What happens when marketing software starts learning from customer behavior instead of following a campaign rule that someone wrote weeks ago?
In this episode of Tech Talks Daily, I speak with Neha Mittal, co-founder and CEO of JustAI, about AI agents, reinforcement learning, and a different approach to personalization. Neha spent over a decade working on growth and retention at companies including Twitter and Pinterest. Her experience repeatedly brought her back to the same problem: teams were running sequential experiments while the interests and behavior of their users kept changing.
Neha describes how a series of small A/B tests could consume weeks without producing a durable answer. Changing a button from red to blue might teach a team something, but the larger opportunity was to keep experimenting and learning across many decisions. Seeing similar problems at different companies helped shape the idea behind JustAI.
We discuss why she considers marketing a promising environment for autonomous agents. It involves a high volume of decisions, frequent feedback, and experiments that can often be assessed quickly.
Opens, purchases, and ignored messages can all contribute to learning. Neha contrasts those conditions with higher-consequence fields such as medicine and law, while explaining that strategy and judgment remain responsibilities for people.
The conversation also separates generating content from choosing an action. A language model can help produce an email. The reinforcement learning approach Neha describes aims to learn which decision supports longer-term customer value. The question becomes what message, timing, or channel makes sense for a particular person in the current moment, rather than which static segment or campaign they belong to.
JustAI divides the work across Strategy, Creative, Decisioning, and Data agents. Neha explains how opportunities inform messaging, how the decisioning system learns from responses, and how the data agent feeds insights back into the process. Marketers can review customer-facing creative, provide brand guidance, and adjust strategy when the company's products or priorities change.
Her hypermarketer vision raises another question about how teams grow. Neha imagines one marketer having the capacity traditionally associated with a much larger team, with agents handling repetitive execution and experimentation. She presents this as a direction for the business rather than a demonstrated productivity ratio for every customer. The marketer still needs to translate business goals into instructions and judge whether the results make sense.
We return to current examples. Neha describes Coursera and ClickUp delegating the creation and selection of messaging variants. She also discusses the importance of human review for Better's financial-services communications. These examples show why the level of autonomy should depend on the task and the consequences of getting it wrong.
The supplied booking notes report over $100 million in customer revenue influenced and a $17 million funding round. Those company claims provide background, but the interview does not establish a revenue attribution method, incremental lift, or a complete return-on-investment calculation. The practical value of the conversation lies in understanding what work is being delegated and what people continue to own.
For me, the key challenge is creating a feedback loop that improves relevance without letting operational speed outpace customer trust or business judgment.
Which marketing decisions would you delegate to AI agents today, and what would you want to review before a message reaches a customer? Please listen and share your thoughts.
Useful Links
[00:00:00] Do you struggle to keep your AI agents from acting outside of compliance? Well, Denodo provides an AI data layer connecting your data systems, keeping guardrails consistent across all of your data platforms. So start scaling your business carefree with Denodo. And you can do that by visiting denodo.com to learn more. But now, on with today's show.
[00:00:29] What if the next great marketing team is one person supported by a small army of AI agents? I know, it sounds like the sort of claim designed to make every marketer clutch their campaign dashboard a little tighter. But my guest today believes it is already becoming an operating model. And she is the co-founder and CEO of JustAI.
[00:00:52] And following over a decade building growth, personalization and retention systems at Twitter and Pinterest, she's going to share today why marketing could be the ideal proving ground for autonomous agents. Because we're going to discuss how reinforcement learning differs from generating another perfectly pleasant email, as why customer behavior cannot be reduced to static segments.
[00:01:18] Instead, we're going to explore the idea of the hyper marketer. The agents handling strategy, creative decisioning and data. And where human judgment will always remain firmly in charge. But enough for me. It's time to take you all on a road trip to San Francisco. Get your coffee ready as we sit down and take a practical look at what happens when personalization finally learns from every decision.
[00:01:46] So let's get my guest on now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Sure. Yeah. Thank you for having me. My name is Neha Mitha. I'm the co-founder and CEO of JustAI. Before starting this company, I've spent over a decade building growth, personalization and retention systems in very large companies like Twitter and Pinterest.
[00:02:14] My, the simplest way to think about my job is some version of how do you get millions of people to come back and engage with platforms like Twitter and become loyal customers. And just to put this in context for everybody listening, when you see your email inbox or your, you know, phone screen flooded with notifications that are saying, hey, we miss you, come back to this app. Most of the times you would go and ignore those.
[00:02:42] I'm sure something that might not be obvious to you is those marketing decisions are actually a huge set of rules and campaigns that are maintained by an army of marketers who are babysitting every single notification every day. So what JustAI is doing is it's replacing the guesswork with a reinforcement learning system that's learning how people are responding, what they care for and what they don't, and it self-improves to get more and more relevant with every send.
[00:03:10] And that's one of the reasons I was excited to get you on today. I think getting a little peek behind the curtain, and when we look down at those notifications, I've no idea where they come from or how it's done. So before we get there, though, I mean, if we go back, I mean, you spent more than a decade working on growth and retention at companies such as Twitter and Pinterest, to name but a few there. But right at the beginning, what problems did you repeatedly encounter with things like personalization
[00:03:39] and marketing technology that would eventually lead to you building JustAI? I feel like there's got to be a story there. Yeah, yeah. There's tons of stories. The funny thing is every company thinks that their problem is unique, but underneath they were all wrestling with the same exact issue. So I guess going back to my time at Twitter, we would spend lots of time running one-off A-B tests. So we'll run a test, we'll change the color of the button from red to blue,
[00:04:08] and then wait for three weeks. And again, if something works, we will say, all right, blue is the way to go, and then let's try another test. So now we are sequentially running lots and lots of A-B tests to get to understanding what people want, but the way users work is they've moved on. They're now starting to think about something else. The trends have changed and so on. We saw this at Twitter, and once in a while, something interesting would happen. One in 10 A-B tests would show us really good results,
[00:04:37] like, hey, 10% movement on our daily active users. So the insight was not figuring out red versus blue. The insight was we need to be scaling and continuously A-B testing as much as we can, and machines were not designed to do that. So that's when we first started building the stack that lets us run scaled, or what we call now dynamic A-B testing. At the time, we would joke about building a company around this software.
[00:05:02] But, you know, as Elon acquired Twitter, I went on to Pinterest to run growth and retention there. That's where all the patterns started to repeat in terms of how do we personalize our messaging for users. And the silver lining in Twitter's acquisition was a lot of my colleagues went to different companies, like Snad, Reddit, and so on, and they were all seeing the same problem. And so that's what started to validate the idea that, hey,
[00:05:28] personalization is this lofty buzzword that everybody talks about, but nobody has seen in reality. What a great story. I absolutely love that. And I was also reading before you joined me today that you believe marketing could become one of the first enterprise functions where AI agents move from just co-pilots that assist people to operators that can actually execute work. So what is it that you think makes marketing particularly well suited to this transition?
[00:05:56] And what should marketers still remain responsible for? How do you see that? That's a great question. We were thinking a lot about this as we were starting the company, just because you're seeing AI everywhere, you know, from legal to health tech and whatnot. There are three characteristics that make marketing really the perfect environment for autonomous AI. The first one is we're making millions of decisions every day with subject line, which notification, what timing, what channel.
[00:06:25] And AI is suited to do that because that's a scale problem. The second one is marketing is a function where every decision is producing immediate feedback. Did somebody open the email? Did they make a purchase? Did they ignore it? And that's a constant learning loop that we can leverage. And the third one is that the cost of experimentation is relatively low. So, for example, if an AI chooses a different subject line, you can immediately learn if it performed better.
[00:06:51] Versus if you're talking a field like law or medicine, those decisions are obviously higher risk and feedback takes much longer. So that's where AI agents and really going autonomous with AI makes sense from marketing. But to your other question, humans will still define the strategy. There is a good layer of judgment that will still be human driven. But all of the operational work is what we'd be delegating to AI.
[00:07:18] And personalization is something that's been promised by marketing platforms for years now. But much of it has amounted to segmentation and predefined sets of rules. So what is it that changes when AI enters here and has it continuously interpret customer behavior and decide the next best action for an individual without accidentally drifting into creepy territory? Because it's always somewhat of a balance, right? Yeah.
[00:07:46] It goes back to what we were discussing earlier. So the problem is that customers don't behave like rules, right? So, for example, just to again put it in perspective, when you see an email, you're not thinking, am I part of campaign one or campaign two from this company? You're thinking, is this product that has a 40% discount now, is this still relevant to me? Right? So that means we're now, instead of asking the question, which for the last decade we have been,
[00:08:14] which journey should this user enter or which campaign should we target this user with? The system is now asking, given everything I know about this person right now in this very moment, what is the single best action I can take? Which starts to become a fundamentally different problem to attack, right? So now we're not talking first names. We're definitely not talking segments. We're talking, can we get extremely one-to-one with users in terms of giving relevant content, not particularly making things creepy, if that makes sense.
[00:08:45] Yes, it really does. And I was also reading that you introduced this idea of a hyper marketer, where one person equipped with AI agents could potentially operate with the capabilities that previously were associated with a much, much larger team. So what does that operating model look like? And how does it change how companies should think about headcount skills and marketing budgets? Yeah, yeah. Kana hits our core vision.
[00:09:13] So our vision is that one hyper marketer can do the work of 20 marketers. And it's not particularly about replacing marketing or marketers. It's really about removing the operational work. So today, a marketer is building campaigns. They're monitoring dashboards. They're writing copy, analyzing experiments, talk about localization. They're now doing translations across 20 different languages, right?
[00:09:39] All of that work is something that AI agents can now continuously do and automate. And so think like robo investment, where you're saying, hey, what's your risk profile? You're putting in your guardrails, allocating goals, and letting the system do the rest. It's very much that shift that we're going to see. And sort of what we expect of a hyper marketer, one, they're able to shift altitudes. So they're able to go and say, hey, this is what my CEO and CMO needs to see in terms of where the company is going.
[00:10:09] But also, I can go down to the level of guiding AI in terms of what needs to get done to get there. So we are going to see hyper marketer as somebody who is fairly technical, but also has a really good judgment on what the business needs. And when I was reading about just AI and doing a little research on that, it seems that you divide work between strategy, creative decisioning, and creative decisioning and data agents.
[00:10:38] So just to bring that to life a little, can you walk me through how those specialized agents can collaborate? And especially where humans enter the process too, and what you've learned about making multi-agent systems work reliably in production and not necessarily replacing people, but working side by side. Yeah. So the construct that we use is basically we want to create a feedback loop throughout. And the feedback loop is managed by four distinct jobs and hence four agents.
[00:11:07] The first one is the strategy agent, which is basically deciding what opportunities exist. So given your users, your campaigns, your company goals, what are the biggest opportunities you should attack right now? It then passes those opportunities to the creative agent, which goes and evolves messaging. It both learns from your brand and tone guidelines, but also the previous messages you've sent and there, and how users have responded to it.
[00:11:31] Once those creatives are in place, the decisioning agent decides who gets what message. This is where our reinforcement learning model comes in that learns very quickly in terms of what messages are resonating with users. And in real time, it optimizes for how messages should be sent. And then to complete the loop, what happens today is we're running all these decisions and EB tests and whatnot, but we're not taking those learnings and reapplying them back as insights.
[00:12:00] That's where the data agent comes in, which basically takes away the need for having multiple data scientists trying to look at patterns. So the data agent kind of proxies that data scientists work for you. They're constantly talking to one another. The piece where humans come in is every time we're doing something that is visible to the end user. So for example, if the creative agent is going and creating 20 emails that users can see, this is where marketers will have a chance to review them, edit them, and so on.
[00:12:29] Same thing goes for strategy. It's not as autonomous as, oh, here are the best three things to do. Typically, every company is evolving. Their products are changing. So the world is very much actually, I don't know, Figma just launched, you know, Figma AI and whatnot. And now that context has to be placed in marketing, right? So all of that stuff is where humans would come in to sort of tweak how we proceed.
[00:12:53] And much of the current AI conversation revolves around LLMs and generative content. But of course, you're also using reinforcement learning to optimize marketing decisions. So tell me a little bit more about that and what does reinforcement learning actually allow you to do that maybe the traditional LLM can't? And could this approach have applications moving far beyond just marketing, do you think? Yeah, yeah.
[00:13:20] In the most simplified ways, so large language models are phenomenal at generating language, and hence they can do a great job of generating text, email, whatnot. They're not designed to optimize decisions over time. Reinforcement learning is asking, again, a completely different question. So instead of asking, can I write a good email? It's asking which decision today is going to maximize long-term customer value for the company.
[00:13:48] And to do that, it's now processing millions of data points every second, every hour, and then trying to get to a better decision. So in some sense, it's a self-learning loop where it's experimenting, observing outcomes, and continuously updating its strategy. And think about, the simplest way to think about it is also that's how humans learn. So a baby who's just born might not know what a tree is, but within minutes of being on the road,
[00:14:15] they would start understanding the difference between a tree and not a tree. So that's the concept of reinforcement learning. Marketing is absolutely just one application of reinforcement learning. It is actually applied in a lot of different places already. So for example, when you see Netflix and all the movies it's recommending behind the scenes, it's using that. Meta has used this technology for the longest time for ads and understanding which ads to show at what point. So, you know, it's existed there.
[00:14:43] How we use reinforcement learning and how we tweak the algorithm for different use cases, that's where some of the work lies. And this year, I've seen and heard so many ambitious claims about agentic AI, whether it be on a tech conference or on my news feed. And I always like to bring things back to measurable outcomes and return on investment. And I know that you're seeing hundreds of marketing decisions being delegated to AI every month
[00:15:10] and more than $100 million in customer revenue being influenced. So I'd love to bring this to life further by asking you, what are customers like Coursera, ClickUp and Better Mortgage and so many other big name clients here? What are they actually delegating to agents? And what have you learned about where autonomy delivers genuine business value? Because I think this is something that every leader is chasing right now. So we talk about a few things.
[00:15:38] We talk about how we can shrink four years worth of testing to one month. Yeah. Right. And that really means what they're in this case, at least let's take Coursera and ClickUp's example. They've delegated creating thousands of messaging variants within a month and then letting AI pick the best one. That would have taken them four to five years to do by hand. Right. So that's one example of, hey, we're not doing one-off tests.
[00:16:04] We're not letting AI pick the best message, phrase it correctly and understand who should it send it to. Those decisions start to scale when they're starting to talk about which channel is the best channel. If this user uses push notifications, maybe I start there or maybe they come to the app quite often. It's a prompt that I show on the app. So those are, again, decisions that AI is starting to take over.
[00:16:27] Again, the parts that still stay with them and they work really closely with AI is on every single nuance and sort of the last mile of things. So, for example, Better, that's in fintech. They have to take a lot of care to make sure that every email that goes out is properly written and, you know, it's compliant all the way. So that's where we still see humans in the loop play a huge role.
[00:16:50] Well, and I also believe it's an exciting time for you and indeed just AI because, I mean, you started today talking about how your idea was validated, the journey you've been on. But if we fast forward to present day, before you came on today, I was reading that you've recently raised $17 million and also growing investment around the infrastructure needed to make agents useful inside businesses.
[00:17:15] So if I ask you to look ahead now, what do you think the marketing organization or what do you think the marketing organization becomes when AI agents can handle more execution, handle more optimization? And what skills will separate the marketers who gain the most from this new model? Because it feels like somewhat of a crossroads and there's exciting things on the horizon here.
[00:17:37] Yeah, so one of our core visions that's not very cleanly understood is, so when we talk about, hey, what we do, a lot of companies would say, okay, so you are a plugin on top of an existing Salesforce or HubSpot that sends out messages. And we try to explain the idea, no, they're going to be commoditized very quickly because they just become your sending layer that's finally sending out the message.
[00:18:03] And we're looking at a world which is very, very flexible and not based on deterministic rules, which means now you just tell your agents what kind of marketing you care for and agents are going and running execution at scale. So that means we're not just building a better Salesforce. We're not just building intelligence on top of it. We're saying, hey, your workflows are going to get disrupted where you're going to be instructing agents on what to do. And to do that, you have to build infrastructure ground up.
[00:18:33] To support that. Because now the infrastructure is not, again, based on deterministic rules. It's very flexible and it can be tweaked very quickly depending on your requirements. So that's sort of the world we're building towards. It's very, it's so new that we're also playing this out in terms of adoption and users comfort. But we see it coming. We see people adopting something like this faster than we thought they would. So that's sort of the exciting piece of it.
[00:18:57] And again, back to sort of what skills people we see in like hyper marketers are people who are starting to experiment and fail, honestly. So people who are just going and building their own thing and starting to see how much they can tweak the AI towards their use cases. So, yeah. And that's probably the most fun part about this job is working with those marketers and seeing where we can take this.
[00:19:22] Well, I love your story, the journey you've been on and how from the idea that you had there and that story that you shared to how Just AI is now generating over 100 million in customer revenue while challenging everything that you thought you knew about scaling growth at the same time. Amazing story. And for anyone listening that would like to find out more information about Just AI or anything we talked about, where should they go? Yeah, well, I'm fairly active on LinkedIn.
[00:19:49] So if anybody wants to continue the conversation there, please just feel free to DM me. You can also find us on getjust.ai. That's our website. And there's a button in there. If you click there, you will go directly to my calendar. We share a lot of learnings about what we are seeing in reinforcement learning, AI agents and so on. So I'm always glad to hear from people who are thinking about the same problems.
[00:20:13] So many big takeaways from our conversation today and also some big questions and whether the future of marketing isn't a bigger team with a bigger budget or another software dashboard, but a single marketer working alongside an army of agents. So many big possibilities there. I'll have links to everything that you mentioned. But more than anything, thank you for sharing your story today. Appreciate your time. Yeah, thanks for inviting me.
[00:20:37] I think marketing AI begins with a better decision rather than just another mountain of generated content. And that distinction between language models and reinforcement learning, I think matters so much. Yes, one can write the email while the other can learn whether sending it, changing it or leaving the customer alone will ultimately produce a better result over time. And that final option may earn a cheer from crowded inboxes everywhere around the world.
[00:21:07] But the big lesson here is keep humans responsible for strategy, judgment, brand and compliance. And then give the agents the repetitive experimentation that they can perform at scale that no human team could dream of matching manually. So thank you to my guests for sharing how just AI is putting this model into production.
[00:21:31] But over to you, could you your next competitive advantage to smaller marketing teams that learn faster rather than larger teams that just send more and create more? TechTalksNetwork.com. Let me know your thoughts as always. And I'll be back again tomorrow with another guest. Speak with you then. Bye for now.