Reverse-Engineering the AI Buyer — Aliisa Rosenthal, Acrew Capital

Hi everyone, uh, my name is Aliisa, and I joined a little research lab called OpenAI a little over 4 years ago. When I joined, it was, you know, a couple million in revenue, we had a self serve API, we had some developers who found it and swiped a credit card and started building.

Um, and over the next few years I helped grow the organization with several hundred people, grow revenue to several billion on the enterprise side, and launch multiple products. So I really had the opportunity to be at the bleeding edge of bringing this somewhat alien technology to organizations, from startups to massive enterprises, and I'm going to share a little bit about what I learned along the way.

So I think most organizations typically have said, okay, I'm ready to build out my go-to-market organization, I'm going to hire a bunch of salespeople, I'm going to hire rev ops, I'm going to hire sales engineers, and I'm going to build out my sales organization, and then I'm going to figure out where the bottlenecks are, and I'm going to add AI, or I'm going to add automation to figure out how to scale this thing.

My current advice to founders is do the opposite. So start with what you can automate. Start with, there's so many tools available now in 2026 for a modern go-to-market or sales organization that have never been available before, which really turns the entire playbook on its head.

And you'll find that a lot of traditional sales leaders are very used to coming in and hiring a really big team and augmenting everything with humans, bringing in sales operations, sales engineers, lots of salespeople, BDRs, SDRs. You can do so much of this through an automated approach now.

So my, my advice to founders is build the machine first before you build the team. Figure out what you can automate, figure out where the bottlenecks are and what's broken, and then add humans on top of that to get through those bottlenecks.

So just a story about how this played out at OpenAI. So we launched ChatGPT in the end of 2022. When we launched it, we had no enterprise features whatsoever, and I got bombarded with organizations saying, hey everyone, my company is using ChatGPT, can I get SSO, can I get an NDA, can I get an invoice.

I had none of these things. I had no enterprise features whatsoever. Uh, so over the next few months I begged and begged and begged, our technical team, can we build an enterprise version of this product, can we build some enterprise features, and finally, nine months later we got approval.

Now over that nine months time, the companies that had been the loudest, the companies that had gotten to me, were the bigger companies, were the large enterprises. So I thought, okay, we should go out and build a really high-end product for all these enterprises that want to use us.

So we added every bell and whistle, we made it lightning fast, it was a very expensive product that we shipped, we went up market, we went way up marketright out the gate, and you know, we did fine. We, we did pretty well, um, you know, and, and I think for most companies we did great.

Uh, but after about a quarter we said, well, we should probably also offer a self serve version. Not everyone wants to opt into this heavy sales process and start with these really high price points. So then we released our self serve motion in January of 2024, so about four months after our enterprise version, and it just completely cannibalized our enterprise business.

Uh, it grew so much faster. It turns out most people just didn't want to talk to a salesperson, and our mid-market SMB sort of self serve business was much higher growth than our, than our heavy enterprise business. So it frustrated the sales reps because they now had to compete with self serve, it sort of cannibalized the customer base.

They sometimes preferred to just, um, move into the cheaper self serve option. And so that was a really valuable lesson. We should have launched with self serve first, listened to feedback from our customers on what we they weren't getting from the self serve product, and then figured out how to add, build a more expensive enterprise offering, and then hire the team to sell it.

So we did it all backwards. It was a really, really valuable learning experience, and from there we decided not to do that again. For every other product we launched at OpenAI, we made sure to, to do it self serve first, learn where we needed to tack on humans and add them.

Uh, the other thing I'll mention, and, uh, you know, uh, if Everett's still here, nice, uh, shout out to him. We used Clay at OpenAI and it was really, really useful for us. So in the beginning we were just completely overwhelmed with inbound, which I know sounds like such a champagne problem, but it was me and four sales reps and we were getting, you know, 10,000 inbound a day.

I mean, it was just the most staggering problem to deal with this tidal wave of inbound. And, you know, if I could go back in time, there's so much I would have done differently. I would have, first of all, added way more fields to our signup form like phone number, because eventually when we did catch up and we built all the automation and we figured out how to handle these leads, I didn't have a lot of information to go follow up with people.

So add a very in-depth sales form. It can be optional, but if people are willing to give you their phone number, you should get it because at some point you might want to call them or have AI call them.

Um, so gather as much info as you can. Try to respond to everyone. And I, I, you know, I wish I would have done this, at least just sent some email out saying, hey, we hear you, we don't have our enterprise product ready yet, you're on our waitlist, you're on our list, tell us XYZ about what you're looking for.

Just some sort of automated email campaign to make them feel like we cared. Because when we finally did launch our enterprise product nine months later, almost every company I went out and talked to said, well, you never got back to me, and I went out and I brought Microsoft Copilot.

Um, so that was a, that was a really hard lesson. Um, and then similarly on outbound, uh, you know, there's lots of great tools out there now. Clay is one of them. I hear really great things about Nooks.

Uh, there are really wonderful tools to start researching, sequencing, dialing, and automate as much of the outbound pipeline gen as you can so your reps can focus on the qualified leads and the high-value conversations.

So this is my biggest advice to any company starting to build their go-to-market motion is just make it easy on your buyers. Uh, don't give them homework. Don't make them go and fill out a form or, if you can avoid it, integrate something without you hand-holding them through that.

Never say, hey, go do these five things and then come back to me, or go ask this person about this and come back to me. You do as much of the work for them, you hand-hold them through that.

Maybe that means you have to set up hackathons and come in their office and do it with them in person. Um, maybe that means you have to set up some sort of Zoom session and do it live with them.

But whatever you do, don't send your buyer away to do something and wait for them to come back because you lose total control of the sales cycle and it makes everything take longer. Secondly, remove as much friction as you can in the buying process.

Fewer options is better on pricing. Make the paperwork easy and fast. Make the approval flow as manageable as possible. We've already talked about investing in self service, doing that upfront before you add on humans. And the other thing I'll hammer home is to avoid pilots at all cost.

Um, and this is something that comes up when I talk to founders a lot is they say, we are stuck in pilot hell. We've got all of these pilots going on, none of them are converting, there's so much work, or if they do convert, it just, I have to run an entire sales process all over again.

I sell the pilot, I invest really heavily in managing the pilot, and then I have to start all over again from scratch once the pilot, I get a yes to move forward on the sales cycle. So use pilots sparingly.

They should be not the default, but the option for your most valuable, uh, highest paying, biggest revenue opportunities. Now, how do you get out of a pilot? Because every customer will ask you for one. They'll say, I need to try this out, I need a POC, it's part of our process, I need to validate this works in my environment, it meets my needs, we have, you know, all of these things we need to validate and check on.

Just remember, as soon as you give someone access to your product, you're giving away a lot of power and leverage in the deal cycle. So there's lots of ways to get around these objections without a pilot. You can introduce them to another customer, you can do an eval on part of their data, you can do a demo with their custom data and you do it over Zoom rather than handing them access to it.

You can give them an opt-out clause. It's not something you do when you're a giant company, but when you're a startup, you can get away with these things. So sign a contract, you have 90 days to opt out if it doesn't meet your needs, and that shifts the onus of trying out the product and validating it, uh, onto them because the clock is ticking on them now.

Um, so it gives you a lot more control over the deal. And then another one I like to do is I say, well, our normal, uh, contracts are three years, but we'll do a one-year POC with you. So there's all kinds of ways to sort of negotiate your way around a POC without putting tons of resources and getting stuck in them.

The other thing is security. Security is where deals tend to stall out and die over time. Uh, there's a lot you can automate now. There's so many wonderful tools on the market to build these trust portals where, you know, companies can come in, they can auto sign an NDA, they can self serve for your pen test and security documentation.

There's AI products that can auto fill out your security questionnaires. So automate as much of the security as you can. And most companies will tell you, oh, I need to get on a call, I need to talk for two hours.

Push back on that. You know, push them towards your trust portal and then make them come to you and say, here, you know, here's what I couldn't find, here's what wasn't available, here's why I need a call. And try to automate security because a lot of it really can be automated now, which is really, really, uh, speeds up the sales cycle.

Um, so another thing we learned, another expensive story, and this goes back to sort of building the Porsche before the Prius, is we built this, this really high-end product, this enterprise product, and we set the price really high.

We set it initially, ChatGPT enterprise was $60 per user per month. Now we were really the first ones on the market and we didn't know what pricing was going to be. Uh, so we, we basically priced it according to how expensive it was for us to serve it.

And then all the other products came out on the market, Copilot and Gemini and Anthropic at a much lower price point, and we realized we had set the price too high. Um, so we eventually went and we did lower the price, and what we did, we changed the model to basically a license fee, a bare fee to use the product, and then moved to usage base.

And then we just saw, A, the barrier to entry went down, so we had so many more companies signing licenses and signing contracts with us, and then usage went up over time as more, as we sort of the barrier to entry went down.

Because when it was $60 per month, organizations would come in and say, well, I'm only buying this for a subset of my team, or only for my developers, or only for my investors, I'm not going to buy it for my whole company, it's too expensive.

But once we lowered the threshold, it spread like wildfire. Everyone started using it, and it was a much better pricing model that we learned the hard way. I know over time now seats are out, this, this has become common knowledge, we learned it the hard way.

And I also know that now usage is also getting pushback, so it's company pushback on as companies are seeing cost of usage skyrocket. There's easy ways around this. So you could, for example, cap, you say, hey, we have a dashboard, you can cap your spend per month.

Most companies don't end up using that, but they like to know that it's there. You can cap certain employees' spend. But people like the idea that it's a low barrier to entry. People can use it. If they're using it, ideally they're getting value out of it.

And then there's ways to handle the objections of what if the cost spirals out of control.

Allright, so we've talked about building the machine. When do you hire the humans? When do you start bringing on salespeople? Salespeople are expensive. Uh, you know, they're, that's a big hire. They're, they're, you know, how do you know when the time isright?

It's basically when you're starting to hit, you've gone as far as you can in your sales process, and you're really having a lot of companies saying, I just need to talk to somebody, whether it's, I need a demo, I just need to look someone in the eye and build the trust here.

And you're going to have that as you start to move up market naturally. Bigger companies are going to want to look, shake the hands of a human and have a conversation with them. On that note, I will say, avoid the pull-up market as long as you can resist it, because once you get a big customer, they'll come and they'll say, I can spend a million dollars, ten million dollars with you, and it's very tempting to want to orient your whole business around that.

They will take so many of your resources. As soon as you have this enterprise customer, they will take so many of your legal, security, product, engineering, sales resources. So you want to be really, really picky about who those first big enterprise partners are.

Eventually, you're going to need, most likely, not everyone, but eventually you most likely are going to need some humans to help have these conversations. So who do you hire? You hire someone who is AI native, who really gets AI, who lives it, who can authentically sell it because they're really bought into it.

This is not every sellerright now, but this is more and more becoming the norm with sellers, with account executives. Uh, you want them to practice what they preach, you want them to be efficient, you want them to be bought into the culture of your organization.

Why do you need them? I'm calling this the revenge of the steak dinner. As more and more of this becomes automated, as you have the automated outbound, the automated demo, the automated security checklist, more and more companies are craving in-person times with humans.

So conferences, conference booths work really well, dinners work really well, and they're, by the way, not that expensive to set up a dinner with some buyers, some customers, some prospects. It's a very effective tool. And you do, still at this, at this moment in time, need humans to run those dinners.

Uh, so let's talk a little bit about hiring salespeople and how you pay them. Um, this comes up a lot. At OpenAI, we were notorious because we still have not given out comp plans. Um, it was not necessarily my intention to wait this long.

I just kind of kept kicking the can down the road and saying, eventually I'll pay commission. And then we just, the team got bigger and it got harder and harder. I do think OpenAI is finally about to put in commission.

Uh, but we, we made a, we made it really far without any comp plans. I would not necessarily say that is theright path for every company to follow. I think OpenAI could get away with that because of the equity packages and the way the company was increasing in value.

But in general, it's really hard to hire the best enterprise sellers in the market if you don't pay a variable. That said, you can get away with it for your first few hires. Those first few sales hires are usually wired a little bit differently.

They're there to build the company. They're there for the equity. They want to roll up their sleeves, they want to contribute. They're not necessarily as motivated just by commission and by a W-2. So you can get away with it really for the first two or three sales hires.

Eventually, you're going to want to hire kind of more coin-operated enterprise sellers. They're very motivated by the cash compensation, and so eventually you'll need to put in a comp plan. So just, I get this question a lot, so, uh, just wanted to put it out there on how to think about building out comp plans.

Simpler is better, but have upside for outperformers. Um, so I love this slide. I saw this early in my career in sales, and I just think it's, um, it's a really great reminder of what you're actually selling. You're not selling your tool.

You're not selling your product. You're selling what that tool or that product does to your champion or your buyer. And this is where I do think humans are still very helpful. They're very good at value selling, and I think AI does still have some ways to go on value selling.

So until we totally crack that, I do think we're going to need those steak dinners and we're going to need those humans to sell value. Allright, so that's me. You can grab a one-pager of this with that QR code.

Um, also really easy to find on LinkedIn. I'm the only one with my crazy spelling of a name with A-L-I-I-S-A. Feel free to reach out if you have questions. And it looks like I have a couple minutes for Q&A here.

Oh, we have first questionright here.

Yeah, yeah. Can you hear me?

I can.

Okay. Yeah, I was just curious about, um, you talked about the POCs and pilots should be the exception, but then there's also the self-serve first motion. How do you reconcile that? Because, you know, the POC or pilot then becomes just, do they get help when they go off and do self-serve?

And do they need your help to be successful?

Great question. I am sort of thinking of PLG, so people coming and self-serving and just using your product is very different from a POC, which is usually a contracted period of time where they have access to your product and they have support and resources from your team, and you're giving them sort of services and support along the way.

So you'reright. That is access, like self-serve, access to your product is totally fine and great, and you can stage gate that or limit that with, um, consumption, uh, time box it, um, or, um, uh, feature gate it. And those are all really valid ways to have people sort of build a natural PLG funnel into your product.

When I talk about POCs, I'm, I'm specifically referring to more handheld services, uh, oriented POCs that require a sales engineer or salesperson to sort of supervise.

Hey, Aliisa, Roy, um, quick question on forward deployed engineers. How do you think about them fitting into the sales cycle?

Yeah, I mean, in an ideal world, we wouldn't need them. Uh, eventually, hopefully we get continual learning and these models can just figure out the organization and how to just, uh, immediately become smart. Uh, I do think for now, the thing with a forward deployed engineer is on the one hand, they're expensive, they're hard to hire.

I think there's something like 55,000 job reqs for forward deployed engineersright now and about 5,000 people who know how to do that job. So they're hard to hire. They're, you know, they're, you're getting more and more of them available.

The positive is they make the product really sticky. Because once you've invested in that sort of services element of we're going to have someone come and help you, uh, integrate the product, your customer's unlikely to go anywhere soon because they feel like they've got a customized tool that works uniquely for them.

So I would say, obviously the, in an ideal world, your product works without the forward deployed engineering, uh, because they are very hard to getright now. But the, the positive is then you have a, you have a sticky customer who's with you for a long time.

So if you can, if you can find good forward deployed engineers, they, they certainly do pay dividends.

Yep. Hi Aliisa. Um, for startup working on like their first 10 customers, would you go like the clay route or would you do something else?

Uh, I'm sorry, when you say the clay route, what do you mean specifically?

Like using something like clay for automated outbound.

Got it. No, your first 10 customers are going to be different. Um, they're going to be really more design partners, um, and they're going to need to be much more handheld. So I, I'm talking a little bit more when you're beyond the first sort of design partners and you, you have some product market fit and you're starting to scale the organization.

For those first partners, you want to be very particular and picky about who you choose. On the one hand, it's, it's sort of a gift to a company to let them be your design partner. There's exclusivity to it, there's scarcity to it.

They really help give you a lot of feedback on your product. But you want someone who's going to work well with you, who's going to understand this is realistically where my product is, this is where my enterprise features are.

Um, so you want to be thoughtful about who those first few customers are. One tactic I love to do on the design partner side is go to a company that has a great logo, but work on like their internal product or some sort of product that's not their main production so that they are willing to take a little bit more risks with you and bypass some security processes.

But then you get the logo on your site and you can go to companies and say, hey, I work with this company. And it opens a door to potentially sign a bigger agreement with that company over time. And I think we're at time here.

Sorry.

Are those usually like relationships you already have?

Ideally, they're relationships you already have, either companies you've worked with or your investors are introducing you or they're, you know, friends at, at companies. Because you, you want these to be very friendly, uh, uh, uh, processes and experiments in, in the beginning.

Great. Allright. Thank you.