Inside Avalara’s Agentic AI Tax Compliance Platform

Hey everyone, welcome back here to Techstrong TV. I am happy to have this gentleman back with us. His name is Hugo Sarazin.

Hugo is the CEO at Avalara, and I had the pleasure of interviewing Hugo, was it about two weeks ago now? Three weeks ago. That's right.

Yep. Hugo recently has become the CEO of Avalara. You could go watch the first video if you're interested in his background.

It's fascinating. PhD in some advanced technologies and work down here, for what was Ultimate Software before they merged. But now at Avalara, and a lot of people are saying, "Avalara?

" Well, yeah, they've built a hell of a business in tax compliance and reporting and so forth. Hugo, I read somewhere that you guys actually clear something like $2 trillion dollars or handle the tax work on $2 trillion dollars plus of transactions. Is that true?

That is true. We have an amazing aspiration to be part of every transaction in the world, and we do sales and use tax. And we have a calc engine that is built for that.

So whenever you go into e-commerce transaction, when you get to your shopping cart and somebody needs to say in real time what taxes you're going to need to pay when you deliver a good in a state, a city, a county, we do that in real time. So you can imagine that on Black Friday- Right ... yeah, we're behind every one of those transaction.

There's more going through our network than many of the payment transaction because we help every payment transaction do their sales tax calculation. So a lot of the commerce of the world is being powered by Avalara. We're behind the scene.

Really interesting, right? A behind-the-scenes sort of powerhouse there. Now, excuse me, I know a lot of our audience out there, Hugo, is saying, "Great, Alan, but where's the cutting-edge technology here?

This sounds like a CPA convention or something. " So there's the- There is a tech conversation, right? Right.

Talk to us. You guys are, like any good company in your place should be doing, staying at the very edge of what's possible and adopting agentic AI and things like this. Tell me a little bit about Avalara and AI.

Yeah. So before I get into the AI piece, let me just set the stage. We were birthed in the cloud era.

So we're cloud-first. And we're one of the very few players out there that is multi-cloud. So we work on the Google Cloud, we work on Amazon Cloud, we work on Microsoft, we work on Oracle, and then we work in a way that you can move the work from one of the cloud at any point in time.

So if Google goes down, we're up. So we have super incredible resilience. Incredible resilience.

And then we deliver these sales tax calculation with incredibly low latency. You don't want to be on a Shopify and be waiting for three seconds to get your shopping cart. 9999%.

And then we do it with low latency. That's today. That has been going on, that's our claim for fame, that's why people come to us.

We've also built now to migrate, to shift to the agentic conversation. We've built a lot of agentic capability. We keep the core deterministic because you need the same answer no matter which day of the week, no matter what transaction.

So our calc engine is a deterministic calc engine, and you can rely on it delivering the right answer for you. The same thing when we create your tax return because we have all this sales tax information, we can calculate your returns. That calculation is deterministic.

That doesn't change. What we're doing is we're putting AI around everything so that we can monitor in real time if there are things that are changing in the business. So if you are missing tax code or there's a thing in our industry called Nexus.

So when you do e-commerce transaction and you do reach a certain threshold in a state, like Georgia or North Dakota, which was the Wayfair state. When you reach a certain threshold, now you need to file tax. We monitor that for you in real time.

We use AI to, when you submit exemption certificate, to look at the exemption certificate, take it apart using optical character recognition and natural language processing to identify what's in the exemption certificate. " So there's a lot of things going on in the background. The same thing for cross-border.

I could go on and on. So this business is being reimagined now using AI everywhere. Like so many others are and will, right?

I think this is a great case study. " And I think you just laid out the case better than anyone could, Hugo. So thank you for that.

I want to talk about Crush a little bit, though. Yeah. Right?

We're recording this, by the time people see it, we'll be maybe a week, two weeks at most out from the event. It's down here in Fort Lauderdale at the Omni, new Omni Hotel down there. Is there still time, still tickets available for people who want to- Yeah ...

attend? Yeah, there's absolutely tickets available. I would encourage anybody who's interested in understanding where this space is going to come down, see us.

We want to share with you some of the latest thinking. There's some great content, great program. We're going to also introduce a number of really breaking, leading technologies for the future, so you can really learn about this.

And then you can meet your peers. There's going to be a lot of time where you can network with partners, with other people involved in sales and use tax, and you can compare notes, learn from each other, and get ready for, again, 2027. It's going to be a spectacular year.

There's so much change happening, and the pace of change is not slowing down. It's only accelerating. Number of things that we need to keep track of, the speed at which the regulation is changing in this country and around the world is only accelerating, and you want to be ready, and we're going to talk about all that.

Absolutely. Hugo, there are some people who are saying, well you said it best, right? Taxes is rather deterministic.

" And this is dollars and cents we're playing with here. This is real money. And finance being as tightly regulated and everything else.

Isn't there a pushback on using the AI? Well, I think there's a few things. One, at the core, we need to be able to be 100% clear that a tax calculation will be the same no matter which day of the week And that's why our deterministic core is not changing, and that's why you can't go to one of these LLM and ask it to calculate things for you.

You can ask it a question, but that LLM is trained on public data. Now what we have, which is really super interesting, is we have 22 years of proprietary data. We've seen transaction flow.

We've seen calculations being done. We've seen returns being submitted. We've seen what came back, which one created audit, which one did not.

So now all of a sudden we have the ability to do an end-to-end closed feedback loop on this process. And now we can train our agents to be smarter and better. We still do the calculation in a deterministic way, but we can anticipate problem in a way that is a bit more informed by all this knowledge and all this data, this proprietary data which others don't have.

That's one example. The other one I mentioned, monitoring Nexus. Again, we can do that using an agent that does it for you in the background.

We can also assemble the return from all these different sources, get the data in one place, bring it here, normalize it, clean it up, bring it from another place, normalize it, clean it up, assemble a return, then surface it and say, "Tax professional, this work has been done for you. Now we need you. You're in the loop.

You make the final call. " We've surfaced and we've identified a few areas we think could be at risk. There's some judgment calls.

But we've eliminated all the grunt work that people have to do. So I think it allows people to focus on what is the highest value activities, and it's going to fundamentally change the work of tax professional and even accountants. Absolutely.

Here's something, though. The programs you work with, Black Friday, and I don't know if Shopify's a customer, but those kinds of programs- Mm-hmm ... are increasingly themselves using AI, right?

So we find ourselves in a world where AI to AI communication, no human in the loop, if you will, it will become normal more and more as time goes on, right? Yep. Are you guys using, and I don't want to get too geeky, but are you using MCP type- Oh, yeah ...

of protocols for this kind of thing? We've been using MCP for a while now. We've created a fully agentic core platform.

It uses A2A MCP, and it's an incredibly powerful way for us to do a lot of the work. We've also created purely APIs. We can use AI to interact with third parties.

We can have agents to agent conversation. You're exactly right. Every ERP now has their own agent, and we need to be able to interact.

We can also live in other surfaces, which is kind of an interesting and wild thing. So in the good old days, what we did is only we created a way for us through a connector to be part of an ERP tax calculation. But now we could also be part of a Slack conversation.

We could grab an invoice in an email. And we can live in other surfaces because we're using these standard protocols and providing the right information in the right context. So it's a very different experience, and we're also getting ready for a world where agents are not only going to be doing commerce with each other, but we're going to need a tax framework for these agents, too.

Right. It doesn't exist right now, but that's going to be a part of it, and we're obviously trying to be ready for that world. That kind of segued into my next question, which is what is next?

So here's one thing that's next for sure, right? That is one. I do think the other one, here's an interesting thing.

Let me contrast two things. " It will do its very best based on reading the internet to come back with an answer to that. And you could decide that that is good enough to put a tax number on a transaction you're doing.

But if you do get an audit, where's your audit trail? What's the log? What connects all the decisions you've made since the beginning of time?

So now what we can do is the ability with an agentic framework is to document your entire return with all the decisions and the law and the reference and everything we've done in a way that is incredibly hard to do today. Now we've completely created a new way of doing that, and that's pretty interesting. It is interesting.

Yugo, I'm looking back. I don't know if I gave the dates for Crush 2026. It's September 22nd and 24th.

Absolutely. When are you delivering your keynote? On the first day.

I'm going to have the opportunity. It's my first keynote here at Avalara, and I can't wait to share some of our perspective on where the industry's going, what we think is exciting, where we hope to play in that, and most importantly, spend time with customer. Get to meet more of them, spend time with our partners, and learn from them as well.

And then just be part of a dialogue so that we can continue to push the boundary of what is right, use AI with the right governance, the right understanding, the right data, to make everybody's job better, and with trusted outcomes. Excellent. Yugo, we're about out of time.

I want to thank you for coming on here. The next time I see you, well, I'll be able to meet you in person at Crush. We'll be doing videos there.

Absolutely. So we're excited for our coverage. Thanks for coming on here and keeping us posted.

It was a busy holiday weekend, and everyone was talking about the new Astra release from Chat and just how fast this AI stuff is. Are we at AGI? Are we not at AGI?

What you're doing at Avalara is a great example of this isn't just affecting techies who are doing the code development or stuff like that. This is going to have repercussions waffling all through our economy. And it'll be interesting, and Avalara is a great example of it.

Thank you. Thank you. Thank you.

Looking forward to seeing you. All righty. Yugo Sarrazin, CEO, Avalara, here on Techstrong TV.

We're going to take a break. We got more to bring to you today.