The AI Outbound System: Your Judgment, Agent Execution

Julia Nimchinski:

With that, we transition to our next demo. Welcome to the show, Mike Wang. Founder, thank you, and CTO of Avina, Mike? Welcome to the show, yeah, the AI Outbound System. Super excited for this.

Mike Wang:

Awesome, me too. Thanks for having me.

Julia Nimchinski:

Phenomenal. Let’s jump into it.

Mike Wang:

Alright, I’ll share my screen. Let me know if you can see that.

Julia Nimchinski:

Awesome, yep.

Mike Wang:

Cool. Okay, so I know I’m filling in here for a different demo that was slotted for this time, but I hope this is all interesting to everyone, and I’m really excited to be here, so… Yeah, without further ado, so I’m Mike, I’m one of the co-founders of Avena. We’re an Agentic go-to-market automation platform. And just a quick intro on myself. So… Before, Avena, I previously co-founded a company called Bowtie. It was an AI receptionist that, we ended up selling to MindBody, and through that, we scaled our AI receptionist into thousands of SMBs. So. really got to see, kind of, how to deploy custom AI software into many, many different use cases and deploy that at scale.

So now we’re working on Avena and taking all of those learnings of deploying AI into those companies and applying it to go-to-market And we also grew that business primarily on the back of cold outbound. So emails and cold calls, primarily. And, now Avena, we built this to help other companies also scale their intelligent cold outbound, the same way that we did. So, before I get into what Avina does, I think it’s helpful to kind of frame where we fit into the process. So we’ve, again, we primarily focus on helping, with cold outbound. And, that’s emails, LinkedIns.

And also just getting signals data for your top-of-the-funnel pipeline. And so we’re focusing on what the SDR… is doing in their role, and how we can bring automation and intelligence into that process. So, I like to think of this as 7 steps. There’s targeting, you know, which companies are you going after, timing. Is there a signal that indicates they have a pain point? Persona, who you’re reaching out to, research? What do you know about that company that can inform your pitch? Pitch actually crafting the message that’s gonna resonate with them. The delivery, domain reputation, deliverability, not landing in spam, all of that good stuff, and then handling the replies and converting those into meetings.

Okay, so… that’s kind of what your best SDR does. Now, this is what we call the AI-enabled SDR, and don’t worry, we’re not another one of those AI SDR companies that’s trying to kind of replace the person with an AI system that doesn’t work. We all… you know, for anyone who’s followed this space closely, you may have seen that those systems don’t really work well, because the AI just isn’t there yet. It’s getting a lot better, but it’s not there yet. So this is how we think about Avena. Avena is a go-to-market automation platform that really, integrates with your process and helps you execute each of these steps, where you still make the decision, you’re still adding your judgment, and bringing the art of whatever you’re selling into the system, and the AI is helping you execute.

Mike Wang:

Okay, so switch over to the demo here. So, AVENA’s really designed to, again, help you with all seven of those steps, starting with targeting, timing, and persona, who you’re reaching out to. And the way we do this is through what we call AI prospecting. And essentially, this is using AI agents to take any search query and turn that into a list of companies. So I’ll give you… a quick preview of what that looks like. We’re just gonna try this search, because I think it’s a cool one. But, this is something that we’re actually using for our own sales, so these are companies that we’re looking for ourselves.

Companies that sell to dental practices, raise the seed round, and are hiring for SDRs. So we’ve got, kind of, three components here. It’s a very specific segment of companies. That would be pretty hard to find in a typical prospecting tool. There… and there are two, signals here, so they just raised a seed round. That’s also a qualification criteria, and they’re hiring for SDRs right now, so that’s a timing-based signal. And you can search for literally anything here, as long as it’s something that AI can find on the web. So imagine you hired an SDR, and they’re, you know, going and spending their whole day searching the web for these companies.

So the AI took that query, it structured it into these more detailed search criteria, and we can go ahead and preview the search results. And what this is going to do is kick off a search in real time to find those matching companies. While this is running, I can quickly talk about some of the other signal types we have here. So this is our AI signal type. Again, you can search for anything. We also track web visitors, we de-anonymize web visitors to your website, and we have dedicated signals and prospecting types for new hires, job listings, tech stack, and social posts.

And you can think of us as a unified data set for all of those signal types. There’s dozens and dozens of signal providers out there for each of these, but we integrate with all of those for you, waterfall them, and bubble up the best data. So… On the right here, we can see the preview results coming in, and it cites all the evidence here, so, you know. Take the first one, for example, Overjet provides dental AI solutions. They are hiring an SDR, and… Raised seed funding. So all of those qualifications match. Okay, so another thing that you want to do here is define who you want to reach out to.

So we found these companies, and we want to actually find the right folks to reach out to. So in our case. Again, I’m showing you our personal account that we use for our own outbound. We want to target the executive, so that could be the founder or CEO. The, go-to-market were sales leadership, rev ops, marketing, and any founding sales members. So… Every lead that comes through, we’re gonna get all of that… all of the contact details for all 5 of those personas, as long as one of them is found. And I’ll show you what that looks like in a sec, but you basically turn on this signals audience that we’ve created, and then you get new leads every day.

And the searches run every day, so it’s fully refreshed data, you don’t have to worry about the data being stale. And this just becomes a dynamic growing audience, that will continue funneling leads to you over time. You just create it once, and it just keeps running. So… I’ve actually run this for a few days now, and we can take a look at the actual lead data that’s come through. So, switching to the Leads tab here, we can see that a bunch of companies, have been found, and for each company, there’s multiple personas found as well.

So, for example, we’ve got, you know, Joss. Joss here, the founder, of Prakti. So on and so forth. We’ve got all the emails, we can search for phone numbers as well, we have their LinkedIns, and then each of these companies is also scored against our ICP criteria. So… This allows you to qualify the companies as well. So this is another piece of that. Targeting. Piece there is actually making sure that these companies are all qualified. And these are all A’s, because… They’re seed-funded, that’s one of our qualifiers. See if we can find any unqualified companies.

No, they’re all pretty good. So yeah, we have a clean lead list here, and just to show you a different example. To show you, kind of. the breadth of, you know, the data we can find here. I’ll just pull up one example now, we can go over later… more later if we have time. But, in this search, we’ve actually just searched for a very specific segment of shipping and logistics companies. So, you know, this is an AI prompt, so you can kind of write anything in here. Company sells products or services for freight procurement, transportation, or the company has language on their website targeting shippers, you know, or the company lists integrations with any of these highly specific integration platforms.

And the AI was able to find a lot of companies that matched that criteria. So just to showcase the power of that.

Mike Wang:

Okay, moving on to the next step in the outbound process. After finding these leads, we wanna… we wanna do something with them, right? We wanna reach out to them, put them in our CRM, and actually turn these into meetings. So, with any signal audience that you create. you can reach out to them in a number of ways. One is to use our AI sequences feature. So, if we go ahead and click Launch in AI Sequence… we can actually get that going. So what this has done is it’s selected that signal audience that we just created, the seed startups. hiring, the seed startups that sell to dental practices hiring SDRs, and we integrate with these 6 AI sequence providers.

And what we do is we generate the AI emails and personalize them, and then deliver that to the sequencer that actually delivers the messages. So, let’s try an example with Lemlist here. And in the background, this is actually creating the campaign in Lemlist. So, we can stay within Avena and construct that. So… Avina gives you a lot of power here to actually construct these emails and have the AI personalize them well. So that you’re actually going to get replies. I’m actually just going to jump to a pre-filled example, because I want to show you what a kind of fully filled out Set of instructions here. would look like.

So this is, again, the same audience. that we just created. We’re targeting, the… those startups selling to dental offices, and… what we’ve done here is given the AI a bunch of instructions on how to write a good outbound email. We typically… what we do is we use AI to brainstorm. Through Claude Coworker, Claude Code. To actually help write these skeletons. And then, we have the AI on the fly. fill in the personalization token. So we found that that’s the most effective way to run this currently. If you just have the AI straight up write the email from scratch, you’re gonna get what, often sounds very AI.

So we found the best kind of combination, again, is the human judgment paired with the execution partner. The judgment here is making sure that the overall structure, language of the email looks good, and the execution is the AI filling in the personalization slots for every single recipient. So, this is an email that… that, you know, we’re currently sending, but… Just to showcase the personalization here. saw on your Ashby page that you’re hiring an SDR, If I were the new SDR, the first audience I would want is dental practices. So we… dental practices on open dental, whose Google reviews mention billing confusion.

So this is a highly specific, custom signal that we think that company would want, and the AI has actually filled that in based on the template that we’ve given it here. for that recipient, and based on all of the instructions that we’ve given it below. So you can see, you can give it pretty detailed instructions, you can tell it how to research the recipient. How to pitch your company, and you can even attach Pretty deep context into your brand voice, the language that you use, how to pitch your product, and so on. So this isn’t just a black box where the AI writes your email, you have a lot of control over how to do it.

Okay.

Mike Wang:

So, the next thing that I want to quickly explore is… The kind of settings that allow the system to run, that you set once, and allow it to run very effectively and execute each of those steps. Every day. So, again, showing you the personas that we targeted. You can create as many of these as you want. And this is how you ensure that the data you’re getting is complete for every lead that’s found. And, You can define your ICP criteria here. So, this is how you qualify those companies that come in, and ensure you’re only reaching out to highly qualified leads.

And you can define all of those, context entries that define, kind of, the GTM brain that you want the AI to have as it’s writing those emails on your behalf. Okay, so… That’s Evina in a nutshell. I’m gonna just jump back. to the… slides here. And kind of one, you know. philosophical position we have here is that technology doesn’t do the thinking for you as an execution tool. That’s how I would view Avena. It’s helping you execute on each of those steps to get you to highly intelligent, performant outbound. But it’s not gonna, you know, find product-market fit for you.

It’ll help you do that, you can use it to experiment, but ultimately it’s amplifying your sales. process. So this is… Something that, my friend Gerard, said, and I think it was very apt. So, and I would just add one caveat to it, is the technology doesn’t do all the thinking for you. Because it’s starting to do more and more, and, We’re actually moving towards… operating Avena for our customers, and the way that we’re doing that is by creating this external, kind of, sales brain. That, is starting to make, you know, with… with GPT-6 and Fable and all these incredible models, we can start to make a lot of these high-level decisions that AI previously was not able to do on its own.

So we’ve started creating, within Claude Cowork. These kind of external sales brains that contain much of the knowledge about how to write in your voice. What are your competitors? How to pitch your company? What are all your proof points? You know, what has been customer feedback? And that all goes into a system that is making decisions and strategizing, coming up with good signals audiences. Writing good emails, and ultimately, Moving closer and closer to a point where the human can really make those high-level decisions. And… we’re really excited to be hiring our first forward-deployed GTM engineer.

It’s a little jargony, I know, but We think it’s really exciting because we’re moving more towards kind of this embedded model of go-to-market, where we know that even though we can provide you these execution tools, it’s still so much work to get all of these systems working and… And, really getting meetings, and… There’s so many pieces that have to be get to, to be deployed, and they all have to be done right. So… For a lot of our customers, we’ve, again, started to… To kind of help them out with a lot of those things, and we’re scaling that up by having this external sales brain that our forward deployed engineers will be able to leverage as they’re embedded in your organization.

So, yeah, that’s the end of the demo. Thank you, and happy to take any questions.

Julia Nimchinski:

Great demo session, thank you so much, Mike. Speaking of forward deployed, just curious, you know, your take on AI ROI. Do you believe that that’s the best model to actually see it for mid-market, enterprise, and bigger companies, or… How do you see it?

Mike Wang:

Yeah, 100%. I think that the… Kind of embedded model is very important, because any AI system. needs some level of human supervision at the moment. I think that’s just true across the board. AI is still very spiky, right? It can do some things extremely well, and other things very, very poorly. And it’s not always clear which is which. So that’s why I think it makes some… it makes a lot more sense for organizations to go step-by-step and figure out, you know. which parts can AI do really, really well? And… For example. Right? We were able to build a system where you tell the AI what you want, it finds those leads.

That’s… there’s not a lot of judgment there. You’re giving… you’re making the decision of which leads to look for. The AI is executing that, and it’s… what the AI’s really good at is… Consuming thousands and thousands of websites and… Data sources and bubbling up the results for you. But if you were to ask the AI right now. which leads should I go after? Like, who should I sell to? You know, it’s gonna come up with some stuff. But it’s probably not going to be very sound. So I think the right model is to figure out what the AI is really good at for your use case.

And stick to that. Stick to using the AI for that. And what that requires is having, kind of. some level of human supervision embedded alongside the AI. In, in most organizations.

Julia Nimchinski:

Thank you. You’re representing.

Mike Wang:

iceberg.

Julia Nimchinski:

Yeah, yeah, definitely. I’m just curious your take on, you know, the future, kind of, go-to-market organization. How do you see it, from the perspective of being in, you know, a truly AI-native company? Do you see the functions of marketing, sales, customer success blend into one, or how do you approach it internally, and what are you seeing, you know, in all of your customer stories currently?

Mike Wang:

Yeah. A lot of it is in flux. Things, things change quickly with… Especially with AEO and new channels popping up and things like that, but kind of the fundamentals are the same, right? You want to get in front of… you have… you’re offering something of value, and you want to get in front of someone who feels the pain that your product or service solves. So that will always be the fundamentals of sales, like, whether it’s outbound, inbound. SEO, you know, ABM, GEO, whatever it is, or even, like, agents selling to agents, right? That’ll always be the equation.

It’s… it’s, it’s… I have a product or service, and do you have a pain point? If those match, then… then… that’s where… that’s where we want to make a sale happen. So, I think right now, again, I view AI right now as an execution tool, so I think the overall channels are slowly shifting, like, there’s AEO is a new thing and everything, and some… have found that outbound is getting more saturated. That’s definitely true. However, I would still say outbound is an incredibly important channel, Because if you don’t have it, then you’re kind of just waiting for people to come to you.

But that being said, I think in each of those channels, whatever it is, AI should be viewed as an execution partner. So if it’s SEO, for example, you don’t just say, hey, AI, go run my SEO, you know? You… you give it, you want to give it as much… of your judgment and input as possible. So… so maybe that’s, you know, hey, these are industries that we target, and here’s some content that we have that we think will resonate with that industry. I want you to generate some You know, break this content up, give me some social posts, give me some blog content.

And use it as a generative kind of execution partner. That’s how I see the future go-to-market in the near term. In the longer term, that’s where we’re getting into more of you know, it’s kind of just a continuation of that, where the AI is able to make more and more high-level decisions, and you can start trusting it to do something like, hey, here’s my website, go around my SEO. You know, eventually we’ll… we’ll have that, but there’s a… there’s a series of steps. Before we get to that level of automation.

Julia Nimchinski:

Definitely. Can we address the, I guess, the first point of skepticism when it comes to, you know, all things autonomous, semi-autonomous, specifically data enrichment and… you know, making sure that the data is relevant, and, you know, up-to-date. How do you approach it internally and, you know, product-wise?

Mike Wang:

Yep. Data… the data is and always will be pretty messy. There’s so many different data sources, it goes out of date, you have to clean it, you have to get into the right systems, they have to talk to each other, there’s just, like, so many moving parts. And so… Kind of our philosophy on that is… We want to be kind of, like, the central… spigot of data that you can rely on, and I think that… I think that’s… Generally, how this… Will continue to evolve is you’ll have all these Dozens, hundreds of data providers. feeding data into the market.

And then you’ll have companies that kind of sit on top of those, like ours and a number of others that unify the data together. And so you can kind of think of it as a pyramid structure, right? You’ve got, like, raw data sitting on LinkedIn. on the web, whatever. And then you’ve got companies that are offering that as an API, they’re sitting on top of that raw data, and then you’ve got companies unifying all of those APIs, doing the cleaning and so on. And then you’ve got the end consumer, that probably just… Your, like, your goal as an end… consumer, as in your sales team, is you just want that clean data.

You just provide the filters, you get the clean data. So I think it is, kind of like, everyone’s kind of, like, standing on the shoulders of giants, type of equation.

Julia Nimchinski:

Makes sense. Mike, as a CTO, how do you approach, actually, your harness internally? Are you, like, in terms of architecture, is it open models, closed models, combination of the two?

Mike Wang:

It’s everything. I mean, we have, Our goal will always be to To, provide the highest quality results, and… the results that provide the most coverage for the lowest cost. So we’re constantly testing… every single new model that comes out, we just add it to our test and eval harness and see where it lands in terms of that, cost performance. Matrix, and… Yeah, in terms of the harness, I think there’s a kind of… sometimes… people get confused between an agent harness, where the agent is kind of, fully autonomous, like Claude Code, for example. You give Claude Code a task, and it kind of just starts reading… it decides, like, which files to read, like, which APIs to call, Just, like, making a ton of decisions for you.

And our system is mostly not that. So, we, again, want all of this data and each of these steps to be reliable, repeatable, high quality, and inconsistent. And the way you do that is… is by… Making much more, Kind of scoped… LLM calls. So, similar to how, you know, you want to figure out what AI is good at, right? If you give AI too many decisions, that’s when you run into problems. So we give the AI as small of a problem as possible. We say, hey, this is… there’s a lot of context engineering involved, so we say, like, you know, here are the… Five things you need to know.

Just generate the output, and we do that in a series of steps to get you a consistent output. We do have an MCP, so that’s… One way where you can access all of this information and functionality with kind of like an Agentic partner, so you can spin up Claude Code, say, hey, use Avena to find me this type of company, and get all the contact info, and create an AI sequence that, will go to them. And, you know, Claude Code will do a bunch of thinking and use our, endpoints that are really good at doing those specific things, and kind of make those decisions on top of it.

But again, you want to supervise that, so… So we’re providing the tools to… To execute really well on targeting timing, personas, enrichment. generating emails and so on. And, As it stands today, the harness can… use those tools, but I would still be cautious about kind of letting it just do it all on its own. You wanna… Ultimately, you know best. How to sell your own product? And if you don’t, then, You know, that’s the first thing that, you should… you should work on. And you can use AI to help you brainstorm etc, but… Yeah, that’s how I do it.

Julia Nimchinski:

Thank you so much. And what’s the best, next step for our people to support you? Where should they go? Is there a test drive of this?

Mike Wang:

Yeah, you can sign up on our website, it’s a free trial. I’m happy to do a personalized demo for anyone who’s interested in diving in, and We can test an audience that you want to, see for yourself, see what the data looks like. And go from there. We also do white glove for folks that, are interested in having someone with expertise kind of just, like, run their outbound. So… Yeah, I’m sure my email will be shared. And feel free to reach out, happy to chat.

Julia Nimchinski:

Amazing. Thank you again. And…

Mike Wang:

Thanks, Joy.