Breaking Barriers: How accessibility and agentic AI help cities serve every resident

Thank you all for joining us today at our Breaking Barriers webinar. We will kick off with introductions now.

Hello, everyone. My name is Stacey Kyler, and I am Managing Director at Zammo. And we are so excited that you all are here today to learn about this partnership between ourselves and A11y. And we're just going to do a quick round of introductions, and then we're going to jump right into it. So, Gina.

Hey, everybody. I'm Gina Doll. I'm a Technical Specialist with Zammo and created a fun demo that I'll be showing you.

Thank you, guys. And hello, everyone, as well. Thank you for being here. I'm Dariana Lago, co-founder of A11ySolutions. We are an IAAP-certified digital accessibility team. And just for the ones that don't know yet, IAAP stands for International Association of Accessibility Professionals. And here with me is also Jose Rodriguez, who leads the engineering behind our tools and will be running our live demo a bit later.

So, for years, accessibility and AI lived in different worlds. Accessibility sat with compliance and ADA coordinators. AI sat with innovation and IT, right? So, different budgets, different vendors. That split does not hold anymore. If a resident, or the AI assistant they are already using, cannot finish a permit, a payment, or an appointment, it does not matter which team owns the project. These are now two halves of one solution.

And on our side, A11ySolutions builds that foundation. We reinvented accessibility expertise. We audit and remediate websites, mobile apps, and documents, including PDFs. And we use our own AI-powered tools to turn accessibility backlogs that have stalled for months into work that actually moves forward.

So, the goal is not a report that sits in a drawer. It is a digital estate that is conformant, defensible, and usable for a person using a screen reader and for an AI agent trying to complete the same task. Now, that foundation is not just compliance, it is the ground everything else stands on. And now, Stacey is going to show you what that looks like when the experience sits on top of it.

Thank you very much. Yeah, so, really good introduction about what A11y does. And they really are the folks that handle your website, right? Making that fully accessible. Whereas Zammo is the agentic solution that sits on top of that. So, the chatbot that you see, or even for us, the IVR that you call into in that experience.

And what Zammo Agents offers is, we looked at the product from years and years of doing conversational AI and all of the building that goes into flow building. And we realized very quickly that two things are very true when it comes to cities and counties and states across North America as well as Europe. One is that the IT team is backlogged and they don't have the capacity to build projects for every different department across their organization. Number two is that the citizens expect the same service that they receive when they order food, or when they contact their bank, that they would get from their city.

And so, our goal was to take that knowledge and turn Zammo into an agentic solution that allows the business user to build those experiences, because they have the knowledge, while IT maintains that oversight and that technical oversight. So, that's what we're going to be showing you today. And I hope that you can see now how the two of us paired together are the full package deal. So, with that, we can go ahead and move on to the next slide, which Dariana is going to take for us.

Thank you, Stacey. Okay, so why is all of this landing right now, as Stacey was mentioning before? And these three numbers here help explain it.

So the way people start looking for public services is changing. Your website is still the destination, but it's no longer always the first stop. McKinsey, one of the largest global research and consulting firms, finds that in consumer search and discovery, 44% of people now prefer AI-generated insight over traditional search. Pew Research, a nonpartisan US fact tank, finds that about 4 in 10 American adults already use chatbots to look up information.

And at the same time, the web is still not ready for that door. Each year, WebAIM, which is a respected US nonprofit that tests digital accessibility, reviews the homepage of the 1 million most visited websites. And that study is called the WebAIM Million. And in 2026, it found detectable WCAG failures on almost 96% of those pages. Now, when a page is not accessible, people get stopped and AI agents get stopped too. Both need the same thing: structure, labels, headings, and controls they can actually use.

Now, this matters especially for the public sector. Not because teams have done the worst job, not at all. In many automated rankings, .gov homepages actually outperformed the commercial average. But the real issue is harder. You operate the most complex inventories: years of PDFs, third-party portals, applications, legacy content. And now you have a clear federal standard and a date. Because under ADA Title II, most large cities, counties, and school districts must meet WCAG 2.1 Level AA by April 26, 2027.

So the gap is the risk: residents and AI tools arriving first, while part of the site still cannot be completed. But at the same time, it's also the opportunity, because the same work you do for Title II is the same work that lets a person and an AI agent finish the task instead of dropping off. So, now, to see why this matters so much, in the next slide I will show you what actually happens when an AI agent visits your website.

So, when an AI agent visits your website, it doesn't look at it the way you and I do. We see the colors, the buttons, the images. The AI doesn't. It reads the structure underneath the page. So, the labels and the code that says, for example, this is a button, this is a form, this is a menu. Think of it like this: we see a picture, but the AI reads the blueprint behind it.

Now, agents do this in a few different ways, and this is the important part. The most reliable ones, and the ones the whole industry is moving toward, or at least trying to, read that underlying structure directly. The other approach just looks at a screenshot, and that one is slower, it's more expensive, and it's less reliable.

And here's why that matters so much for you. That structure the best AI relies on is the same accessible structure that makes your site work for someone using a screen reader or a keyboard only. So, accessibility and being AI-ready aren't two projects. They are the same job.

So, the real question becomes: how well does that structure actually work? And that's exactly why we measure with our AI Navigation Score. It's a simple zero to 100 score of how much of your site an AI agent can read and operate when a resident asks it to get something done — renew a permit, submit a form, find a service.

Notice what it measures: not whether the agent will finish the task, but whether it can. Because when it cannot, that request doesn't disappear. It comes back to you as a phone call, as a counter visit, or as a resident who just gives up. So, the score also shows you the exact step where the journey breaks and what it takes to fix it. And because the audit behind it is a full accessibility audit, everything you fix for that agent also helps your human users. So: same work, two returns, and it puts you on the path to the DOJ deadline in April 2027.

Now, I told you the score measures whether an AI agent can complete a task, not whether it will, right? So, let me show you in this next slide why that gap matters so much. Researchers at UC Berkeley and the University of Michigan tested this year. When an AI agent works on a site with a healthy, well-structured foundation, it completes the task about 78% of the time. Take that exact same agent and put it on a site where the structure is broken, and it drops to 42% — nearly cut in half, just from the structure underneath.

And the things that cause it are never complicated. It's almost everyday stuff: controls with no labels, missing headings, and page interactions you can only click with a mouse. So the agent, or the person using a keyboard for example, gets stopped. Now, remember what I said a moment ago, that when the agent can't finish a task the request doesn't disappear? So that 42% becomes a phone call to your office, a visit to your counter, and so on.

And this isn't just a university study to us. This is exactly what we find in accessibility audits every single week. It's the pattern our A11yDetector flags first on almost every site we look at. Which means it's also completely fixable, which is really good news.

So, let me show you exactly how we fix this. And this is the part that I'm most proud of, because it's built for the reality of a government team: a small staff, a huge website, and a deadline. So, we do this in three steps. Detect, Resolve, Ready.

First, Detect. This is where our A11yDetector comes in. It's a Chrome extension. Your team just installs it in the browser and it scans any page in just a few minutes. But I want to be straight with you. Automated tools, ours included, can only catch around half to 60% of accessibility issues. No scanner on the market catches everything, and anyone who tells you otherwise isn't being honest.

So here's what actually makes us different, and this is the most important thing on this slide. The A11yDetector is just our starting point. The real coverage comes from our expert manual audit. Our IAAP-certified specialists go through your site by hand, the way your residents actually do — using screen readers, keyboard-only navigation, using zoom, et cetera. That's the only way to catch that other 40%, the things a machine simply can't see. So, between the automated scan and the human audit, nothing slips through.

Second, Resolve. And here's where our A11yResolver comes in. Finding problems is the easy part. The hard part is fixing them, and that's where government teams get stuck, because the backlog grows faster than the staff can clear it. So the Resolver lives right inside your developer's IDE, the environment where they already write code. So the fix happens where the work already happens. It inspects the page, it traces the issue back to the source code, and hands your developer the AI-guided fix right there. So, no context switching, no separate tool to learn. So the backlog actually moves instead of piling up until the next audit.

And finally, Ready. The result is a clean, well-structured site on a clear path to WCAG 2.1 Level AA, exactly the standard the DOJ deadline requires. So, that's the process: catch it, fix it, and keep it clean. And now Jose will show you this live.

Hello, all. Let me show you a short demo. Do you see my screen? Okay, perfect.

Let's take a quick look at how we turn an accessibility finding into a verified fix, and build a stronger foundation for AI agents. First, A11yDetector scans the page for real accessibility issues. Once the scan is complete, we can see the findings and the evidence behind them. Here, we can drill into a specific issue. In this case, a visual heading isn't exposed with the correct semantic structure.

Now the finding becomes the fix. We pass that finding into A11yResolver, where an AI agent can inspect the live page, trace the issue back to the source code, and determine the appropriate remediation. It then applies the change directly in the code, and we can see exactly what was modified. The build passes, so we're ready to verify the result.

But we don't stop at the fix. We run the accessibility check again to verify that the targeted issue has been resolved. Found. Fixed. Verified. A stronger accessibility foundation for people and AI agents.

Thank you, Jose. Okay, so that's one issue, start to finish. But the question I always get from any government team, or any of our clients, is: okay, that's great for one page, but can you really do this across our entire site, thousands of pages and documents, on a specific deadline? And the answer is yes. And here's the proof.

We work with a client, anonymized here of course, that had exactly that shape: dozens of sites and apps, thousands of pages, and a deadline they had missed before. And we delivered the full review on time. And A11yDetector and A11yResolver kept the backlog moving, no matter the volume. Just for you to have an idea: in our projects, A11yResolver has cut up to 70% of the repetitive remediation work.

Now, this was a hospitality client, but look at the bottom: many properties, many owners, one clock. That's the exact same shape as a city or a county — lots of departments, lots of sites, one deadline. We did it there, and we can do it for you as well.

So that's the foundation: accessible, defensible, and AI-ready. And now that foundation is what makes a powerful AI experience possible on top of it. And for that, I'll hand it over to Stacey.

Thank you so much. And just to give a real shout-out to Dariana and her team: we at Zammo use their team for all of our accessibility. So when we want to make sure that our chatbot code snippet that you are creating at your city, county, state is accessible from the moment it goes on your website, we have used them to make sure that all of that is pre-built and part of the product. So I just want to give them a shout-out and tell them thank you for all of their help, because they truly are very, very amazing to work with.

So let's now take a look at what is Zammo, and how does it help your city, county, state? So truly it is an agentic AI experience that is built for public-sector scale. And four key things that I want to call out here.

Number one is, of course, the no-code agentic creation. So any user is able to use natural language to create bots that not only answer informational questions but also answer transactional. And if you really get down to it, anybody that comes to us and says, oh, we just want an informational bot — at the end of the day, they always want something transactional, right? Because when a citizen comes, they want to solve an issue. They want to find out when they can submit for a permit, when the trash is going to be picked up, what is okay to put in the recycling bin, and the list goes on and on. So with Zammo, the permitting office can create their bot for their agency, and then the recycling office can do their own, while IT maintains that oversight over all of them. So it's built for scale using natural language.

Next would be that true omnichannel experience. So when you build a bot, let's say it's for permitting, and you want to take that bot and publish it to your website chatbot — or maybe you want to publish it to IVR for the residents who want to call in — you only have to build that experience once, and then you can publish it to all of the different channels that Zammo offers, which I'm sure Gina will be showing you as part of her demo.

Next would be the security and data sovereignty. So there's two different ways that you can purchase Zammo. One is our SaaS solution, which is all on Zammo's Azure tenant. And the other way is through our managed application, which is available on the Microsoft Marketplace. So that is installed on the customer's Azure tenant, meaning that all data stays on your tenant — which 99% of the cities, counties and states that we work with want, because of that data sovereignty piece.

And then finally would be the global language support. So out of the box, you create your agent and 140-plus different languages are supported. So if you build it in English and a customer comes and wants to speak in French or in Spanish or any of those 140 languages, they can interact just like someone who speaks English. So making that accessible, not only for accessibility but also for languages, is a very important part of what it is that we do.

And with that, again, everything that we have is built on that foundation of Microsoft Azure infrastructure, which is so very important because it is the cloud that most of our customers already trust. And so all of the security that comes with Microsoft Azure comes with Zammo out of the box. So, enough of hearing me talk about it. The best part about Zammo is getting to actually see it. So I will turn it over to Gina.

Thank you, Stacey. I'm going to show how Zammo removes the complexity from AI adoption, so teams can build AI agents that actually evolve with their organization without heavy engineering dependency or months of effort.

So first, let me show you how easy it is to create an AI agent. Here in the Zammo product is where I told it to create a 311 support agent that answers questions using this retrieval-augmented generation project, which contains the location of the content and the guardrails to prevent hallucinations. Then I just told it to support multiple languages, and I chose Spanish. Then I told it to allow users to create tickets in ServiceNow if they want to report an issue. And then I said, you know, I don't know how to set this up, all I have from my IT team are these credentials — which aren't sensitive, by the way, these are just a demo instance. And lastly, I told it to escalate to a live agent if the user is frustrated, and Microsoft Teams live agent was automatically detected and configured and set that up.

So in minutes, an AI agent was created and it's ready to go live. What you'll notice here is there's no heavy technical lift. Teams can build and deploy agents quickly, which means that ideas turn into real solutions faster, and more people across the organization can actually participate and add value.

Also in here, let me show you the different channels, as Stacey mentioned. Zammo makes it easy to meet customers where they already are across channels, including voice, without needing separate tools or teams for each one. And for the web chatbot, the chat widget is fully accessible and can incorporate your logo, color scheme and more.

Now I'm going to show you the agent in action with the web chat channel. And I will put this on the right side of the screen; on the left, just for reference, these are the public web pages that we told the agent to use to answer questions. So I'm going to ask a very common 311 type of question. I'm using a demo environment, so you might see some latency in the OpenAI responses due to heavy interest from other Microsoft customers. But for your production use case, Microsoft offers dedicated OpenAI resources that deliver immediate response times.

So here is where in the sources the agent is going to get that information — the permits can be gotten online or by mail, some documentation is needed, et cetera. So over here on the right, here's that response. The AI agent used retrieval-augmented generation to provide an accurate and really easy to understand response, along with a clickable link to the source of the response.

Okay, I'll go back to the chatbot. Now, a question like this, as we instructed it in the design, is to have it trigger ServiceNow. And it just does a confirmation, and I'll say, yep, that looks good. Now, on the left, I'll pull up ServiceNow, and the agent is creating that ticket. The ticket ends in 083. On the left, I'll refresh ServiceNow, so you can see that new ticket being created with the description that I gave it. And this integration was made just by using my words.

So I can be very clear and say, I want to speak to a human, or I could express frustration, and it's going to connect me to a live agent. So now on the left is Microsoft Teams live agent, and I'll act as the agent and accept the chat. The transcript of the conversation is available, so the agent has context before engaging with the user. And you can see the agent's responses come into the chatbot — they have a back and forth conversation and the user is very satisfied.

And lastly, I want to show you the multilingual capability. I'm just going to paste in a question in Spanish: what items can be placed in the green waste cart? And let me pull up that page so you can see where this information is coming from. So here's the green cart, and this is the source of the material. So here's where you can see language stop being a barrier. Teams can support customers globally in their own language without adding complexity behind the scenes. And there's that response in Spanish.

With Zammo, you can confidently build and evolve AI-powered experiences without long timelines, heavy engineering dependency, or risk. Thank you. Back to you, Dariana.

Thank you, Gina. That was amazing. Okay, so as Stacey and Gina have just showed us, let's bring it all together. What you've seen today really comes down to two pieces that work best when they are combined.

On our side, A11ySolutions is the foundation. And I want to be clear about what that means. Being accessible isn't optional and it isn't partial. Your website, your applications, your documents, your PDFs — the PDFs residents actually have to fill out. All of it has to be accessible for every person, and now for every AI agent. And that's the work we do at A11ySolutions: we audit it, we remediate it, and we make sure it holds up — conformant, defensible and genuinely usable, faster and at scale.

Zammo is the intelligent layer that turns that foundation into real services, as you just saw a few minutes ago: secure, multilingual agentic AI, already helping cities answer residents across web, voice and chat.

So here's why they belong together. You can have a powerful AI, but if the foundation underneath it isn't accessible, it stalls. And you can have a fully accessible site, but without the AI layer on top, you're not giving residents that always-on, any-language service they now expect. Put the two together, and that's when a resident gets a real answer at 2am in their own language and it just works.

So that's the whole idea. One of us makes you inclusive and AI-ready, and the other turns that readiness into actual service for the people you serve. So if any of this is resonating with you, Stacey will show you the easiest, no-risk way to get started with either of us.

Fantastic. And just like Dariana said, we definitely see so much value in our two companies joining together, but we also don't know where you are in your journey. You might just be starting, and you might need to start with the accessibility piece before you're ready to add in that AI layer on top. Or you might have a fully accessible site already and you are here because you are interested in the agentic solution.

So Audrey is going to be posting in the chat window links to meet with either of our teams. And if you want to meet with both of us, that's obviously more than okay as well. So you can use those links. And then there also will be a follow-up email that will include this deck, where you can use the links on this page as well. So either way, it gives you both ways that you can reach out to us, and we'd be happy to do a custom demo for you using your data, your website, et cetera.

So with that, I know we have two minutes to spare. And if there's any questions that anyone wants to put in the chat, you are certainly more than welcome to, and we can take a look at those and answer. I know that can sometimes be a little bit awkward, so if you'd rather just meet individually, we also very much understand that as well. So we will stay on for the remaining two minutes. But if you have no questions, then that is what we had to share with you today. Thank you for joining us.