This post was originally published on this site.
Engineering has accelerated dramatically because AI can now help build products, but advertising hasn’t changed at the same pace. AI promised to free marketers from operational work, and advertising platforms are becoming increasingly agentic. Google, Meta, TikTok, and others are deploying their own agents that can recommend, and increasingly execute, actions for advertisers.
Yet much of performance marketing remains manual. Teams have access to ChatGPT, analytics platforms, and a growing array of AI tools, but humans still act as the glue connecting them. They spend significant time pulling together data, investigating performance, moving decisions between tools, and manually implementing changes. AI can accelerate individual tasks, but people remain responsible for holding the overall process together.
Adtech startup Adlyse has developed a platform that helps paid media teams keep pace, use advertising budgets more efficiently, and drive growth and revenue across the business. I spoke to CEO Anna Stepura to learn all about it.
The company was co-founded by Ukrainians Anna Stepura and Roman Kuzmych, along with Yaozhong Kang, bringing together experience across entrepreneurship, adtech, AI, and engineering.
CEO Anna Stepura is a serial entrepreneur who previously co-founded AI recruitment startup HireGPT, which was acquired by ApplicantQ, and AI sports coaching company AiSport.
She has also been an Entrepreneur in Residence at Antler and a Menlo Fellow at Menlo Ventures. Kuzmych brings more than a decade of experience in advertising technology, optimisation, and AI, including serving as Director of Product at PulsePoint, while Kang is a former Meta engineer.
Adlyse believes that the next generation of marketing organisations will bring people and AI agents together as one coordinated team across advertising platforms and the broader growth funnel.
“Paid advertising has changed. Teams need to operate at a speed and scale that manual campaign management alone cannot support,” said Stepura.
“That requires reliable collaboration between people and AI agents, with shared business context and clear responsibilities. People retain control over strategy, boundaries, and approvals, while agents execute without requiring human involvement at every step.
We built Adlyse to give growth and paid media teams the infrastructure to make that operating model reliable and scalable.”
As advertising platforms develop AI capabilities within their own environments, businesses need a way to coordinate decisions across them. Budgets, customer journeys, and business goals span multiple channels. An improvement within one platform does not necessarily produce the strongest outcome for the business as a whole.
Adlyse provides a shared operating layer for that work. The platform connects advertising channels, business data, and tools across the growth funnel. Its agents continuously monitor performance, investigate issues, recommend actions, and execute authorised workflows, requesting human approval where required.
Marketing teams define the goals, permissions, rules, and approval requirements. Agents work within those boundaries, with shared context and visibility into their actions.
From manual campaign management to agentic advertising
According to Stepura, paid media remains one of the main growth channels in advertising, particularly for ecommerce and D2C companies.
“Ultimately, businesses don’t care about having the best-performing campaign on Meta or Google. They care about revenue and how much they get back from every dollar they invest.”
“If Google isn’t performing well this week, for example, we could recommend reallocating part of that budget to Meta or another channel where we’re seeing an opportunity,” explains Stepura.
Adlyse agents can monitor campaigns 24/7, analyse them deeply, identify what’s working and what isn’t, and try to understand why.
Once Adlyse identifies a problem, it can run more than 100 hypotheses about what it could do next. “We explore different directions and predict the potential outcome of each.
“We then select the most promising action and go back to the human and say: this is what’s working, this isn’t, this is why, and this is what we recommend doing next.”
The analysis and decision-making process can happen autonomously, but the human remains in control and approves the proposed action before implementation. The campaigns still require human approval because they’re dealing with real budgets — often very large ones.
Once the action is approved, the agents return to the advertising platforms and apply the changes. That could involve changing a creative, headline, budget, or another campaign element.
How Adlyse’s agents decide what to do next
Under the hood, Adlyse uses different LLMs for different tasks, depending on which model performs that particular task best. It also has reasoning models specifically designed to perform well for performance-marketing tasks.
“We believe agents should proactively move entire workflows forward, understand the nuances of paid advertising across channels, and know when to act and when to ask for human approval,” Stepura said.
“They need to see the full picture because success for a brand means growth and revenue, not simply a campaign performing well in isolation.
Agents should work toward those business outcomes, wherever the opportunity exists, across any campaign or platform.”
Real-time optimisation catches problems before budgets are wasted
Crucially, Adlyse works with real-time data. One problem with traditional campaign management is that money can keep being spent on something that has already stopped working.
Stepura explained that by the time somebody notices a creative is declining, you’ve potentially wasted a significant amount of budget.
“Our agents work proactively, so they can identify that decline as it’s happening and recommend an intervention.
“We also analyse historical campaign data and use predictive models to look at what could happen next.”
Building an operating layer across thousands of marketing tools
Adlyse focuses on mid-sized companies through to enterprise customers, typically businesses managing advertising budgets starting around $100,000 per month. They usually operate across multiple platforms, which creates significant complexity. Their accounts contain a lot of data, and analysing everything manually requires considerable human input. These companies suffer particularly from delayed decisions and delayed optimisation.
Previously, there wasn’t really a way to do that work much faster because somebody still needed time to dive into the data and understand what was happening.
Adlyse currently operates across six advertising platforms and connects with more than 3,000 tools across the growth funnel, but it aims to eventually work across every advertising platform.
“For us, another advertising platform is simply another traffic source where a company can invest money and potentially generate a return.”
Why cross-channel advertising matters more than ever
Managing advertising across multiple channels is becoming more important because it increasingly takes multiple interactions before somebody converts. People might encounter a brand on the street, hear about it from somebody else, see something on Instagram, and then encounter it somewhere else.
“That’s why we’re so focused on being cross-channel rather than optimising one platform in isolation,” explained Stepura.
Despite the noise within the era of GEO (Generative Engine Optimisation, aka visibility in generative AI platforms), Stepura says that, depending on the category and type of business,
Google still generates a lot of conversions. For B2B brands, LinkedIn is expensive, but if it’s done correctly, it can work very well. There are also underestimated channels.
“One we’re starting to implement is connected TV, where the infrastructure increasingly allows advertisers to buy advertising in a similar way to other digital channels.
It’s not suitable for every brand, but we’re seeing interest.”
Back to LLM platforms, OpenAI advertising is still very early. Stepura believes that while it’s interesting for companies wanting to experiment with something new:
“But if your immediate goal is return on ad spend in terms of revenue, I wouldn’t necessarily expect that from OpenAI ads yet.”
What happens to performance-marketing jobs when agents take over?
I was curious about what happens to performance-marketing jobs as more of this work becomes automated. Stepura thinks junior-level, and even some mid-level, roles will disappear.
“There isn’t a need for humans to continue doing so much of the manual clicking and repetitive work. But we’re also seeing new professions emerge.”
She cites the example of a growth architect — someone with deep performance-marketing expertise who also understands how to design systems in which AI agents work together.
“These are people who might have 10 years of performance-marketing experience and deeply understand the field, but they also understand how to design a system in which agents work together.
“They don’t necessarily need to build the technical infrastructure themselves. They need to understand how to architect the system.”
In this model the marketing department becomes increasingly agentic, with humans and agents working alongside each other without creating bottlenecks.
Closing the loop: measuring what happens after every AI decision
Crucially, Adlyse doesn’t just optimise and automate campaigns. The platform also tracks what happens after each change.
“We call this the learning loop. We monitor what happens after an action is applied until the point where that change stops affecting the campaign. That allows us to measure the impact and determine how accurate our prediction was.”
According to Stepura, the scale of the improvement depends significantly on how well the account was managed beforehand. “
We’ve worked with campaigns managed by very established agencies where perhaps 10 people were working on a single account. Even there, we’ve been able to improve performance by around 10 to 15 per cent.”
But in well-managed accounts, she says, much of the value comes from eliminating repetitive work rather than delivering dramatic performance gains.
“We’ve seen reductions in manual work of around 55 to 60 per cent. Our goal is to reach around 90 per cent automation, leaving humans to spend the remaining 10 per cent on strategic work.”
The gains can be considerably larger when Adlyse takes over an account that was previously poorly managed.
“We’ve seen return on ad spend (ROAS) increases of around 150 per cent in those cases.”
What comes next for Adlyse
Adlyse also operates with an AI-first approach internally, so it doesn’t necessarily need to hire large teams. It’s currently going through the SOC 2 compliance process, which is important for reaching more enterprise customers. The company is already in active conversations with a number of them.
The next major product milestone is integrating as many advertising platforms as possible, so humans can orchestrate advertising across everything without being limited by individual platforms.
“Longer term, of course, the ambition is to become a unicorn as quickly as possible and bring this way of working to as many advertising accounts as we can,” Stepura concludes.
The company is backed by ZAS VC, Accel, Boot64, Aperiam VC, New York Angels, and other investors across the United States and Europe.
Lead image: Adlyse founders Yaozhong Kang, Anna Stepura, and Roman Kuzmych.