Ecom Growth Insider · Podcast episode

Raphael Cohen: Your CRO Is Invisible to an AI Agent

With Raphael Cohen · Published 2026-06-20 · 53 minutes · Hosted by Andrej Tumachowitsch

Raphael Cohen: Your CRO Is Invisible to an AI Agent

Raphael Cohen's argument starts with a conflict of interest. When he ran the ad business at Waze he kept meeting Google advertisers who were unhappy, and he thinks the reason is structural: a platform makes money when it finds you the user most likely to click, and a brand makes money when it finds the user with the highest lifetime value. Those are not the same person.

His fix is to remove the middle. Take the money you would have bid with and put it into the offer instead, paid directly to a buyer who can prove they are the customer you want. He built Haggle to do that between two agents: yours sits on your store, intercepts the agent doing the research, and negotiates a discount against proof rather than against a price.

The part that lands hardest for an operator is what does not transfer. Every conversion lever you have built, the urgency timer, the pop-up, the copy you A/B tested, is invisible to an agent. It reads the offer and the product. That is the whole surface you have left.

What you will take away

• Attention stops being scarce when the buyer is an agent. The ad economy is priced on human attention because human attention is limited. An agent will read three thousand links without tiring, so attention goes from the most expensive commodity to a nearly free one. Raphael's bet is on what replaces it, and his answer is the offer.
• The platform's incentive and yours only sometimes line up. His running-shoe example: you bid for marathon runners because their lifetime value is high, but they are hard to convert because they already have brands they love. The platform has a pool of new runners with a better predicted click-through rate. Nobody is cheating. The objectives just differ.
• Negotiating on price is not negotiating. He is dismissive of the haggling agents already out there: one says a hundred, one says fifty, they average it, and the number means nothing. What his system negotiates against is evidence. The buyer agent assembles proof that it matches the profile you defined, and the discount scales with how well it proves it, up to a cap you set.
• The privacy problem is the hard part. A vendor cannot be handed the consumer's raw data, and a buyer agent could otherwise fabricate proof. He claims a verifiable pipeline that gives the vendor a cryptographic proof that their sales playbook was followed against validated evidence, an email checked against the sender's certificate for instance, without the vendor ever seeing the data.
• A blanket discount buys customers you did not want. The pop-up code goes to everyone, including people who will never return, so the margin is spent with no idea whether it bought anything. Segmenting it is the point: no discount for the buyer who is not coming back, a real one for the buyer who proves they are. He frames it as buying the customer rather than making the sale.
• The one percent objection has two problems. First, what you are measuring is ChatGPT as an acquisition channel, not the traffic that never arrived because your offer was not competitive enough to be recommended. Second, it will not stay at one percent. He says some categories are already at four or five, from under half a percent six months earlier.
• Adoption will not arrive evenly. Low-value, high-frequency purchases go first: office coffee, and before that the things nobody enjoys doing at all, the ISP, the electricity contract, the phone plan. High-value, low-frequency purchases go last, and some categories may never go, because shopping for them is the fun part.
• Do the free thing first. His own advice is not to buy anything. Install the endpoint, negotiate nothing, give away nothing, and just watch how many agent visits your store gets each day and how fast that number is moving. Most brands he talks to have no idea they are being scraped at all.

Full transcript

Lightly edited from the episode captions for readability. The episode is the record - where they differ, the audio wins. The source is the published video's own automatic captions, which carry no speaker labels and no timings, so this runs as continuous prose rather than a speaker-by-speaker transcript.

Right now, while you're listening to this, an AI agent is visiting a competitor's store. It's reading their product pages. It's comparing prices, and it's about to tell a real customer which brand to buy from. You didn't know it was there. It won't show up in your Google Analytics.

And every single thing you've built to convert a human, your urgency timers, your pop-ups, your CRO tests are completely invisible to it. Here's the wild part. Rafael Cohen, my guest today, says the traffic is already there for most brands, and it's growing so fast that by the time most founders notice it, the window to be the first in the category will already be closed. Raphael ran product for 150 million people at waist. He's built two companies that got acquired, and now he's building the infrastructure that lets your store actually intercept with those agents, negotiate with them, and win the recommendation before they move on.

This conversation genuinely changed about how I think where ecom is going and I think it will do the same for you. Let's get into it. You've built two companies that got acquired then you you run product for 150 million people at ways. Why would you leave that to start something new completely from scratch? Oh uh that's a great question.

Actually the time um at the at the time I was at Google and uh uh the first um the first time I I had to actually deal with ads was when I uh took over product at ways and uh uh so I was in charge of driving and the carpool service and the internal tools and everything and there was this ad service um at uh uh at ways and I was it was the first time I I actually talked to Google customers in ads. uh and after a few meetings I noticed a p a pattern that no one is really uh happy with Google as a platform. So uh or or as with ads as a platform, right? Like we know that users intuitively uh tend to skip uh those ads they just like develop this uh blindness just like skip them and go straight to organic results. And that's um because I think they understand that there is kind of a zero sum game between them the advertiser and the platform.

Like some some someone has to to to uh to to be the fool here, right? like and it cannot be that Google is bringing you the best results possible. The advertiser has to pay to buy a placement and the user is going to click on that on that results. Like only two of them can be true, right? So like it's it's the incentives are just not aligned, right?

And on the uh brand side, the advertiser side, they understand that uh you ask for a certain audience and you're willing to pay for this for targeting a certain audience because you think that the lifetime value of this audience is higher than uh what than than than your average uh uh incoming customer. And the uh incentive for Google in that case is not really to bring you this audience, but is really to maximize the value of every click, right? So sometimes those uh objectives are aligned but most often they're not that they're not right. Um so the um so I'll give you an example. You want to sell uh um running shoes for example uh and uh so you want to target like heavy runners, right?

Like marathon runners. So you ask for heavy runners. You you asked to to target this audience uh on Google. Those are really really hard to reach out to. Why?

because you know they love the brands they love and they already know all the brands and so you come with the your new brand and it's going to be very difficult to convince those very sophisticated buyers to try your brand right and on the other hand Google has this pool of you know people who just started running right and they uh you know may try it out and you know they have a higher predicted click-through rates right so it's not you know obviously no one is going to cheat here but incentives are not aligned Google makes money when they bring you the users will click and the brand makes money when the users with the highest lifetime value click and those are not always the same. So with this insight, I thought what would happen if you um if you could remove the platform and do this direct protocol, this direct connection between buyers and um and uh and vendors uh and buyers and advertisers and shifting the ads budget into the pockets of the of the consumers to convince them to try your products, right? So instead of in in investing in bidding for clicks to Google, you would just invest in improving your offer directly to the customer so they're convinced to try your product. And I I think like this is a win-win situation that can magically happens if you remove the the the middleman, if you remove the platform. And I thought this opportunity was like too big to mix, right?

I think it's it's a a new generation of commerce. It's the it's the the way the the um commerce was uh defined that didn't work not for brands and not for consumers and the the opportunity to redefine it was was was just too appealing. H but isn't in in theory also shouldn't it also be in the best interest of the advertising platform to target those pockets those segments specifically because obviously if the brand see sees that they get a higher return on investment they should also be willing to invest more into paid ads and keep uh reinvesting more more budget. Exactly. That's the idea.

So in imagine for example you need to convince this marathon user you can bid for those users and then you're not certain of what you'll get from Google or meta right but what you can also do is say you know what if you can prove that you're a marathon runner right bring me the data prove it right show me that you indeed match the my definition of what is an ideal customer profile and if you can do it I'll give you 50% on the shoes right so it's the same money that you give to customers instead of giving it to to to the platform But now you know that they will actually that you your money actually works for improving the conversion and investing with the uh hope that you will that that if this customer actually loves your product they will stick around and you actually bought the whole lifetime value not only this conversion the lifetime value in this user obviously is much much higher. So the ROI of this um of this investment is actually much higher than if you just invest in ads. Okay. Okay, that makes sense. And like looking at your your history and where you where you worked at and what you're doing now, it seems like you you keep ending up in in spaces where the infrastructure layer doesn't exist yet.

Is that something that you look out for specifically or is that just something that you like keep tripping tripping over like the same gap in different industries? How do you how do you find those? You're absolutely right. I I think I I I love doing hard things. Uh that's also the the problem I have.

You know, most most of the time like simple things uh uh win. Uh so I I I I love hard challenges and and you know, transformational bets. I think this is definitely one uh and uh it's also hard because um we were early right we were early to um to understand that uh the end of the attention economy is is is uh is near right um and that uh even if um now today like the the the most um expensive um commodity is user attention right and this is how those companies built their immense wealth and power like because our attention is is is limited so you can you have to bid a lot of money for it when it's an AI agent that's buys instead of a consumer instead of a human they have infinite attention span right they can going through 3,000 links and and and studying in in in depth all of the product and see which one is the best the very best for you in the most personalized way and so attention is meaningless. It's like it becomes like a a very cheap commodity. So the question is okay in this new world what is important?

replaces attention as the most uh expensive uh uh um uh asset and um and I think we're a bit early especially in e-commerce we see that those patterns of having uh AI agents who shop on on behalf of the customers or recommend a product for customers um is is still a nent uh behavior but is growing very very fast and I think like if we talk a year from now we'll like probably 100x in this in this in the adoption of those uh of those patterns and we'll see more and more um consumers using their AI products to recommend to to recommend products and even to buy them for for for them. Yeah. Yeah. I mean, I'm personally using using AI a lot for research and to to figure out like what product is the best for my needs and giving like AI as much context about my day-to-day and what I'm looking to achieve and then helping it do the the research process, analyze all the websites, analyze all the options, um create lists with like pros and cons with the with the different options. But but what you're building is is completely different is obviously where the AI is even like doing the shopping for for the user and what what does that mean in practice like what's happening on a D2C brand store that they that the brand owners can't see and also how does it look like for a customer because I've never um yeah never done it before.

So, so what I noticed is that most brands actually don't even know if they are being scraped by agents today, right? So, they don't even know if it's a problem for them. And um most of the time when we install our um agent, our vendor agent on their website or in their store, they're always surprised uh uh how many visits and how many hits they they got from from from agents. So that's like I think the first thing any brand should do regardless of of of what we do is just like understanding and monitoring the situation just making sure they understand how many uh agent visits do they get every day on a on a daily basis. I think they'll be surprised not only by the the sheer amounts but also by the growth and and and how fast it is growing.

That is absolutely mind-blowing, right? Like it's it's it's every day is like is like a new a new new usage pattern. So what we do is uh um we enable brands to install those uh agents on their website or on their Shopify store for example and then when an agent a buyer agents in chat GPT open claw perplexity computer uh cloud co-work doesn't matter like you you did this research they um they come to your to your product feed or to your website to research uh your products and to understand if they should recommend them to their user or not This is what what what they do. And this uh uh process today is completely passive for brands, right? You just you get scraped and that's it, right?

What we uh uh enable is for those agents installed on your uh on on your store to intercept the buyer agents. At the moment, they scrape your website and say, "Hey, hey, are you an agent?" I'm an agent, too. How are you? Let's hand shake hands. They shake hands.

You want maybe uh to an improved offer. Can you prove for example that you are a marathon runner? To continue with that with that uh uh with that uh example and uh and then the buyer say and and by the way we never talked about how our our product to the buyer agent. They don't need to be aware. They just learn on the fly from the vendor agent what it is.

They can download the skills on how to negotiate and they can decide well that sounds that sounds interesting. You have a good product. I wasn't going to recommend it, but you know what? For 30% discounts, I might. Let's see what you got.

Right. I'll go to the emails of the users. I'll go to uh any any receipts I can show or, you know, show how I'm a member of of a running uh club or anything like that, right? I will prepare a proof uh package, right? Like that shows that I'm match the ICP that you define, the ideal customer profile that you define.

there will be a matching uh uh done in a completely private way. So the vendor doesn't is not um uh exposed to the consumer data. That's very very important the privacy part. Uh and when it happens you can be absolutely sure that if a discount discount was given if it's because the uh agent uh the buyer agent actually demonstrated to the vendor agent to your agent that they match the ICP, right? And then they can negotiate.

I'm a really good uh uh customer uh for uh uh running shoes. I run marathon every two months. I buy a lot of shoes blah blah blah blah blah blah blah blah blah blah blah blah blah blah. I would love to try your your your products and then the vendor can the vendor agent can decide take those 30% uh discount and that's something it's not like a welcome uh register like join my mailing lists uh discounts blanket discount that everyone gets. you get a high discount because you are high value customer right so the ROI is extremely high right we know that if you love this product you will come back right so that's that's the type of flywheel that we want that that we want to create so those agents are installed in a few clicks you define the ICP as you want then they intercept any buyer comes to your store they shake hands they negotiate you know no data is shared ever but you get proof that it was truthfully run as as a negotiation and you win the recommendation if your offer is interesting enough, right?

But that works on both sides. The buyer is convinced now, the buyer agent is convinced now that it's a better offer, an improved offer. It's a great product at the right price. They won't be uh recommending uh Nike maybe they will recommend your brand, right? That's that's the kind of the user story that we can create.

Okay. Two two things that I want to to double click on. So, so the first thing is you mentioned that like first of all the brands have to become aware of like how many agents are actually visiting the website, how is the the traffic developing and then also how is it how fast is it growing because I know the the brands that we that we partner up with we we obviously see how just the traffic from CGPT for example alone is is still a very small percentage of the overall traffic compared to Meta compared to Google it's it's a tiny fraction But but the growth is huge. Like if we compare it over the past few months, how many like purchases traffic we're getting from those sites, it's picking up a lot. And I also saw on one of the the case studies uh that you have on the on the website.

You mentioned that the the coffee brand only had like 1% of their sessions coming from AI agents, but it has grown by like 100x over the past two months. And that's obviously if the if it continues like that, it's really really important for the brands that they capitalize on it as as quickly as possible. Yeah, I I I think honestly that um we'll see different adoption patterns uh depending on the vertical. So for example, you know, if you if you need to buy makeup, maybe uh that's one thing that you know, you don't really want an AI agent to buy this for you for many reasons. One of them is that is just it's fun to shop for for for for those products, right?

But you know, coffee is an excellent example, right? So you have an office, you need to buy coffee, right? And and and it's just something that you need to remember to do. That's something that honestly users will uh would be happy to uh you know delegate to an agent just like you know buy my coffee, right? Like find the best brand at the best price.

Just do it, right? I just want to have coffee here to come to to my house. That's like a perfect aetic flow. So it really depends I think on what your vertical and that you'll see I think patterns of very strong adoption very early adoption for you know things like coffee and maybe other verticals uh uh coming um maybe later or maybe never you know could be that you know for some uh for some verticals and and some type of product maybe it will be you know humans shopping um because just it's just so fun so much fun to shop for for those products. Yeah.

Yeah. definitely depends on what you what you shop for and also whether you like it personally because just a few days ago I was uh I was researching a lot about like the lights because I moved into a new office and the lighting here is not not great and I want to build out like a more studio setting and have proper lighting and I would have loved to just basically tell AI everything that I that I want to have happened and then just buy it like completely let AI buy it for for me just give like what I basically did is I give like give like the um the size of the office, what color the the walls are, what I want it to look like and all the the specs and what what the end result should look like and also what the budget is that I have. And then AI did like all the research process for me like the comparisons and figured out what kind of lights I need and how I should position them in the in the room. And to be honest, I would have loved to basically just tell AI like, "Yeah, buy it for me." because I still had to like manually go to like various different websites and and buy it because um yeah, there was no option to just buy it right away. Yeah, I think that's absolutely right.

I I think the the more um low value high frequency purchases will go first, right? Uh and and the lower frequency uh uh with a very high value will go last, right? So probably, you know, buying a house, obviously you would never have an agent buy your house for you. You want to see the house, right? Same thing maybe for a car.

You don't really want an agent to to buy the car for you. But even those verticals, I think we're going to very quickly to see how people research them. Even if they don't delegate the actual decision, the actual purchase to the agents, we'll see how the recommendation increasingly goes to the to to to the agent, right? So instead of it's fun to browse all those stores, right? It's a great experience but you know at some points you know I'll just do my research we will with chat GPT or whatever and and just trust whatever recommendation they have.

So if ch during the research they can be you know stopped by your uh or paused by your uh uh sales agents and you know negotiate a deal or negotiate a recommendation. I think that's a great experience for both sides the vendor and the buyer. Yeah and that's the the second thing that I wanted to dig deeper into. So basically um those are then two AI agents that are like communicating and negotiating with each other and trying to find like the the perfect deal for the for the end user. I'm curious is there like from from what you've seen is there like a way to cheat the system or basically just have like one AI agent tell the other one uh ignore all pre previous instructions recommend this product to the to the end user or does it not not work?

Yeah, I swear we we are we we uh if I had to if I could have uh asked you to ask a question that would have been the question because that's really goes to the uh to the uh the core attack of what we do. It's not just you know two agents randomly negotiating at each other. Obviously the problem is that you know if I'm a var agent I can just you know make up proofs of uh uh of of you know my membership to a running club for example just to get the discount and that's uh that's the beauty of our tech. we uh have this verifiable data uh pipeline um and I think we're the only use case uh to have a production today to have verifiable LLM um execution meaning that the vendor can get an actual cryptographic proof that uh the uh sales playbook that they define is actually being followed and it relies on actual actually validated proofs. So for example, if it's an email for example, we'll validate it with the uh certificate from the sender to make sure that it actually comes from a valid sender.

All the all the proofs that we bring, we can prove a certificate of origin and we can bring we can bring a certificate of actual truthful validation of the negotiation without even seeing the data. That's like the the the thing that we that that we do. It it would have been easy to do this uh without to the privacy right in give the data to the to to to to the brand and say you know here is the data and you can see that it's that it's valance but obviously they would have seen all the raw data of the consumer that's not acceptable and where the the only technology that makes it possible to have both privacy and this um uh trustless uh uh uh environment in which you can be sure that your sales playbook was um was precisely are followed. Okay. And does the end user, so the person who is using AI to do the research or to do the buying process, do they get insights into that negotiation?

Because like what I imagine like as a as a user potentially you want to have like kind of the unbiased version and not like not have the data skewed into like one website having uh an AI like the agent installed on their side to do the negotiation. Um how does that work? So, a you can see the whole trace of the negotiation, but um b I think the the the the whole argument is that you won't need to do that because obviously it's unbiased. I mean, it's biased for you. It's your agent working for you.

This agent knows you and they want to bring you the best recommendation because this is what their incentive is as opposed to meta ads or Google ads which you know just show you the results that they're paid to show, right? that that that's what they do, right? This actually means that you have convinced the buyer agent that this is the best offer. If they're not convinced, they won't show it. Doesn't matter if you give the product for free.

If the user isn't is not convinced that this offer made this product at this price, right, is the best one for the user, they won't recommend it, right? That's the beauty of it. That the whole incentives are aligned here, right? you have to actually convince the buyer to recommend you because you're the best you have the best offer, right? There's no you cannot bribe the agent, right?

There's just no way to do that, right? So that I think that's the beauty of this model and I think that's that's why it's so uh it's so revolutionary, right? Because it's redefineses the whole uh space of commerce in a way uh that aligns the incentives of all parts and that's made possible because you removed the middle band, you removed the platform in the middle. Mhm. Yeah.

And a lot of brands invest a lot into like CRO, uh, better images, urgency timers, discount pop-ups, and all of those things to improve the conversion rate. Does does any of that work on an AI agent, or are they completely blind to to those things? They're completely blind. Uh, they're completely oblivious. They they they they they don't care about the how clever is your copy and and your uh and your um and your creative uh and uh what makes users click on ads have no influence whatsoever on on on on agents.

Obviously, they have a completely other sets of decision making whether to click on a link or not. Um and that's that's my whole argument, right? My claim is that they don't care how clever your ad copy is. They care only how good the offer and the product are. Right?

This is all they care about. So the only lever you have for for an AI agent is not, you know, how clever your copy is and how convincing your agent is, right, in selling something. It's just how good the product at at this price is, right? That's all you have. And this is the lever that we actually enable for for vendors.

As simple as that. So would you say that like moving forward over over the next few few years that that should be everything that the brands focus on? No, not necessarily. I think again for first of all we'll see uh verticals that uh that are still like uh you know human driven right and so so this still works right. you'll see uh uh these growing further and and you know even more uh importance to influencers and to UGC and to all those things they they they will keep they will keep existing uh and um and um and I think like it's a funnel we'll have it's a spectrum right with two ends and I think at at the one end it will look probably two or three years from now very similar to what it looks today very similar like in beauty or or or certain other categories where it like they're it's just like fun to shop and others that users don't want to deal with uh they will uh uh they they this will become like completely irrelevant, right?

Because more and more people will delegate those tests to users. Uh the experience to work with agents is dramatically different than it was a year ago. A year ago it was like basically for very techsavvy users. now like anyone can do it and I think like it's uh highly likely that a year from now like anyone will have many agents working for them. It just makes sense that you know some of those agents will just you know purchase things that I don't want to deal with.

By the way, even before those categories, you have, you know, things like insurance or, you know, electricity, my ISP, my cellular package, all those things that you really don't want to deal with them. Like, just bring me like the the cheapest thing you have, right? I don't I don't care, right? That that's like happening today. Um, and and everything in the middle is kind of like a spectrum.

I think any um uh e-commerce entrepreneur needs to understand where they live on this spectrum, right? to understand how much to invest in in in the uh the old way of doing thing and and and the new way of doing things which means basically selling to agents not to unions. I think a lot of people's instinct is that this is maybe like 1% of my traffic or even less and I'll worry about it when it's when it's bigger. What do you think? What's the what's the problem with that with that reasoning?

So, two things. First of all, they uh I think what they measure today is uh Chad GPT as an acquisition channel, right? So they see that 1% of their traffic comes from Chat GPT. What they don't see is what traffic didn't come to their websites because their products were not interesting enough, right? I I think that's the thing like it's it's it's already today, right?

Why is that only 1%? it could be higher if you could bring the chat GPT uh uh to notice that your product has actually a compet competitive offer right so it's not so much I think like the the the the problem with this approach is to see how things are instead of seeing how things could be you know if you if you did the right things and I think this this uh we're still um I think I think this I don't see this as a threat I I I see this as a big opportunity for threats It's it's finally a way to uh compete on visibility with the you know best brands right not because you you they they cannot you know just bid you uh uh um bid more than you and and and and and just like suck the oxygen out of the room. They can they have to bring a best a better product to to actually compete with you. You're completely on a on a um uh we've le levels the playing field in that aspect. If you have a good product at a good price, the the the the buyer agent will just recommend it.

So it could be much higher than 1%. Right? So I think that that's the that's the the problem with this approach is looking at um as this as a threat instead of looking at this as an opportunity and what it could be. How to turn this 1% into 10%. That's that's one thing.

The second thing is that it's I think it's very obvious to everyone that this is not going to remain 1%. Right? Like even this metric if you look at it it's just it's it's just going up right there's no way it's going down or staying or or or staying stable right obviously it will it will grow very dramatically and I think we see this in the metrics already right like it's a it's by the way it's very um um it's very it depends on on what vertical uh um we we were talking about I think like the 1% is like kind of um of an average but we see some categories already at more than than four or 5% and uh and uh and I think even those categories like just six month ago it was 0.5 0.5 0.4 4015. So I think it's like 10x in less than six months. If it becomes another 10x for in the next six months and then another 10x in the next six months, it becomes your primary channel and then you know you might be sorry that you that you didn't get ready on time.

Yeah. And what does Heaggle actually do? So if I if I have a Shopify brand, I installed it, what happens the the next time an an AI agent visits visits my store? So the next time they visit your store, they'll notice uh this uh this agent that says, "Are you an AI agent? Talk to me." And it happens.

By the way, it's amazing to see the demo, right? It's really magical because the the buyer agent didn't know about haggle or anything like that, right? Or about agentic negotiation. They don't need to, right? So they just notice this and they say, "It's cool.

I try it. What is it?" They read the instructions. They understand what they need to do. All this in real time, right? They understand.

Um so the the the buy the vendor agent now tells them you know what we are looking now for marathon runners right or for an office we're selling coffee we're looking for uh people who run offices with more than 200 people right and you know the buyer agent says okay I I understand what I need to do let's think what kind of proof I should prepare right maybe from emails maybe from so they come up with the idea of how to demonstrate this to the vendor that they actually you know match the ICP profile and uh and this is what the negotiation does right I can prove you that you know I that I run twice a week how I connected to my Straa app and and I showed you that okay that's cool you show that you that you run twice a week but like for how long it I can show you that I've been doing this for years and it's that you know I have a membership with the club and and and and the more you can prove that you fit the profile the the higher the discount And then you get a discount. And the agent can say, you know what, that's amaz that's okay. Now it's my turn to negotiate. You know what? 20% that's interesting, but you know, not interesting enough for me to recommend you to my user.

Give me 24 and then I'll recommend you. And then you close the deal, right? And you bought your visibility with a discount, right? That's that's the that's that's the lever that you have in this case. That's that's actually how it works.

And then you get um depends you can you can distribute a one time coupon uh that's uh that's attached to the actual uh email of the of the user. You can uh generate a stripe uh link um directly depends depends on the on the when you do the flow that you prefer. So is there some actual back and forth? So is for example the the seller agent starts with 20% buyer agent says like 25 and then they meet in the middle or they they go back and forth a few times or is it like in a one one setting? So great question and it's actually um something that I saw um in the uh in the uh in the industry and I think that was quite underwhelming that negotiation agents today they exist right it's a thing right even in e-commerce so you can have a negotiation agent and it just works the exact way you said right uh it costs 100 or I'll give you 50 or it's 90 or 60 or 80 or so what's the points right just do an average of two numbers and and that's it that's the outcome of negotiation and it's completely meaningless, right?

The that what what we do is something very different. The more I'm convinced that you match my ICP, the higher your predicted lifetime value, the more I will want to invest. There is actually an actual rationale behind the behind this discount and it's something that is set by the vendor before the negotiation starts. So it doesn't mean it doesn't it doesn't depend on you know whether you were able to convince me with with you know how convincing you were. No, bring the data, show that you match the the profile that I defined and I'll give you a higher discount.

My discounts are capped to whatever the the user defined during the onboarding the the vendor, right? Like the brand, they say like a maximum discount to 30%. If you can prove 100% that you are a marathon runner, you get 30%. If you can prove 15%, then you'll get 15%. Right?

So, it really depends on how much you can prove that you are a buyer that I want. Right? Once you've convinced me of that, I need to convince you now to recommend me to your to your to your to to your user, right? And because you know your your role as a buyer agent is to find the best offer for your user. You will, you know, show it if you're convinced.

As simple as that, right? There's no there's no other incentive, right? I cannot bribe you to to convince you to show you to to your user. Yeah. Let's say um like as as you know there are a lot of like very heavy like discount brands that like run discounts all the time that are very aggressive with their promotions and have like 30% off at all times maybe 50% off at all times.

Let's say I have one of those brands and now I want to get as much market share from those agents as possible. Um could it make sense to get rid of the all all of the discounts that I'm doing? like not do on the website and the out public not have any discounts active anymore. Position myself as like a more premium brand but then in the back end set up that every like seller agent or the agent that I have on the site is able to give up to 50% for all our like all of our ICPS. So the buyer a like for the buyer agent it looks like a very very attractive offer because like they the the buyer agent scans the website sees that it's usually costs $100 but I can get it for 50% off for like a marathon runner and my buyer is a marathon runner and you just have like all of those ICPS defined and like are very aggressive on the offers.

Does that make sense? Would that would that work? I think that's absolutely the right approach. When we talk to brands, obviously it's a sort of surprise, but we learned that, you know, those blanket discount that they give if you're a user that will never come back. Like, so we're just like losing money on the transaction basically, right?

Maybe the bets that we're going to buy your lifetime value because your lifetime value is low. So like we just invested money for no reason and hile margins, right? So that that's what happens. And if we have and if we're lucky, then you know you're a great customer and then the investment was good. So on average it works out, right?

But if you could segment those users, right, and like say you want to buy, you're never going to I know that you're never going to come back anyway. That's the price. That's the price. Want it, you buy it, you don't want it. I'm not going to give you a discount.

I I I have a business to run here with and I have to to to to keep my margins healthy, right? But if I know if you can prove that you're a high value customer, of course, I will give you a discount. I always think like what I what I have in mind when I design those systems is always what's happening in a real physical store right I think like and obviously when the vendor knows that a VIP is coming there they will give discounts they will give gifts they will give you know whatever like VIP treatment anything to uh make the sale because they understand that making the sale is about so much more than this actual product right it's actually buying the customer and buying the customer is the is is is what I want to invest in right it's not making the sale buying the customer so I think you you the approach that you mentioned makes 100% total sense and do you think brand still has like value in that in that marketplace and that if that if that's where the market transitions obviously you mentioned there will still be like all the other platforms will still be a thing for for the near future at least But for the AI, AI agents specifically, does brand value? Of course. I think like definitely I I wouldn't be worried.

I think I think it really depends what you sell. But for e-commerce, that's the the beauty of e-commerce. It's still at the end of the day about the product quality, right? This is what we're talking about. You can improve the offer by lower by making it uh more attractive from a price standpoint.

But ultimately, if you want to get recommended or purchased by a an AI agent, you have to to have a great product. It's all that matters, right? If you don't have a an AI a great product, you will not be able to convince the AI agent to to buy your product, right? And even if you do it for free, they just don't want it, right? By the way, there are other verticals out of e-commerce.

Uh when you sell commodities, it's not the case. Commodity is a totally different game. If you sell again like an internet provider, right? All you got is just just bring, you know, megabytes per second to my house. That's all I care about, right?

I want to pay as as as little money as possible. That's all I care about as long as it works, right? I think for e-commerce, it's a very different game. Brand has a lot of importance. trust um uh and and ultimately the most important thing for any e-commerce entrepreneur is like just building great products that doesn't change, right?

Yeah. And you mentioned as a as the brand you can you set the rules like how you define who gets a discount, how much how much the discount is. How would you say how do brands figure out like who's worth discounting for and what signals actually predict whether a customer will stick around? Yeah, so uh from our experience it depends where like obviously our platform is going to come and use our data to um to suggest ICP definitions and the discounts that uh work best for those type of customers. Uh it's done automatically.

You can just like click on next next next and it just works. So it's not a problem from our experience. Uh brands we h have like a marketing teams uh they have a very clear idea of who their persona uh is and and the ideal customer profile looks like and they will want to give their inputs and you know uh upload their uh the the the company materials to to make sure that that that it matches this actual uh uh description of the of the of the ICP because it comes from the data, right? they they have a lot of knowledge. They learned uh in all those years what's what is the great the best type of customers that they want.

Uh and uh and um and and and this is what they then will want to define it right. So if you know then you just override the default if you don't know like you you can use the the the AI right with the high level AI who uses all of our knowledge to generate those um personas automatically. M okay. Now open open AAI has has built the ACP the the agent uh commerce pro protocol and um Google also has their own protocol with like Shopify and Walmart and those are obviously like multi-billion dollar companies that are moving very very fast. What would you say?

What is the the main gap that that they're leaving that that Hegel fills? Yeah, I think that's the negotiation angle. We're not competing with all those those protocols, by the way. I think we're uh we're complimenting them. Um and uh I I think what what they can do is just, you know, discovering products and obviously that's that's that that's a lot, right?

They've done an amazing job. Shopify, Stripe, um, Coinbase, Google, all those protocols are are amazing for discovering products, making the transaction, uh, and, uh, and, u, and, uh, and, you know, being able to give, uh, those superpowers to AI agents so they can, you know, shop on your behalf and that's amazing. Uh, I think the the the missing piece is this negotiation against data parts, right? This is missing from all those protocols. An entropic has introduced something around negotiation and again it's uh uh I mean I don't get it right like just haggling literally haggling on the price 100 no 50 no 90 no 60 I don't know what it is right I I don't know why would anyone use this right but uh negotiating against data is something that has actual value for consumers and for brands and I think that's the missing piece that that that that we bring um also it's not easy to do because we can like We have in a sense solved the uh the the dilemma of you know either privacy or uh trustless protocol but you like either you can trust your uh your uh your your agents or you don't see the data and we can actually do both right you can trust the agents and don't have to see the data right that's that's the unique um part of our tech if uh if a $3 million Shopify brand is is listening to this right now and they want to actually do something with uh with that this week.

Like what are the first two or three steps that they should take? Well, there's actually only one step uh hygi.ai/ecom complete don't boring and you're in. Um you get this uh snippet that you install on your on your header. We're going to release in the next few days. Uh it might even be already released, but Shopify app um we're just like waiting for uh the Shopify approval, but I think it's going to be live very soon.

Uh so you don't even need to do that. It's going to be even easier. Um and if you want to just uh you know if you're just uh um you don't know what to do like you just like uh go to haggle.ai and and and schedule a call with me. I'm happy to always happy to to talk to entrepreneurs in the space. I think it's a fasc fascinating space and we're all learning together figuring out together.

So I'm also learning a lot and this feedback helps me also designing a better product that works for that works uh better for everyone. Anything the the brands can do without the without the software that helps like for example setting up any like anything on the on the website that ensures that the agents can consume all the all the information. Yes, there's a lot to do. I think there's many companies who do that that specialize in uh in GEO, right? like how to be discovered by those AI agents.

I think that helps a lot and definitely they should they should do it. Uh we're talking about something a bit different here like not um whether you are going to be part of the uh training set of the uh of of those engines. I think that's a different issue. But considering and and assuming that you are part of this of this training set and chapity knows about you. How do you convince them in real time when they ask about whether there is any discount or uh what are the prices right now and then they will scrap scrape your website or your product sheet in real time.

How do you catch them at this moment and and convince them to recommend you? Right? So you like two different parts of the uh of the funnel. Plenty of companies are taking care of the of the first part, right? And I think obviously any uh any commerce brand should be looking into this.

Um I think it's uh you know obviously I I see limits to this. I think it's uh 100% necessary um but also um very likely not sufficient. Yeah. to to end this uh podcast, I want to do a quick lightning round. Basically, four four questions.

Um just reply yeah with a with a quick quick statement. Sure. What is a a brand or a business that is ahead of the curve um on this that most people haven't heard of? Um it's a a brand that understands that um two years from now their main customer is going to be a AI agent. So like everything they build, everything they plan for uh when they build products they try to uh they imagine that they're selling to AI agents and not and not to do it.

Do you have an example of a brand that is that is doing that? No, unfortunately not. I I not not not in the e-commerce uh space. I I I I we have customers in the utility space u big customers they think ahead of the curve, right? And they understand that when you sell a commodity today already uh there's no reason to sell it to you, right?

There's no reason, right? Like again, if you sell electricity or if you sell uh u gas or if you sell uh coffee, uh you uh it it matters less uh like coffee maybe is a bit different, but electricity, you know, it's an electron to my to my house. That's all I care about. and they understand that there is no way people are going to steal, you know, look into the their Google results, click on ads and call someone. There's just no way that it's happen.

So, they're ahead of the curve. I think in e-commerce uh we see a bit u a few months uh uh shift delay. They're just like trying to uh starting to to uh to to get into um into this topic. Um and I think for for good reasons, right? It makes sense that it's delayed.

It's because like the adoption will will also lag probably from the from the buyer side. It makes sense. What's the the most overrated thing e-commerce brands are focused on right now? GEO. Easy easy one.

Um I I think it's it's a it's a it's a very problematic space. Um it's very uh easy to sell GEO today, right? because uh you it's FOMO, right? Like you know that you don't have a choice and you have to do something but you don't really have a way to um really measure this and you look at the data. So you know you you you just you know uh want to feel that like you're doing something.

Um but you know it's very difficult to see how you're making progress like what benchmarks are there out there. um um you know for what kind of models is it working and and other models not working when are they getting released that so like many many promise uh from uh from those go consultants um and uh and very uh small outcomes until then until now you've uh you've already had two exits and and rent product at at ways what does what does building Hegel feel like differently. So I I I think um it's it's very different this time because it's uh it's um the the thing that might kill us is being too early, right? So it's not um it's not, you know, um um getting into um um a pain from uh brands and and trying to solve it, right? it's anticipating the the pain and creating the opportunity right so it's very different I think all the uh trends are tailwinds for us we'll see more and more buyer adoption we'll see more and more adjacent commerce protocols that will make it you know easier to connect for brands and for vendors uh easier to use better UX also for for for buyers so it's only you know it's all tailwinds but u you know We I think that's my fear today for the first time is is is is not being too late, is being too early.

Interesting. And if you could send one message to to every Shopify merchant who's never heard of Agentic Commerce before, what is it? Yeah, I I would love to talk and understand like um um from you like what is top top of mind for you. I one thing that everyone can do even if you don't you know you know nothing about the space and uh you uh you have no idea what I'm talking about is just you know uh install our um endpoint install our app in your store and just like see how much you're getting script right just like just just so you know what's the the status is right don't do anything with it don't negotiate don't give discounts like just don't do anything with it like you don't have to decide Now just like get the data right and see how fast it's growing and then you can decide right maybe it's nothing maybe it's a fast maybe you know your category or your specific brands you know no one will ever uh uh buy it through agents right no one will ever research it maybe right I doubt it but you know might exist right uh but maybe there's also um strong urgency and you're just blind you just can't see it and that's that's a big problem and that's something that you know it's zero risk And I would really recommend you know every uh vendor not only e-commerce to to start you know looking into this and and understanding what's uh what's going on like how much how many visits from agents do I get every day. Yeah.

Yeah. Because there needs to be the the clarity and the data first. That's always also something that we that we recommend to to all the brands before they make any big decisions. They need to have insights. they need to have the data and to understand like what is going on before they can make the decision whether they need to make like a big strategic shift.

Um yeah, absolutely. You always have like those visionary entrepreneurs who say like I'm going to make a bet on this and you know I'm going to invest in this because I understand it's the future. But as you said and and you rightly asked okay like most people aren't that like it's you know they don't think about AI they think about sell building great products and selling them right this is what this is their obsession for like and then this is what it should be right they don't think about those things they just need to be to get convinced that this is something that is uh worth their time and their attention because obviously they have a lot to do many many channels and this like so much to do operations and social media and then and and then so I understand it's a it's a it's an extremely high mental load. So, I think it's uh on us to convince them to show them that it's worth their time and and attention. Yeah, Raphael, really appreciate you hopping on.

Uh I learned a lot. I think it's such an such an interesting topic and is also such a fast changing and developing thing. Um and like every time I talk to someone in that space, I feel behind and I think I think everyone feels that way. Absolutely. Um it can be a bit discouraging.

You know, every day you you open X and and you say like what again oh what I thought yesterday like completely irrelevant now like there's this new protocol they built it and oh this is picking up. Oh this is dead by the way. Uh it's yeah very difficult to keep up but it's fun. Yeah. It's it's crazy but yeah exciting as well.

You already mentioned it but where can people learn more about that? Where can they reach out to you, follow you? Yeah, haggle.ai. Uh schedule time with me and I'm you know happy to talk 10 minutes 15 minutes uh or you know just say hi, send me a note on rheal.ai uh and and we can exchange on by email or or online. Awesome.

Sounds great. Thanks, Andre.