The Agent-Native Merchant Lifecycle: A Playbook for Shopify App Partners

Sakshi: Hello everyone, I'm pretty sure most of you don't know me. I'm Sakshi, I lead the Wishlist Plus engineering and product teams. And I'm pretty much the only one who doesn't have the founder title in this list of speakers.

But there's a reason for that. Like I said, you don't know me, I'm pretty sure you don't know Swym as well, even though I know a lot of you have worked with us and we've been around since 2016. is built with founder-minded people.

It's built with builders. It's built with people who come with high agency ownership. And that's how I became part of Swym.

I was looking for an option, and this came to me, and I couldn't have been happier. Because after coming to Swym, I've had some amazing opportunities here. I've built a gift registry app, which made around $100k in revenue so far.

Now I'm leading the Wishlist Plus app, which I hope most of you know about. But just to give a sense of the scale of Swym, here is what it looks like. Every second, there are thousands of shoppers adding items to their Wishlist, showing some sort of an intent, subscribing for a product that's out of stock, getting a notification from Swym.

So that keeps happening and I thought I'll put this out here so that because we don't talk about it enough, we're very silent that way, and today is a good day as any to let the people know here. So if you do need access to this though, to understand what we exactly do, please hit me up and I'd be happy to do that. Okay, so now the introductions are done, I'll finally get to why I'm here, and hopefully it makes sense why I would love to share our experience of what it means to build in this world where AI is, on the go, on the fly, and what doesn't really change even after that.

I think it has always been about the merchants. It will always be about the merchants, and that's what I'm trying to show through our experience at Swym, where we have like 50,000 odd merchants that work with us every day, and what it means with this fast-moving world at AI. So moving on, this is not a talk about AI, even though I've used AI almost five times This is not one of those talks where I can come and tell you, hey, I am the one who's going to teach you how I am, or I am the judge of how well you are.

But this talk is about being honest about the people in this room, what we all share, what all of us can learn from each other. This is more about learning through each other's experiences, and I'm going to share my experience, our journey at Swym, how we have been thinking and rethinking our strategies with these changes. So quickly just recapping the people in the room.

What do we share? We share the same merchant, the same Friday launch day panic where the merchants like almost like, I need this done right now. We share the expectations which keep changing over time.

I think Swym has been in the ecosystem since 2016, where it used to be just like a simple heart icon and that was enough then, but the expectations just keep growing. And you know, we all share that. We've seen that happen, you know, right in front of our eyes over the years.

The ecosystem, the partners, the integration partners, the technology partners, the Shopify ecosystem partners. I think we all share that, and I think we all work with each other in one way or the other, and we bring each other like business in one way or the other. And of course, the platform, which is why we're all here, Shopify, and how it has been like on top of its game year after year, and never leaves us not surprised, and how it's been super AI native and like top of its game here with their AI first mindset.

So we're gonna talk a lot about how Shopify has set us up so well for us to be able to be AI first. But that actually brings me to the question. Like we know that Shopify is definitely getting there.

They are AI native already or they are becoming AI native. But like I don't know how many of y'all had the FOMO over the few days of the conversations that you've had, there are we there. Like as an ecosystem, as people who are sitting outside of the Shopify offices, how much as app partners, as agency partners, as solo devs, how much are we there in that journey of being a native?

So three things that I'm gonna talk about before, like talking about what worked for us is, how things did not work for us. So in the last 12 months, we had Swym. We started thinking about how our AI first, AI native journey is going to be.

And that rebuild, restructure, and redo of how we will be building for this future. So things of course did not go well. Like we were just like all over our heads that, hey, we got this and the tools are amazing.

Now we're gonna be like supercharged and we're gonna solve every problem out there. But the reality, we build the tech. We built the MCPs, the iPods, the automations, but we did not realize that we were not building the solutions.

We were just building technology, and we were just experimenting to be very honest. And the best and the worst part about AI is that it lets you build things super fast and faster than anything that I've ever seen. And a lot of times, it's a lot of wrong things that it helps you build as well.

It's very easy to build today. So that was the first mistake. Next, we decided that, hey, we're all going to come together as a group of people at Swym, and make sure that we have a common pulse.

We have a common place to get any answer that we needed to. But we did that without really having a common place for data. And data here doesn't really mean the data that merchants have, which is in terms of their product data, their customer data, but it also means the conversations we were having.

It means the code. It means the daily check-ins with the customer tickets. It means like, hey, what incidents were we having?

How was that infra working? All of that data wasn't at the right place. And we were trying to build like a common pulse for anybody to be able to answer any question whatsoever.

Next, we tried to build something that was very incentivized towards Swym and how we could like make sure that our merchants were able to adopt features, rather than be able to get outcomes out of our platform. And we built an agent based on strategy of almost making Swym look good. And we thought that if we made some good merchants will look good, but definitely like we were missing something So these are the things that you know like the top three things that I wouldn't repeat again And I wish somebody had told me 12 months ago but There was a time I think six months or eight months where like we got together and we huddled down and we're like Hey, where is it all going wrong because yes, we were using the top tools We had access to every possible high tech AI tool that was there.

But everything wasn't coming together. And then we realized that we had forgotten our basics. It needs to be about the merchants.

So we stopped building solutions for merchants, but we started thinking of imbibing ourselves as, hey, we are like the merchants. How would we want the solution to be built? even though this sounds like a very small difference, but trust me, the thinking, the strategies, they're very, very different.

And like that thin line, once you cross over, it's going to change the whole game. So that brings us to the main part of the talk, which is the lifecycle that we try to create around our merchants. So the first part, the product itself, how well could our product adapt to this AI trend?

like we decided that we had to meet the merchants where they are. We couldn't meet the merchants, like you don't tell them, "Hey, go try this really fancy new skill or like MCP tool that we have created for you." So that you're going to see new value out of it.

We had to be where they were. We had to be in the conversations that were happening on our customers' remote. We had to be there in the emails, we had to be there on the Shopify admin.

We had to be there in Sidekick, which is where Shopify was taking us. and that's what mattered. As long as we got to them in a way that it felt natural, it felt like it was within the system, it would work.

What the merchants really want is, they don't want to feel alone, and the older version of an onboarding experience used to be like, "Hey, there are a couple of steps." You click next, next, next, and then you'll somehow set up the system. Sometimes you'll get lost, sometimes you'll have to look at hundreds of docs to make sure things are set up.

But that had to change, and AI did make it possible, cut to what we see here. This is Intercom where most of our merchants try to reach out to us. This is a time where Sidekick did not really exist.

And I'm glad Sidekick did not exist for app partners at that point of time as an extension, because we built this within Intercom, which is the highest recorded chats between us and the merchant. And what you see is a live onboarding happening right there where we're able to take them through. you know, let me set up a store.

Tell me who you are, tell me your persona. I'll set up the store exactly for you. I will look at what themes you have.

I will set up the wishlist button in the right place that you want to have. And all of this is happening as if I were just talking to like a support agent, but this is pretty much Intercom's own Fin AI agent. And earlier there used to be, I think we had the chat, we had the talk about 2,999 conversation.

We used to also have 4,000 to 5,000 conversations They used to be half of them are non-bodding. This single thing, it just killed it all and it also made it very real that we could reach to the merchants where they exactly were. Next, of course, Sidekick came in and as soon as it was launched, we knew we had to be there because that's where we were going to see the merchants figure out what to do next.

We also had to get them to some success. How can we help them understand their products better? What is working?

What is not working well? them, you know, restock the right items, restock inventory, which made sense with the intelligence built within the system that they were already using. We couldn't go and be like, hey, use this other, you know, application, other interface to get this knowledge.

It had to be where everything else was. So this is what you're seeing on the screen right now is how the Sidekick extension that we have, it helps the, you know, merchant understand what do they need to restock, when do they need to restock, how many people are liking a product, not liking a product. and almost like a conversation with them.

Next, after all of the beautiful onboarding and beautiful support with Sidekick, we'll still get chats and that's not gonna change. So building the onboarding wasn't enough. We knew that if we needed to be a leaner team, if we needed to have the team being able to spend time on problems that were more interesting, we had to solve the easier problems within the product.

So this is using something called as Intercom procedures is able to understand us through our MCP tools, it's able to actually answer non-English language questions which would have taken us way longer time in the earlier days because we're all like just English speaking support staff and this would have gone on for days and days and this is like within a few minutes, a user was able to set up the header icon on the wishlist on the mobile phone and their website. This was unreal in the previous days then of course AI has made it happen. Moving on to, like yes, we solved the problems, we've onboarded them, but there will be reasons why they are not going to be doing well, and we need to make sure that we are on the path to show them more value, to make them more successful.

And we need to let them know that we are here on your path. So this is an email that pretty much is part of something called as our value agent, where we try to get to them and tell them, hey, this is how you're exactly performing, and this is how much you're converting, your total actions, your AOV, et cetera. But you can't just give them the bad news, you need to give them actionable items.

So hey, how can we help you? Some of it, yes, we can directly help you. Some of it will be just like, why don't you restock this item?

Because it's our stock, and you have 100 people wishlisting it. It's gonna be really good sales for you. So these are the next steps.

It's also telling you data that the merchant's getting within the email, because that's where they're accessing you. They're not coming back to your admin again and again and again. They'll read this when they see value out of, hey, you're leaving so much money on the table.

Let me get that to you. So that goes to the success part. Now this was pretty much like a sneak peek into how we're thinking of product.

Similarly, we need to think about the people who are building the product, the operators of the product, because we cannot be AI first without our operations being AI first. So what I'm going to try to show you is how we have built or rebuilt our organization around the AI first vision. And it's not a single agent for everyone because everybody has a very different persona.

In a product company, you'll have product, you'll have engineering, you'll have marketing, you'll have GTM support, finance and whatnot. And each of these functions need to get on top of it in their own way. But it cannot be a single agent catering to every persona.

It has to be like a team of these agents. And I know it sounds sci-fi, but trust me, getting on to some of the other side where we're still in progress, this is what it looks like. So what you see here is what Swym's operating system today looks like.

We have a bunch of external connectors obviously, some of the internal data, which is where we struggled in the last where we went wrong. We tried to bring it all together. Beacon is where all our merchant data sets, their metrics, the product, the customer data, all of that, systemic knowledge, the most important thing that we realize.

Being in the ecosystem from 2016, we had 10 years worth of code which change hands. I joined five years ago. Arvind is sitting right there and he's built this 10 years ago.

We both are doing different things, but we needed a way to know that what exactly is a system running on to be able to solve problems that exist. That's what systemic law is. It's built on the QMD formula where we are able to talk to our code, really understand what's exactly the code telling you and not just listen to it and read through some documentation.

It's coming from the source of truth, it's coming from the code. It also has data about incidents, it has data about our infrastructure, like how is it built, what are we using in the infrastructure. All of that is our systemic knowledge on top of whatever intelligence that we have from our business.

And finally, like a unified context gateway where you're going to see our APIs exposed to all those different agents that I was talking about. Like, this is how, like, we've been looking at it. The agents need to talk to each other, just like we talk to our colleagues.

And each of them, like, we've tried to name some of them, and you'll see some of the names here. Interest of time, I'm gonna leave this on the board, and if you really wanna get deeper into this, do hit up, and I will walk you through more of it. But the whole idea was that we had to rebuild, and it had to be done in a way that each of it almost mimicked a human life cycle in the agent format.

But there were principles on building that out and it was like, hey, what are those personas gonna be? Like, yes, of course they were roles, like a product manager, an engineer, and who's doing the PR reviews and who's probably doing the testing and all of that. But there had to be like four proper roles for it and that's how they earned a seat to be an agent in that world.

One is somebody needed to notice all the signals. Hey, what is the data telling me? What are customers telling me?

Listen to that sources, use that knowledge, and do something about that knowledge. The other was the person who took that knowledge and did reasoning on top of it. And trust me, the model doesn't matter.

Like you do not need a Fable, but you need to know what you are doing with that model. And on top of it, you needed an actor. Like you needed a way to reach to your merchant, whether it's Sidekick, whether it's some sort of an MCP, the Intercom example that I showed you, we had to meet people where they work.

And then finally, like the learner, because you are gonna make mistakes. So is the agent. And that needs to keep learning and relearning.

And that is what exactly the four different roles had to be here. What I'm gonna show you is what we built. It's Alfred.

It's sort of river, but for very like specific operations as of now, we have built this. And Alfred's not trying to look impressive. It disappears into work.

If you see here was, actually it went ahead. So we had an incident on our infrastructure. Alfred responded, it knew what was happening, it knew the queues were blocked up, came up with the root cause analysis, it came up with the strategy of remediations, and that was it.

Like, this would have taken at least hours of a DevOps engineer to figure out. Next, Alfred was able to help Renuka prepare for, "Hey, I'm gonna be talking to this merchant. "You need to tell me how they're doing.

"What are the insights? "What is their data? what are the opportunities that I can present to them so that they can be more successful.

So that's how Alfred's helping, like Renuka, a product manager, do her job. Finally, Alfred's also helping your sales and your success team pull up their data from the external world. Like this is all their HubSpot data, their support tickets data, and it's helping them have that conversation of, hey, we're here for you and we know where you are.

This would have taken us a support or a success team member a whole two to three hours to find out what they are doing, where are they in their lifecycle, and all of that. And that's what Alfred does for it. And for what it's worth, it's not built on a very high-five model, it's built on a very basic model, it's built on compliance, because that's what the merchant cares about.

They don't care about, like, hey, what model you're using, they care about the compliance, and they care about the use case. It's what we built that matters. It doesn't matter that you're gonna get the most amazing next model to do it for you.

Next, we do have a lot of agency partnerships This is something that I want to open up. We are rolling this out to some of our agency partners, where we're helping them do exactly what we try to do. That is, if you have a portfolio of merchants with us, to be able to know how healthy they are, where are they struggling, what are the opportunities to reach out to them, hit them up and be like, "Hey, here is where you're struggling.

"Let me come help you. "Here is how much I can do for you." And this is all connected to our APIs.

It is connected to how we are thinking about their success. So it should be like a playbook for the agencies to work with apps in the future that they will have to have this united ecosystem. So all this talk for only one thing.

I think it's not about what we built. I think it's about how we thought about it and what mattered. What I want to leave today with is what really matters is how we start thinking from where we want to get.

So working backwards for the merchant, working backwards for the value that we drive for the merchants. start with which phase is it that you touch the merchants? Like where is it they're going to reach out to you?

Where is it they're going to interact with your product? How do you make that as friction-free as possible for them? Then the agent must also know how to serve them at that particular place.

You cannot have it served at a different place. That's where I try to show some of the stretch points. And the other part is the knowledge, because you need, of course, the merchant, what they're doing.

You need how to serve them, but you need a little bit of that knowledge, the data, which is not only like systemic business knowledge, but actually systemic knowledge of the system that they are using. And finally, when you put your agents, they should be able to see it. It shouldn't be like hidden away or it shouldn't be like an extra path for them to hop over and crossover.

So that's the strategy or that's the mindset that most of us are trying to build by. I do not say that, hey, we're 100% right. This is a perspective.

This is something that is working for us. We're learning from it. But what we did try to do is we've tried to come up with a strategy of trying to rate ourselves on, hey, where are we in this AI first world?

So my first question that I started with, yeah, Shopify is getting AI first, are we? Because it is time to be honest. It is time to be, are we going to be talk of the past?

Or are we going to be able to keep up with them? So what I did do is, with Trudy's permission, I actually did create a merchant persona for Trudy and Design Packs. And we tried to hit up without the persona here.

And if you see, it gave a very cold reply first. We're saying that-- OK, so I asked a merchant on an app asked, should they use for the product page? What do you recommend?

I don't know if Trudy-- that's a common question that you get. But that's something we tried to ask. And initially, without the details essentially you would have given about your business, about your data, it gave a very dry answer.

But then later on if you see I did try to give it understanding of, okay, here is what Trudy does, here's what Design Packs do, now can you come up with a plan? It did come up with some plan. It gave exactly, and Trudy, I have not shown this to you, so I don't know how accurate that is.

But it gave a very personalized recommendation based on what you believe, what your product does for them, and I think that is powerful. Being able to think like the merchants and get them answers that they would be able to relate to, I think that is powerful. That's the first part.

The second part which I was talking about earlier, is having a scorecard. This is something that we try to do for ourselves, is we've given everything that we know about the business, we've connected it to external connectors like Shopify's Dev MCP, the Shopify AI skills, and all of that. even our partners like Klaviyo, etc. have their own MCP and skills.

We try to have that and we try to have what we have built along with what I just showed you as like the persona for Trudy and Design Packs for Swym. And then we like try to rate ourselves like, "Hey, where are we walking? Where are we crawling? Where are we running?

What can I do to advance to the next step? What are the other people in this ecosystem doing?" And it really helps us think like the other people in the ecosystem, which are the merchants, the partners, and of course, the Shopify folks who are on top of their game.

So this is something that we've done. And what I'd love to do is give this to the community. It's available.

You can go ahead and fill, I'm gonna just show the template right here, some of these questions. It has the skill to be able to help you, whether you're an agency partner or you are a builder, you can fill this out and you can try to see where are you in your journey. And if that interests you, please sign up for this early access and let me know.

Finally, I want to really take this moment to acknowledge the reason why we all have been able to run so fast. It is because of the primitives. And the primitives were done by Shopify, the Dev MCP that they have.

The skills, the toolkit, the AI toolkit is amazing. Like we are able to do so much for the merchants, which you saw in the demo earlier, is because the toolkit makes it accessible to us. the extensions, the proxies, the UCP, which is giving us a catalog.

I think all of this is not possible without that, and that is what we have to keep our game in parallel with and make sure we are there. So yeah, this is something I wanna end with. Like anything that we build now, we think of two questions.

Would I want this if I were in the merchant org? It's not like what would I love to build, but it is about the merchant. Would they want this?

would like the marketing manager or the e-com manager or like a merchant who's just running a mom and pop shop, would they want that? And what is the value that it will drive to the merchant and their shoppers? So if any of the things that we try to build do not pass these parameters, I would say don't build it.

It's better not to waste your time building something because AI makes it possible. Like in the earlier days, we would think 10 times. Nowadays, I think we do not.

But I think this is just some grounding questions to always think about. But that's about my time. Thank you so much.

And you can connect with me. That's my LinkedIn URL. Also, we love to talk about how Swym's building, what we do.

But we do it in a very silent way on our Substack. So do follow us if you have anything that you'd like to learn or watch from our insights. So thank you.

[APPLAUSE]

Audience: Great.

MC: So we have time for a couple of quick questions. And then there's one in the back. I'm going to go get- way in the back over here.

Audience: Hi. I'm literally in the process of building a similar system. So this is super relevant.

Thank you. I was really interested that you've given partners like an MCP to your app. And I think that spoke to what Gavin was talking about Can you expand a bit or maybe like how they're using it or are they literally trying to plug it into their systems or maybe what insights you're giving to them?

Sakshi: So these are agencies that already have their own LLM and their own systems. They don't need a new system to be handed out to them. What our MCP does is it gives them a hook into here's how we can help them solve the same problems that they are solving and that's what the MCP does.

So a lot of the partners are using Claude. A lot of the partners are using Copilot, ChatGPT. And that you cannot go and give them, I'm going to give you a separate system.

We need to integrate with those LLMs, because what the LLMs can do, we won't be able to build an agent ourselves and do that. So the MCP just provides them the context, the knowledge which I was talking about. It gives them that and helps them solve the same problems, but with that added knowledge.

Audience: So you mentioned that, sorry, it's here.

Sakshi: Oh, yeah.

Audience: Yeah, so you mentioned that you used Intercom where people were able to do the self onboarding kind of thing, right? So what percentage of your customers or merchants were able to do successful onboarding themselves using the Intercom itself?

Sakshi: Okay, so I have an overall stat. We get, if we get 100 chats and out of them, like 30 chats are what we man, everything else is Intercom on-boarded and Intercom solved. But if you ask me, are the merchants ready for it?

Not all of them, I have to be honest about it. You will see even in the on-boarding that I did show you, we do give them an option. If you're not comfortable with chatting with Fin, we give them like, I'll set it up myself.

So they are age old, click next, next, next and set up the on-boarding. But what we do see is they do get stuck up after a while because it's not easy to set up an app, especially when we are an app building on top of an ecosystem. Shopify has challenges for the apps to seamlessly integrate, and that is where they get stuck.

Then they reach out to Fin again. So at some part of the onboarding, 70% of the merchants are actually reaching out to us, but not the whole entire flow. And those numbers, I'll come back to you exactly on.

MC: Okay, one more quick one.

Audience: Hi, thanks for that. Long term agency, long term user of Swym, actually quite a few merchants on your product is really good. I was wondering how you think about managing data privacy, GDPR compliance in your AI stack. You obviously have a lot of information in there that needs to be managed and you mentioned compliance, compliance, compliance.

Sakshi: Exactly. I think I can't tell everybody how important it is because it's very easy to be on the wrong side of these things and AI makes it very easy. So if you look at the Alfred, which is the agent that we use, it is built with those filters where it has its own rules of what data can it really read before which everything is sanitized.

We cannot give it data. We cannot give it ROI data. We can't give it revenue data.

You can't give email IDs. you can't give customer details, you will have to do that sandboxing of the data, and that's something that we built before that layer. So that's 100%.

And so the GDPR part, again, having that solved for is 100% important. I don't know if you know, but we do use cookies, but we have a wishlist experience that works without cookies. It does not have certain things, like once I go away from the system and I've refreshed my browser, I can't come back to it.

So those kind of things you lose, But we have to be on the right side of this because not being on the right side of it will not- you cannot have the merchant sign up for you. Each merchant comes and asks you for your SOC2 certification and whatnot, and they'll have you sign up. I do not want my data to be fed to agents.

And that is real. So it's very important to do that. And we try to do that at the best of our capabilities.

MC: All right, thank you. [APPLAUSE]