How to audit your entire email and SMS automation program — and generate a full omnichannel campaign from one prompt — in under 18 minutes.
What You’ll See in This Demo
Klaviyo Composer is an in-app AI marketing agent built into the Klaviyo B2C CRM that knows a brand’s account data, its customers, and the Klaviyo product itself. In this episode, Gil Hsu, Product Manager leading Marketing AI at Klaviyo, demonstrates Composer’s flow audit and campaign generation — auditing every automation in an account for gaps, overlaps, and performance issues, then building a multi-message email and SMS campaign from a single plain-language prompt. It’s most useful for ecommerce and retention marketers running large automation programs who want AI guidance without exporting their data into an outside tool.
Key Takeaways
- Klaviyo Composer is an in-app AI marketing agent that moved into public beta on June 30, 2026, and runs directly inside a brand’s Klaviyo account.
- Composer’s flow audit reviews every automation in an account — both how each flow is built and how it performs — and flags gaps, gaps in coverage, and overlapping triggers such as abandoned cart competing with abandoned checkout.
- The audit returns a stack-ranked list of fixes ordered by expected return on the marketer’s time, and Composer can estimate how long each fix will take.
- Because Composer is trained on marketing and on the Klaviyo product, it can explain why a recommendation matters and give step-by-step instructions for the specific setting to change, without the user writing a persona prompt.
- Composer generates full omnichannel campaigns from one prompt, producing segmented email and SMS messages with subject lines, brand-matched creative, and scheduling in a pre-draft state that a marketer approves before anything sends.
- General-purpose LLMs require marketers to manually move context in and out; Composer sits on the Klaviyo database, which holds both a brand’s own customer data and aggregated best practices by industry and product type.
Watch the Episode
About Klaviyo
Klaviyo is a B2C CRM and marketing automation platform that unifies ecommerce customer data with email, SMS, and push messaging for consumer brands. Composer, its in-app AI marketing agent, entered public beta on June 30, 2026 alongside Klaviyo’s Customer Agent, with both agents working from the same real-time customer profile. Learn more at klaviyo.com/composer.
About Gil Hsu
Gil Hsu is the Product Manager leading Marketing AI at Klaviyo, building the next generation of digital marketing products. With over 15 years of experience in product, engineering and design, Gil is passionate about leveraging data and AI to create tooling and strategies that drive real-world impact for customers. Outside of work, Gil enjoys traveling abroad, tinkering on bikes and staying on top of technology trends.
FAQ
What is Klaviyo Composer?
Klaviyo Composer is an AI marketing agent built into the Klaviyo platform. It audits a brand’s existing campaigns, flows, and segments, surfaces a ranked list of opportunities, and generates complete email and SMS campaigns from a plain-language prompt. It entered public beta on June 30, 2026.
What is an AI marketing agent?
An AI marketing agent is software that can analyze a brand’s marketing data, decide what work to do next, and execute it — rather than only answering questions or writing copy. The distinction from a general chatbot is access to live account data and the ability to take action inside the marketing platform.
Can AI audit my email marketing automations?
Yes. Composer’s flow audit walks every automation in a Klaviyo account, reading both structure and performance data, and reports gaps in lifecycle coverage, overlapping triggers, and underperforming flows, then ranks the fixes by expected impact.
How do you generate an email campaign with AI?
Describe the outcome in plain language — the channels, the audiences, the offer, the theme, and the send schedule — and Composer produces the full set of messages in a pre-draft state with subject lines, previews, audience assignments, and creative, which the marketer edits and approves before it becomes a live campaign.
How is Klaviyo Composer different from using ChatGPT for marketing?
A general-purpose model only knows what you paste into it, which means manually exporting account data and pasting results back. Composer runs on top of the Klaviyo database, so it already has the brand’s customer data, flow structure, and performance history, plus aggregated benchmarks by industry and product type.
Chapters/Timestamps
00:00 The problem: AI that doesn’t know your customer
00:55 Gil Hsu on leading Marketing AI at Klaviyo
01:32 What Klaviyo Composer is and who it’s for
02:31 Demo: opening Composer inside a Klaviyo account
03:40 Running an audit of your automation flows
05:30 Gaps, overlaps, and conflicting triggers
06:45 The stack-ranked list of what to fix first
07:50 Asking follow-up questions of your own account data
09:10 Step-by-step fixes and the agentic roadmap
10:44 “I have two hours — what should I work on?”
12:09 Generating an omnichannel email + SMS campaign from one prompt
14:11 Why in-app context beats copy-pasting into a frontier model
16:57 Where to learn more
Resources
Klaviyo: https://www.klaviyo.com
Transcript
Greg Kihlström (00:00)
Most AI tools operate without a real time picture of who a customer is or what they need next. That means they lack purchase history, engagement signals, or intent to act on. But what if you had an AI agent that had all of that and was connected to the workflow where your marketing actually happens?
Welcome to one amazing thing about Klaviyo. Today we’re talking with Gil Hsu, product manager, leading marketing AI at Klaviyo. And he’s going to be sharing one amazing thing about Klaviyo Composer with us today. Gil, welcome back to the show.
Gil Hsu (00:38)
Hey, Greg, thanks for having me on. Now for the second time, I gotta keep making these appearances.
Greg Kihlström (00:43)
I know returning champion here. So yeah, looking, looking forward to seeing something else here before we do though. And for those that didn’t catch last time you were on the show, why don’t you give a little background on yourself and your role at Klaviyo
Gil Hsu (00:55)
yeah. So for myself, product manager here at Klaviyo leading marketing AI, building the next generation of agentic marketing products. yeah, largely these days I spend my time thinking about the intersection between marketing and AI, and then specifically around how Klaviyo and its rich database of customer information has unfair advantages to win in this space.
Greg Kihlström (01:19)
Nice, love it. And so we’re going to take a look at Composer today. Why don’t you, before you share your screen, why don’t you tell us, you know, what does the product do and who’s the audience and why is it important to them?
Gil Hsu (01:32)
Composer is Klaviyo’s new flagship product that was actually just announced into public beta a couple of weeks ago at K London, which is our big event in London. And it’s basically a in-app first party AI marketing agent. It knows everything about your account data. It knows everything about your customers within the Klaviyo account. And it also knows marketing. We’ve built it to understand marketing out of the box. And then specifically, because it’s Clavius product, it just knows what all the features are, where all the buttons are. So as it’s not only giving you strategic advice, it gives you actual real actionable advice as to how do you actually do a thing. So we’re really excited about this product that we’ve shipped now with a lot of usership in these first few weeks. And yeah, excited to demo.
Greg Kihlström (02:27)
Love it. Yeah. Let’s let’s take a look. Why don’t you share your screen?
Gil Hsu (02:31)
OK, great. So here we are in the Klaviyo account And right here in this upper-hand corner, there’s a new shimmering star. And this basically opens up our brand new Composer Experience. And the Composer Experience can do a lot of things. It can do anything from auditing flows, which are our automations product, to generating campaigns, to auditing segmentation and just generally getting a better understanding of your marketing program overall. And so I’m going to be doing a couple different demos here. We’ve got a lot of utilities within Composer, but I’m going to be doing two specific things. One is I’m going to be showing how we do, how Composer can do flow auditing or automations auditing, just basically showing you all the sort of nooks and crannies within your automations program, which can get a little bit unwieldy over time.
And then also showing you how we can generate campaigns with basically simple prompts and how that basically will help agencies or customers be able to create campaigns at scale, especially as we think about personalization and making things more custom for each customer. These are the kind of tools that people will soon have or soon be really good at using to be able to build personalization on scale. So this is a lot of like themes coming together that Composer will be able to help with. So first off, we’re going to go and do an audit of flows. So we see this here in the right-hand panel for Composer. We can also see that this account has a number of automations here, everything from, let’s let this load. OK, so we have a number of different flows related to different channels, A-B testing, everything from abandoned car to placed order and so forth. These are sort of normal.
marketing automations that one might have. And so through what is a number of these automations, it can be pretty difficult to understand exactly what every automation does, especially as, you know, owners of these marketing programs come and go, or just because there’s a sheer number of automations, it’s hard to see how all the automations overlap. So this is where Composer really shines. So you can open up Composer and you can do something as simple as, let’s get this to load you can do something called audit my flows. So you click this button and it’s kicking off a pre-built job that we have sort of out of the box. It’s kind of like a pre-built prompt basically you can think of. And it’s doing a very specific thing. It’s walking over all the automations within this account, looking over not only the structural data, meaning how was all these flows built, but also the performance data across all of those flows and the nodes within those flows.
And it comes back with a particular phrasing of what an audit is, but customers will be welcome to make or customize what the audits are or what they prefer audits to be like in the future. So let’s take a look at what this audit came back with. So in this Terraza account, we’ve got 21 live flows that are fully audited. Seems like set up around broad lifecycle coverages. Actually has some issues around several high traffic flows. There’s some risks around welcome SMS and competitor sites high level, you know, some things to look at. It calls out things that are going well, these welcome series, review requests for Klavier reviews, abandoned cart, these seems to be doing okay. But then it’s starting to flag some issues, particularly around what I mentioned for gaps and overlaps across all of these multitude of flows. So it says things like, abandoned cart versus abandoned checkout seems to have overlapping triggers.
Or post purchase and nurture a thing if you exist but no dedicated cross-sell So there’s an opportunity here that that we see as a gap so being able to at any time ad hoc be able to Do an audit and uncover these gaps and overlaps is like incredibly powerful especially as the automation program grows and in this last little piece of what you know our pre-built audit does it gives you a stack rank list of things that you can be doing right now that we think are the sort of most
ROI for your time, let’s call it, that will impact your existing flows program. So in this case, the first one here that’s critical, it’s like, let’s update this WhatsApp welcome series, rewrite the body, open the URL, et cetera, et cetera. And so, at this point of the demo, always tell folks that I show that the audit itself, while great, while very powerful, is really not the thing here. It’s the fact that you now have an AI directly in your account that you can ask any question to. So what that basically means is you now have the opportunity to have a conversational relationship with your data. So if you could ask any question about your data, what would you ask? Well, let’s take some of these. Let’s say, for example, there’s a recommendation here that you don’t totally understand. So you could say something like, you gave me this rec, this, I don’t understand why that’s important. Explain it to me.
So because this is not just an out of the box LLM that you might see from other frontier model providers, so we’ve trained Composer, our AI specifically on marketing, and then it understands your account too. So through that context, you can ask these type of questions without saying, you’re a senior marketer with X many amount of years of experience and you’re trained on X and Y and Z. It already is like that out of the box.
And so it can give you guidance as to, and context as to why this is a problem, how you might fix it, and just generally get confidence between this gap of AI tells me something, do I understand what AI is saying? And is that enough for me to actually do a thing? And so we have this built in and we see that as one of the biggest differentiators with what Composer can do. And then just one last nugget here. Okay, let’s say through all this explanation, you say, okay, you convinced me.
Tell me why. no. Give me step by step instructions on how to fix this. So again, because this is our AI built on top of our system, it knows where every setting is, where every button is, and then can just give you very specific guidance on how to do a thing. And now you have that sort of ad hoc ability to find how to change that one little setting that’s hidden behind three or four clicks and then get to the thing. And so now we’ll actually say that this is an early version where we’re just sort of giving guidance. Eventually, which is where we come like agentic world is that composer will be able to propose the solution for you, not tell you how to do it yourself. Instead, here’s what the fix is.
Would you like me to do it for you? You can either accept it, reject it, or maybe if you want to hedge, you could accept it as a test, give it to 5 % of your customers. And you just simply do this thing. Very similar to say like a Google docs, somebody giving you a suggestion and you can say, accept it. Right. And so now all of a sudden you have this ability to really move quickly in terms of there is a problem here. Let’s go ahead and fix it. And I’ll just say just a couple more things here in terms of like for use cases especially on like the auditing side where we see there’s a lot of room to grow with people not really knowing, sort of like seeing around corners with how to optimize their, marketing program. This is the biggest thing that we get asked at Klaviyo. It’s, you know, what could I be doing better? And so this auditing is one of the solutions. but another thing that we could do, for example, is,
Greg Kihlström (10:34)
Yeah.
Gil Hsu (10:44)
You could say, I only have a couple hours to work on my flows. What’s the more important thing that I can be working on right now? And instead of having to know all the issues and then select the most important one, it can just give you the most important ones. And not only that, it gives you approximately how much time it thinks it will take for you to do that thing. So really, really great there. And then just like moving, of switching gears a little bit, I wanted to show the create side.
So we know that for customers of all sizes, that creation of campaigns is a laborious task, whether it be you want to do original content, original copy, and then thinking about segmentation. Well, if we were to do something like wanting to generate, let’s say, an omnichannel campaign, I have this prompt preloaded, I think, here, where you could say something like, create an omnichannel campaign using email and SMS.
and send this pair of messages to each of my top two segments. So the marketing campaign should have four messages total. Include the top selling product on a 20 % promo and have the copy and content focus on 4th of July. Schedule the emails to go out Friday and the SMS to follow up two days later. So it’s like a pretty specific prompt, but certainly a lot easier than actually doing the work. So you can put that in there. And then I’d say like any good cooking show, I’ve already got the outputs pre-baked.
Gil Hsu (12:09)
So I have this example here, it’s the same prompt and here’s the output. So you have a preview where you can actually go, this is actually what we call a pre-draft phase where it’s not actually a draft message yet. This is still like you’re going back and forth with the AI. And so what we’re seeing here is, okay, we’ve got two segments. We’ve got VIP customers and we’ve got repeat buyers and engaged subscribers. You actually see this is the non VIPs. And then we’ve got two messages for each email and SMS, email and SMS. And then we have them starting from Friday, starting at noon, and then following up on two days later. And better than that, you can click on any one of these. You’d see that it has subject line, preview, has the audience lined up. And actually, this is a pretty good email. And you can continue to hone in terms of design and content. But it’s pulling from the customer’s account.
They’re brand assets. And then for things that are net new, we can actually generate images and certainly copy on the fly. But not only do you get this desktop view, you also get mobile view. You can go back here and look at your SMS and see how it would preview on iOS and Android. And when it’s finally ready, you say, you’ve done all the back and forth with AI. You can click this button. This is demo account, we don’t have it enabled here. But you can create and edit the campaign, which turns it into a draft campaign, which gives you all the sort of levers that you can do to modify your campaign in Klaviyo. And so we see that as still human in the loop will get you 80, 85 % of the way there. And then you’re kind of there to just bless everything at the very end and then actually schedule. So just a couple examples here, two of many utilities that are within Composer, but just really excited about this new age of marketing AI and then specifically how Klaviyo has really put a very opinionated lens on what marketers can do with AI.
Greg Kihlström (14:11)
Yeah, yeah, this is this is great. And so, you know, to go back to the the the first part of what you were talking about, I mean, you know, as you were talking it through, I was, you know, in my head, I was thinking, well, what if I what would it take to copy and paste all of that information into some other frontier model and do all you know, and I think just lots of not only from a time perspective, it would take a lot of time, but also just from I think there’s a cognitive load part that’s
You know, I you have to kind of keep track of like, well, what do what does it need to know that that Clavio in this case has in it versus man, you can just chat with it and ask all the questions and it knows. So, you know, I think just from that, from from so many levels, from my standpoint, it it lets a human actually be more strategic and stop thinking about, well, OK, how do I get this information out of one thing and into another? Right.
Gil Hsu (15:08)
Yeah, absolutely. I mean, I think there’s so much lift involved from, yeah, you describe it as copy and pasting, but it’s basically like context management, right? It’s to say, here are all the things that, you know, organically live in either in this account or maybe you have other connections for, you know, analytics data or content in Canva or Figma. And then you’re saying, okay, well, consolidate all this information somewhere else and then push it back again, into the actual execution services. And so I will say, for some very sophisticated users, they’re certainly doing that. But they’re able to do that because they have really strong opinions. I think the lion’s share of folks, just want it to work. And so we have it in such a way that there’s a term in AI called the harness, which is basically an opinion of what AI should be doing. We’ve built that specifically around marketing and specifically around how Klaviyo works. But I would even say, though that even if you were to very like, you were very, you know, capable in terms of pushing context into these frontier LLMs, there’s still going to be a major difference. And that difference is you can only push your data to these third party LLMs. The benefit of being part of Klaviyo is that it sits on top of what we call the Klaviyo database. And that is this sort of, it’s our B2C CRM that has all this context on not only your customers, but an aggregated view on what’s best practices per industry, per product type. And so as you’re trying to get guidance, help me build a, you know, on brand, on trend content or imagery that pulls from our aggregate data. And so you’re just not going to get that anywhere else. So this is again, why instead of the top, like we have a very unique right to win in this space.
Greg Kihlström (16:57)
Yeah, love it. Well, Gil, I always love having you on here and seeing amazing things from from you and Clavio. For those that want to learn more, what’s the best way for them to do that?
Gil Hsu (17:07)
Well, I’d say, you know, certainly clevia.com for all the things. If you’re specifically interested in composer, it’s clevia.com slash composer. Yeah. Give it a try.
Greg Kihlström (17:17)
Love it. Well, again, I’d like to thank Gil Shu, product manager, leading marketing AI, Klaviyo for joining the show. You can learn more about Gil and Klaviyo by following the links in the show notes.
Related Links
Go deeper on the strategy: Hear the CLV and acquisition-cost conversation with Klaviyo’s CMO on The Agile Brand with Greg Kihlström: #705: What happens to your KPIs when both CLV and customer acquisition costs rise?
More demos like this: One Amazing Thing About Docket with Arjun Pillai · One Amazing Thing About Scrunch with Devin Stevens · #49: One Amazing Thing About Klaviyo with Grant Deken
