AI-First Sales Operating Model — Showcase with Ameya Deshmukh

Julia Nimchinski:
We now welcome EMEA, Head of Marketing at Everworker, and I’m so happy that we won. Zoom background.

Ameya:
Hey, Julia, good to see you again.

Julia Nimchinski:
I was worried.

Ameya:
I learned when I joined you for a second as well. I guess the Zoom virtual background does not play well with my face shape. Perhaps it’s not been trained well on people that look like me.

Julia Nimchinski:
It’s a lot of brightness.

Ameya:
Thank you.

Julia Nimchinski:
Yeah.

Ameya:
But I do have the extra little light here, and the white painted walls in my room likely do.

Julia Nimchinski:
No worries at all. Super excited for your session, and we are diving in into AI for a sales model. In practice, so yeah, let’s… let’s just get into it.

Ameya:
Yeah, I am not a big slides guy, so I’ll open it just by talking. And, after that, I’m actually just gonna go right into our HubSpot and show you all how we do it internally at Everworker. So, I think when you’re first thinking about creating an AI-first sales operating model.
there’s some groundwork you have to lay down, and that is really the fundamentals, right? So, you need to know who your ICP is, you need to know your segmentation, or you need to have some sort of hypothesis around it. So, what are the firmographic, technographic, and demographic segments you’re going after in terms of accounts?
buyer groups at those accounts, and then job titles at those accounts. Okay, great, now you’ve got your hypothesis. Now, this is what everyone is doing based off of that. What everyone’s doing is taking that, going to Clay, downloading those target account lists, downloading those contacts, and then mass emailing them.
And… All the startups doing it have figured out deliverability, because that is the unsolved mystery that was unsolved 4 years ago, but now anyone that’s worth their salt knows how to buy domains, warm them up, get 3 emails per domain, send no more than 5 per day. So, the competitive landscape is such.
Everyone has the same exact access to the same data as you do, everyone has the same ability to segment as you do, and everyone has the same ability to land into primary inbox, just like you. So, how do you actually win with outbound or even inbound lead follow-up with sales today. And that’s the first thing I want to talk about.
And the way you do that is with SDRs. But here’s the problem, and this is why a lot of startups work with us, is you go back and think about how sales teams hire SDRs, and most of them are still following the predictable revenue playbook, which is what was used to grow Salesforce.
Except the economics of that playbook don’t work out anymore unless you sell million-dollar deals. And the reality is, I actually just went and interviewed maybe 15 or 20 different people for an open SDR position on my team. And here’s what I found across the board with the best candidates. they were generating 15 to 20 meetings a month, right?
The best performing SDRs. So that’s net total of 45 to 60. Each one of those headcount hires is going to be anywhere from 70 to 100K. Now, you want to scale, that means you need to scale healthcare, but the problem is, no one can afford it. So, how do you scale what that SDR does?
know you can infinitely scale infrastructure and have all the data that you need to reach out to these people without scaling headcount.
And that’s the first thing that an AI-first sales operating model solves, is creating an AI SDR, but not buying one off the shelf, creating one that’s tailored to your messaging, your signals, the way you research, and the way you tell stories and do outbound. So you do that, which we’ve done for ourselves.
have as a template in Everworker, what we see is, for our inbound marketing-generated leads, which we did 10,000 this quarter, we are converting 5% to 15% into meetings and deals. And the reason that number’s broad is it varies depending on lead source, right?
So white papers obviously convert a little lower, stuff like this webinar and events, a little bit in the middle, and then our inbound demo requests and sales requests obviously convert at the highest, with some of those converting around 35%, actually.
So… that’s where you start, is take the leads marketing is getting, make them automatically get followed up with perfectly personalized, really insightful sequences, all written by an AI, and just scale your SDR’s calendars.
So, the reason I hired an SDR using this model is I had one BDR until now, and she did 153, deals this quarter, opening them up. So she had to take about 200 meetings, and that doesn’t even count all the meetings that AISDR generated for her in the Americas time zone, because she’s in Spain.
And, me and two sales reps took all the meetings that it generated in the U.S. So that’s the kind of volume that we’re doing right now. Now. The second angle to it is your outbound, right?
So, because everyone has the same access to data as you do, that means everyone runs the same damn outbound plays as you do, and everyone uses the GPT spin tax via that API. It just doesn’t work, and we tried that Two quarters ago, and we sent 20,000 emails using that approach. Nothing. Almost nothing from it.
Perfect deliverability, great open rates, pretty good reply rates, but nothing. And the reason for that is that you need signals that are actually relevant, and that’s what has to trigger the outbound. So you have to send that perfect email at the right time. So, now you can go buy those signals from data providers.
But you’re getting upcharged, and they’re making crazy margin on it. And you can’t be sure about the accuracy of those signals, because it’s a black box that they’re giving you those signals from. So, what we did is we built a bunch of signal AI workers. So, the first one we built is one that looks at all of your target accounts.
Mine looks at mine in HubSpot. So, what it does, looks at the target account list, and then for us, if someone’s hiring for, say, a recruiter or SDR, that’s a signal that we should reach out. So it’ll go hit the company, do web research across all the job boards, count all the open roles they have, classify them, push that information into HubSpot.
And that’s just one of about 20 different signals we’ve configured it to grab and push. As soon as those properties get updated in HubSpot, another worker comes in that looks at the companies that have those properties. Sends it to other AI agents, which make decisions about what to do based off what property is filled.
And then use webhooks to go and run clay tables on our behalf. So now, we have the signal, we’re orchestrating off of it, and we’re grabbing the right people, and then putting them in HubSpot and tagging them. Then, I have the third AI worker in this whole chain that’s looking for those people that are coming in.
It’s grabbing them, and now it’s sending them to an outbound AI SDR. Since we’ve done that, now we’re starting to see, as we scale that, 2% to 5% conversion rates from outbound to interested reply and meeting bookings, which I think is pretty damn good for outbound.
I don’t know what your experience has been with the folks on this stream in terms of converting from outbound, but I’m quite pleased with that outcome so far.

Julia Nimchinski:

Julia Nimchinski:
It’s always… it sounds futuristic, it sounds incredible, and, you know, like a dream, and the point where everybody wants to… to basically be it. But, I’d love to address the main, like, question that we received from the community. How long does it take to get there, Maya? And what level of customization?

Ameya:
Yeah, it actually doesn’t take that long to get there at all.
So, if you have absolutely nothing right now, in terms of infrastructure, we’ll guide you through setting all that up, and it’ll take you, say, 2-3 weeks to get your domains and get your inboxes warmed up and connected into a sense system, which I don’t actually recommend using instantly, because marketers are in everything.
So, so many people are using that thing that, it’s actually not delivering very well anymore. So, I’ve switched over to Email Bison, but you use your tool of choice, it almost doesn’t really matter all that much. So, we’ll get you all set up with that.
While we’re doing that, at the same time, we’ll get your messaging and positioning, we’ll take our templates for all of these things, and swap out two parts of them. One is your CRM, the other is your property fields and logic.
And then inside the personalization engine and the things that are writing and researching, we’ll just have you help us swap out the natural language instruction sets on them.
So, here’s how… Here’s what properties to look at, here’s how to classify them to intent categories, here’s the set of web research to do on the company and prospect record to personalize your sequence, and then all we need from you is your messaging and positioning documents and your persona docs, and… Net-net, you’re looking at having a system like this up and running in 45 days, booking meetings on your behalf.
And that’s because We’re platform plus services plus templates.

Julia Nimchinski:
Amazing, and you’re just basically in touch every step of the way, and iterating, and improving, right?

Ameya:
Correct. Correct. We actually do it for you, as a service, because what we realized really, really quickly… Can you hear my dog barking, by the way? Yeah. Okay, hopefully someone won’t get that. One second.

Julia Nimchinski:
Live today.

Ameya:
River. Jesus Christ Apologies. The joys and pains of having a herding dog as a pet that has guard dog tendencies, it hates the UPS delivery guy. So, back to what I was saying.

Julia Nimchinski:
that…

Ameya:
Yeah.

Julia Nimchinski:
Your reality, yeah.

Ameya:
The, reason we do that is we built, actually, the easiest platform for business people to create complex AI workers like I’m describing. What we realized really, really quickly is that our ICP, who are business people that are advanced enough in thinking about AI in this way.
have been forced to become advanced to think about AI in this way because they already don’t have time. So, it doesn’t matter how easy the platform is, they don’t have time to build anything. So, then we decided we’ll build templates ourselves, and then we’ll add on services so it’s done for you.
And then once we create some operational breathing room for our customers. We’ve trained them on how to use a platform as well, and they’re off and running to the races on their own, but that’s why we do it like that.

Julia Nimchinski:

Julia Nimchinski:
Awesome, let’s see it in action.

Ameya:
Sure. So, here is a small view to our pipeline, and… Here are a bunch of leads coming in. So as you’re seeing. This guy came in at 5 AM this morning. These folks came in at 6.34 a.m. last night. And they’re already being sequenced and reached out to.
Now, in terms of signal properties themselves, that’s happening on the company record, but the reason that this is so easy for us to customize TO is that all you’re really doing is working with us to change this.
So the first is the knowledge that the worker’s going to use, and you’re going to use the same exact knowledge documents that you would give to your human team of SDRs. So, whether you have that in Confluence, or you have it in SharePoint, or Google Drive, wherever you’re managing your company knowledge is where you manage it for your AI workforce.
The next part of it is that you’re just going to connect in your knowledge here to the SDR, like so, by checking a box. That’s your messaging and positioning guides. And the beautiful thing is, when your messaging and positioning changes in your system of record for your Knowledge Center, it automatically updates and changes here.
So, once you set up your connection to your agents, you never have to worry about keeping them contextualized again, which is a very beautiful thing, especially if you’ve ever… set up a RAG pipeline using something like N8N and Claude Code, which is cumbersome in comparison, but good. Here, you’re gonna pick a brain.
You pay for your own model token consumption, and that’s because we don’t want to charge our customers margin on what we view as a electricity or utility, so just pay your electricity costs direct to PG&E, whether you’re Running 10,000 emails to folks, or 5 we do not. care and want you running it at scale.
From there, you’re gonna write out instruction set, which is also baked in here as a template and done for you from an interview we do with you, but you’re going to give it a role and objective, and you’re gonna tell it how to do its job. So the first thing is classify based on job title into an intent category.
Then, for us, based on the intent, job title, and company context, execute these research instructions. Obviously, it’s only going to pick one from here. Then follow these best practices for writing a sequence. Output format is for workflow, so it’s going to output as a JSON array.
This is done for you as well, so you don’t need to worry about that. Then final instructions are your guardrails and fences. No skills here, because again, this isn’t a workflow. And then we chain it all up together into a neat packaged workflow like this that runs always on. And does this at scale across all your lead sources.
Now, what happens from there is they all go into our sending platform automatically, which is Email Bison. And these are the folks that are in right now in our… let’s go look at this guy in Zylos. There you go.
Zios, this guy’s an intern that requested a demo, which I actually still will follow up with, because, funny enough, I worked at Techst Group previously.
We won one of our biggest accounts, which was UPS, from an intern actually coming in and teaching, talking about our marketing content, and then wanting to buy the platform, which was a massive deal for us.
So… Here, it’s even gone and checked out that Silas is doing monthly webinars by looking at its website, looked at their services and offerings and their target audience, and brought that into here.
I looked at who he is, found out from his LinkedIn profile that he works on the marketing team as an intern, and determined what he’s working on as well, and then told him from our solution set what we could do to help him address his pain points and needs, and remember, those are now all contextually correct to stuff his company’s actually doing and he’s responsible for.
So, this is the type of outbound and inbound follow-up At scale that an AI SDR is about able to do for you, and this is why it works so well. Now, I’m just gonna show you the… AI SDR right now, but the other things I wanted to talk about were… the other AI workers we’ve built for our sales team.
So… getting the lead to a meeting is just the first part of the sales process. You also have to work the entire sales process with your account executives, and the math there that has been a constraint historically has been that one AE can really only manage, say, about 20 to 30 deals at any given quarter at the same time.
That’s untrue, actually, today. And the reason why, historically, that’s been true is that, well, you’ve got properties that they need to update on RevOps, you know, and we run the statistics on this ourselves.
that to get a lead from the first meeting in a deal to, say, halfway through your sales process is going to require about 14.8 touches in a period of 2 weeks. So, apologies, mixed up the stat. 7 to 11th touches over 14.8 days will produce a statistically significant 72% lift in the likelihood to progress to closed one.
Now, the challenge is, how do you execute those touches across all channels at scale, quickly, and personalized? So, that’s the first thing we did with a the first part of our sales playbook, AI Worker. So that thing, new deal comes in, helps the sales rep. Do that execution, writes those emails, schedules them.
puts them as well onto a LinkedIn sequence, which we do through a Lembless integration, and that way it’s delivering touches. Now, the second thing we found from our analysis was that multi-threading helps lift deal conversion probability.
So, that’s the second AI worker we built for the sales team, which is, you get into a deal, you fill out who you know, what you know about the deal, and what the buying group you think is.
Then that calls Clay, grabs additional contact from the buying group that you think might be in there, pushes them into HubSpot, associates them with the deal, and then goes and writes Touches to those folks as well, to try and get the buying group consensus going and help you build a business case and consensus across that. to the deal cycle.
Now, the other part that we realized sales reps needed help with was, at each stage in our sales process, we want to send the prospects personalized documents and material. So, what we do is we’ll write up agent architecture proposals.
So, we’ll talk to someone first call, then after that, we’ll take the call recording, put it in HubSpot, an agent will extract any insights and proposed use cases from there, then another agent will pick that up.
write out an Asian architecture proposal, send it over into Gamma, which is a design tool where I’ve built a template, then Gamma will design with the template. the agent would export that and upload it into HubSpot, and now we’ve got a shareable sales asset that we can go out.
Currently, that asset’s being sent manually, but that’s going to happen with an AI worker shortly as well. I just want something that’s going to ask the reps for permission before it does things, because obviously you’re working a deal, you don’t want things running fully automated until you’re comfortable and confident in them.
So… Those are some of the early-staged AI workers in the sales process, but the third one we built that’s really, really compelling is the business caseworker. So, that one we actually run live on a call with the prospect, and let me just pull it up quickly here, because this one is very powerful.
So, for example, I’m just gonna scroll through, past that little company information. But, we talked to a company in the Nordics, and these are the notes that we extracted from this call transcript using that first AI worker.
We put them here into the chat on the proposal, and this wrote up an entire proposal along with EBITDA, Calculations, analysis breakdown, and ROI investment summary, which then was sent into Gamma, Let’s see if I can pull up an example here… Here’s one that we did for Honey Love.
And it automatically designs an artifact like this, where key executive summary, proposed solution. overview of each AI worker, and then a financial impact analysis and EBITDA calculation. and total ROI summary, and then business proposal.
And our clients and prospects obviously love this, because articulating the value of our solution and getting buy-in from their executive team becomes a lot easier.
And because we do this live with them on the call, they also, at the same time, see the Power of More Platform, as well as have the opportunity to discuss with us Whether they think these analysis and, and calculations are actually accurate, which we baked in the AI worker’s logic to be very, very conservative with these estimations on purpose.
But that’s just another… another part of our Sales Playbook AI worker.
Then, one of our reps actually built a RFP worker, so all of the reps are using that now, so anytime someone gets an RFP, they go give it to that worker, and then it writes up and answers it from checking our product documentation, and then everything that it writes, it goes and stores to update its shared vector memory space with.
So, as our reps complete more and more RFPs, our RFP worker gets smarter over time, and that shortens that process significantly as well, so you’re not bothering our chief product officer or other technical product leaders anytime you’re working a deal and you get an RFP and technical questionnaire from IT. So…

Julia Nimchinski:
Maya, folks are complimenting you for making a great case for AI Accelerate Copilot. Engineers into the sales pipeline, and they were curious to know what barriers to adoption within the sales ranks, and how do you address it, and how do you enable human in-the-loop control?

Ameya:
Barriers to adoption on sales teams. So what the actual sellers, like account executives, SDRs. they… I go talk to them at even, like, conference booths when I’m working a booth. For example, I’ll go to other booths and just talk to them about their sales process. They love all this stuff, they want to do it.
Typically, the barrier here with salespeople is… that they’re just not thinking about it in this way, haven’t thought of it or heard of it, or… or they don’t know what their sales playbook even is. So if you don’t know what the set of actions are that you need to do to progress a deal from those pipeline stages.
then, I mean, we can give you our AI worker and give you and build out a sales playbook for you, but it kind of requires you to be a little mature about how you’re thinking about sales and running a playbook properly.
And then it does kind of require you to have a good RevOps team, at least to have set up, like, hey, these are the properties we want to track on each deal stage. Here, we’ve been doing it a little manually over time, and then that’s how we did it as well.
We know, for example, that if the BANT criteria properties on stage 1 are filled out, that has stat sig left on the deal accelerating towards Stage 3 or 4. So, once you have confidence in those figures, and you know your process is working, that’s a great time to scale it with an AI worker.

Julia Nimchinski:
Another question here, how do you make AI-written emails sound truly human, rather than often robotic and generic? Slop? This is the most popular question I’m seeing every day here, that is generated, even with training, and it shouldn’t be too overly formal and stiff, and yeah, let’s address it.

Ameya:
Yeah, it’s a great question. It’s really just two things there.
So, to avoid the model with coming up with its own BS, which is what AI swap is, you need to do two things, and this was covered in a paper, that was called, All Computable Models Will Hallucinate Infinitely, and there’s only three things that you can do to prevent that, and that is, one, guardrails, two, fences. slash instructions plus context.
So, as you saw in the instruction set that I breezed through quickly, within it, there are about… 30 bullets, or sections, you could loosely call them, on best practices for writing and copywriting these things.
There’s also sections on deep research tasks that it has to do, so that prevents the model from just hallucinating from its model weights and just coming up with the slop and bullshit that you guys are running into. The second is the vector memory.
So, I’ve built out messaging and positioning docs for each one of our segments there, and that’s what it’s going to retrieve and look at. So it’s not coming up with our messaging on its own. And the third is guardrails and fences.
So, as I breezed through the final instructions, you saw there were about 15 different guardrails and fences that I gave it, and that’s… that’s how you prevent that AI slop.
Now, that looks very simple, but it’s because we’ve intentionally made the front-end UI of our platform simple, and it’s doing sophisticated things like that on the back end to get you to the outcomes that you want, so…

Julia Nimchinski:
And folks want to know, some of the results and, what you’re seeing in production, if you can share some case studies.

Ameya:
Yeah, absolutely. I mean, we’ve rolled out, this AI SDR to… there’s 4 different startup companies right now that are using it and scaling it.
One is doing it, actually, to a technical developer audience, so that one was even more complicated, because it had to ingest technical docs and really think about code and the other pain points someone might have. So we’ll have case studies coming out soon with those.
I’m always happy to share our own internal figures, which… you’ll look at… I’m not gonna do this on this call, but you come book a sales call with me, I will gladly show you my pipeline in HubSpot, and you can see that every single quarter since we’ve been doing this, we’ve been growing it by about 25 to 40% growth quarter over quarter in net new pipeline that we’re generating.
But what I’ll tell you is marketing budget has not gone up, team size has not gone up. And sales and SDR team size has not gone up either. We’re just getting really, really efficient from becoming more and more AI first.

Julia Nimchinski:
Another one here. For your biggest wins, was user adoption and training a new effort or an existing state of excellence? And help us understand who the ICP is, RevOps versus sales ranks.

Ameya:
Great question. Change is hard, right? Getting people to change is hard. That’s why, you do have to make a commitment as a sales organization to become an AI-first sales organization. What you’re gonna see when you do that is a bell curve, right?
So, some percentage of your sales reps are gonna become builders over time and want to build things, like my rep that built the, RFP AI worker. Some of them are just going to be users, and then some, your probably lowest performing ones, are going to require something like a mandate, carrots and sticks, right?
So, they’re going to need a stick, where, hey, you use this, or else. And that’s how you’ll get those folks. But the vast majority of people will just want to use this. So, where RevOps comes into play is actually your, your systems, right?
So, deciding what systems to use, mapping the properties, and then working with us to make sure the AI worker is pulling from those properly and architecting it well. But, yeah. It’s fairly simple.
All we really do is a one-hour, like, roundtable training every week with the sales team, sharing what we’re releasing, sourcing ideas from them, and then teaching them how to use and leverage it. But adoption is very, very fast, because the platform makes it easy.

Julia Nimchinski:
May I… can we speak to the before-after conversion metrics?

Ameya:
Yeah, so… There’s not a whole lot to talk about here for me, because we’ve been going to market for about 3 quarters now, and I’ve been building this as I go.
But I’ll share with you, before we had this system running on fully autonomously, we were only able to prosecute something like 20% of the leads we were generating within a given quarter, so there was just a ton of opportunity left on the table. Now we’re able to prosecute 100% of them, and the other thing is.
Before we implemented this, even for our inbound leads. we got good conversion rates from our demo requests, because obviously those are high intent, and I was doing the standard, old-school thing of build lists for my SDRs, task them to hit the lists every day, and they were doing that there.
But for everything else that was higher volume, it just wasn’t getting followed up with properly, or if it was, it was just getting a generic sequence. So conversions were about, like. 2%. for our inbound, for the ones that we actually prosecuted, which was 20%.
Now we’re prosecuting 100% of them and converting 5-15%, so you do the math on that yourself. It’s dramatic.

Julia Nimchinski:
We have 2 minutes, and a question from Alan here. So, as a founder, he has provided AI with technical papers, investor docs, and a lot of context with this latest startup, Information, and in some email outreach, it literally invented services and positioning. And it wasn’t ever discussed or agreed.
So, I mean, he’s just asking, how do you manage these types of hallucinations, and are they extremely rare?

Ameya:
With us, it won’t happen, and the way we manage it is we built an enterprise knowledge engine, so we have a very, very sophisticated context system baked into the platform. So, let’s take, like, your investor deck, your messaging and positioning, your persona documents, anything else.
we would separate those into 4 different memory items, and then connect those memory items into your AI worker, and then as you write the instruction set, you say, hey, classify the person’s persona, then based on the persona, query the persona’s memory and select the right persona. So then it’ll do that and grab info about that.
Then you’ll tell it query the messaging and positioning memory to determine what part of our service is an offering to promote based on your research on the company in person, and so on and so forth. The other thing you’re probably running into is if you just straight up uploaded your PowerPoint.
deck into context, agents are gonna have a really hard time vectorizing that and then retrieving from it. So… Just convert it into plain text, or .

Steve Davis:
Oh, wonderful.

Ameya:
And you’ll get better performance with your existing tooling.

Julia Nimchinski:
Thank you so much, Yemaya. What’s the best next step to engage here? Should folks just reach out to you directly on LinkedIn, or…

Ameya:
Yeah, you can reach out to me on LinkedIn, you can come to our website, there’s a pop-up that’ll come up to harass you, or you can directly book meetings with me and my team on our calendar. You can also talk to my V1 voice chat, AISDR, on our web chat.
She’ll ask you some qualifying questions and get your contact info, and then try to book a meeting with you. That is pretty much in beta testing phase right now, so it may be a little buggy, but Do that, or just submit a contact us request on our website, and we will reach out immediately.

Julia Nimchinski:
Awesome, thank you so much again, and we are transitioning to our CEO Roundtable, and I’ll start one.