually gonna show a demo, that we’ll kinda be plugging in, so you can get a sense of what these joint capabilities can look like and maybe spark some ideas for how this might be applicable for your own unique use cases and businesses.

Zaccari Scalzi
26:37 – 27:37
Yeah. I think it’s, you know, the proof is in the pudding.
I think a lot of these conversations in general can be a little bit of, like, a gray area until you see it actually in action. So, kind of to set the stage for what this demo is gonna highlight is what a interaction looks like when somebody engages with PolyAI.
Right? They say, hey. I understand what your problem is.
There’s gonna be a certain point within this workflow that the AI says, actually, it’d be much easier if I can understand what we’re looking at to help you guide you. At that point, what’ll happen is the AI will request screen share with, you know, variable consent prompts.
Once that’s allowed is when that SDK gets involved. So it’s not like every single interaction, our SDK is being fired up, and the AI just sees everything that’s happening.
It is always consent driven. So once that consent is given, the AI starts looking at what page they’re on, what error messages they’re feeling.
And that’s where the back and forth between PolyAI and our MCP start kind of doing their magic.

Jenn Cunningham
27:37 – 28:19
Perfect. Awesome.
Let’s jump in. Awesome demo.
Love it. Love it.
Love it. So similar to what I saw in Slack, I’m excited to now share kind of with the wider audience.
Zaccari, thank you so much for joining us. Us.
If people are curious about kind of leveraging these layered technologies, how what’s the best way to kinda move forward? How should they be thinking about it?

Zaccari Scalzi
28:19 – 28:42
Yeah. Of course.
You can go to Cobrowsee. io.
We have a brand new AI page that is sparkling new. If you would like to, we you can also request a custom demo where we can actually install our, you know, MCP within your virtual agents to start testing that workflow within your digital portals.
Or if you want, send a quick email to hello@cobrails. io, and they’ll go out to my entire sales team.

Jenn Cunningham
28:42 – 29:05
Awesome. No.
That’s fantastic. And, obviously, you can always reach us through PolyAI as well, just poly dot ai.
Request a demo or reach out to our team, and we’ll make sure, so long as you mentioned you’re interested in understanding capabilities with Cobrowse, we’ll make sure to kind of have a joint conversation there. Awesome.
Well, thank you so much. I hope you have a great rest of your day, and looking forward to chatting soon.

Zaccari Scalzi
29:05 – 29:07
Of course. Thanks for having me on, Jenn.