Hello, and thanks for checking out the latest AI Gamechangers, your weekly chat with the people putting AI to practical use across the games industry.
This week, we’re talking advertising. Specifically, we’re chatting with Phoena Pang, VP of Americas and Global Partnerships at Mintegral, about how AI is reshaping every layer of the ad tech stack, from creative generation and campaign optimisation to fraud detection and beyond. Phoena’s background spans engineering, product and business development at companies including Google and Vungle, and she has a sharp perspective on how multi-agent AI systems are changing the way mobile game developers approach user acquisition and monetisation.
If you’re new here, welcome aboard! Each edition of AI Gamechangers features an in-depth conversation with someone working at the frontier of AI and games. Our full archive is free to read, and we’d love it if you shared this with a colleague. As ever, scroll down past the interview for our curated round-up of news from around the web.
Phoena Pang, Mintegral
Mintegral is one of the world’s leading mobile advertising platforms. We sat down with Phoena Pang, Mintegral’s VP of Americas and Global Partnerships, in person during PGC Summit San Francisco last month. With a career spanning engineering at Vungle, business development and product leadership at Google, and a stint at Moloco, she sits at the intersection of technical depth and commercial strategy.
In this interview, Phoena explains how Mintegral is deploying multi-agent AI systems to automate creative production and campaign management, discusses the cat-and-mouse game of AI-powered fraud detection, and shares her view on why AI won’t replace creative talent (but will radically change the teams around it).
Top takeaways from this conversation:
Mintegral has built a multi-agent AI system where specialised agents handle different tasks (from creative generation to quality control) orchestrated by a planning layer, rather than relying on a single model to do everything.
AI is compressing entire user acquisition teams into single-operator roles. The skill set is shifting from hands-on execution to orchestrating AI tools and systems effectively.
Fraud detection is an arms race: Mintegral uses post-install behavioural data to train models that spot fraudulent traffic earlier each cycle, but fraudsters are using AI too, making collaboration with advertisers essential.
AI Gamechangers: What’s your own background? Please give us a sense of how you got to where you are today.
Phoena Pang: My background in engineering has fundamentally shaped how I approach leadership in ad tech today.
I currently serve as VP of Americas and Global Partnerships at Mintegral, a leading mobile ad tech company. But I started in a very similar company, Vungle, working as an engineer! I was working on in-game rewarded video ads and building SDKs for developers. Through that, I spent a lot of time engaging directly with developers, which made me realise I was more interested in the business side. Over time, I transitioned from engineering to a business development role where I focused on partnering with developers to bring them on board.
“AI is not here to replace games – it’s here to help make them better. Content is still king. Therefore, creating great experiences, building strong IP, and capturing players’ interest, these things won’t change”
Phoena Pang
Eventually, I moved to Google, where I focused on business development and partnerships. In that role, I learned how imperative it is to build trust with partners based on deep product knowledge. I got very familiar with their tooling, which ultimately led me to transition into the product side of the business. I then spent two years as product lead, working on the gaming ads.
Eventually, I realised I prefer smaller companies – small teams can be more nimble. I wanted a role that would enable me to apply the full range of my experience across engineering, business development, and product. This led me to Mintegral, where I do a bit of everything I’ve done before.
How have you seen the role of AI in advertising evolve?
When I started in the industry around 12 years ago, there was essentially no AI in day-to-day workflow. People talked about it occasionally, but it wasn’t applied in a meaningful way.
That’s changed completely in recent years. Though AI is now dominant across most industries, advertising was one of the first early adopters to apply machine learning at scale. Programmatic advertising has historically relied on algorithms to help determine which ad to show to which user and at what bid. These recommendation systems and bidding engines were already what we now broadly call AI.
What’s changed more recently is how AI is being used across the entire advertising workflow, particularly on the creative side. Traditionally, you needed large creative teams and studios to produce ads: running A/B tests manually, iterating, and optimising over time. Now, you can utilise AI to generate thousands of creative variations in a single day, automate A/B testing, and quickly identify the best-performing creatives with AI analytics.
That changes the scale and structure of teams. Where you might previously have required a full user acquisition team, it’s now just a single operator. The role shifts from execution to orchestration, emboldening skills to leverage AI tools and systems effectively.
So AI is not just improving back end systems; it’s transforming front end workflows as well, making the entire advertising process faster, more efficient, and more scalable.
What tools are you using in the background? Are you committed to a particular AI ecosystem?
It really depends on the use case – whether we’re working with images or video, for example. Different models are better at different tasks, so we’re not tied to a single AI ecosystem.
“An important factor is understanding your audience. Developers know their own players best, and when they share that data with us, we can utilise AI to help acquire users with similar profiles, improving targeting and efficiency”
Phoena Pang
Internally, we’ve designed a multi-agent system. Envision it as a team; one agent focuses on a specific task, another checks the output, and another handles quality control. Each agent is specialised, but they work together. For more complex tasks, we introduce a kind of planning layer. One agent is responsible for breaking the work down (similar to project planning) and then distributing it across multiple specialised agents.
Those agents may leverage different models depending on the objective. Once each part is complete, we aggregate everything into a final result. Rather than relying on a single model to do everything, it’s about orchestrating multiple agents and models to work together effectively.
Using AI at scale can be very expensive – how do you advise people to manage costs?
There can be costs related to token usage. One way to mitigate that is to use local servers. If you build your model on a local server rather than entirely on the cloud, this could help reduce expenses.
Another point to consider is using memory wisely, which allows reutilisation of tokens. For example, at Mintegral, I may ask AI about campaign performance while someone else inputs the same question. One way to address this is by retaining and referencing prior outputs, helping to eliminate any redundancies that can lead to increased costs.
What’s the experience of your Mintegral platform for the end user?
Behind the scenes, we’ve always used AI and machine learning in the back end, but that hasn’t been visible to users. What we’ve introduced now is a front end AI agent that helps our customers navigate and use the platform.
Previously, users would have to figure things out themselves, searching through help centre articles and learning how to set up campaigns step by step. Now, they can simply ask the agent, “How do I create a campaign?” or “How do I set my targeting, or upload creatives?” It can answer those questions in real-time and guide you through the process.
That’s the first version. The next version, which we’re already testing internally, goes much further. It will analyse campaign performance, suggest optimisations, and eventually take action on the user’s behalf.
“AI can improve productivity, but the final decisions still sit with humans. The ideas and creative concepts, especially in games, come from people. AI can execute and refine, but it doesn’t originate truly novel ideas on its own”
Phoena Pang
The end goal is that people can easily have a conversation with the agent, and it will create campaigns, upload creatives, and optimise performance for you. We’re moving from a tool you operate manually into something much more conversational and automated, where the AI helps you both execute the work and understand it.
You’re a global company. Are there any differences in the way people embrace AI around the world? Are Western companies using it differently from Eastern companies?
Companies across all regions are trying to use AI – that part is very consistent globally. Whether I’m talking to developers in Europe, the US, or Asia, they’re all exploring how to apply AI, either in their workflows or directly in their games.
Where things start to differ is in the tools and models they use. For example, in China, there are strong local models from companies like DeepSeek, Alibaba, and ByteDance, which are more widely adopted there. Cost is also a factor – local models are often cheaper than tools from OpenAI or Google.
At the same time, there are practical limitations. ChatGPT or Gemini aren’t easily available in China without workarounds, which create an additional barrier. Teams will ultimately need to use a mix of models depending on the task, local options for cost and accessibility, and international ones when they need broader capabilities, especially if they’re building games for a global audience.
The overall mindset is similar everywhere. The main differences come down to access, cost, and which models are most practical to use in each region.
AI can be used to detect fraudulent transactions. Please tell us your thoughts on the role of AI in that side of things.
Yes – this is something we’re actively exploring and investing in: utilising AI to identify fraudulent traffic in our network. The challenge is that the companies generating it are also using AI, so it’s an ongoing game of cat and mouse.
Much of our approach begins with post-install data to identify abnormal behaviour (patterns that don’t look like real users). From there, we train our models to detect those signals as early as possible. Even if we already know certain traffic is fraudulent, we use that data to teach the model what the early warning signs look like, so we can catch it sooner next time. It’s a learning process where you analyse what’s happened, and then use AI to recognise those patterns earlier in the future, essentially turning known fraud into predictive detection.
Another important aspect is collaboration with advertisers. When they detect fraudulent behaviour, they share information such as problematic bundles or device data with us. We can then use that information to further train our models and improve detection.
At the moment, there isn’t a single, strong third-party solution that fully solves this using AI alone. Fraud detection systems still need to be built into the platform itself, where it can continuously learn from real-time data and feedback to make the models effective.
People are worried about their jobs; they are afraid of what AI might mean for the industry. What’s your take on that? Are people justified in worrying about AI?
AI can definitely improve productivity, but for me, the final decisions still sit with humans. The ideas and creative concepts, especially in games, come from people. AI can execute and refine based on existing data and experience, but it doesn’t originate truly novel ideas on its own.
If you already understand what successful games look like, AI can help you analyse it and iterate, but the games industry is fundamentally creative. If you want to create a hit, you can’t just repeat what’s already been done – you need something new.
Where AI adds most value is in speed and iteration. It helps teams turn prototypes very quickly and test them in the market at a much lower cost. Previously, you might have needed large teams to build and test multiple prototypes. Now, a much smaller team, supported by AI tools, can explore many more ideas in the same amount of time.
I don’t see it as a threat in itself. The bigger question is whether you know how to use AI effectively, because that’s what will really make a difference for a studio.
And how is AI affecting roles or team composition at Mintegral?
We have empowered our team to leverage AI capabilities as a core part of how they work, allowing team members to act as product managers and independently implement features.
“We’ve designed a multi-agent system. Each agent is specialised, but they work together. For more complex tasks, we introduce a kind of planning layer. One agent is responsible for breaking the work down and distributing it across multiple agents”
Phoena Pang
This shift in mindset raises the bar for how people approach their workflows, evolving from pure execution towards more critical thinking and problem-solving. This has transformed our team from executors into more strategic, cross-functional contributors.
You work with a lot of game companies. What advice do you have for them about the value of AI-based tools? When a company comes to you, how do you help them figure out what’s next for them, regarding UA and monetisation?
There are two sides to this: how we provide support in terms of user acquisition or monetisation and then how developers apply AI within their own game development.
We start with monetisation by reviewing how they leverage ads in their games. They can use AI tools to help optimise how often ads are shown, when and where they appear, and which users see them. The goal is to maximise monetisation without damaging the player experience. AI is very effective at modelling those trade-offs.
On the user acquisition side, creative testing is a key area of opportunity. AI can help teams generate and test many creative variations much more efficiently, run A/B tests at scale, and continuously optimise campaigns based on performance data.
Another important factor is understanding your audience. Developers know their own players best, and when they share that data with us, we can utilise AI to help acquire users with similar profiles, improving targeting and efficiency.
But it’s not just about one platform; most advertisers are running campaigns across multiple channels, so internally they also need to analyse which channels are actually driving value. This is another area where AI is becoming very important. Instead of relying on 10-person data teams, you can use AI to analyse performance across different channels, identify what’s working, and make better decisions about how to allocate budget.
Overall, AI helps across the entire process, from monetisation and creative testing to performance analysis and budgeting, making everything more efficient and precise.
What’s next for Mintegral? What are you working on?
We’re growing very quickly, both in terms of product and geographic expansion.
Historically, we’ve been very strong in in-app advertising (IAA) for games, and gaming is still our foundation. Over the past year, we’ve expanded into in-app purchase (IAP) optimisation, developing new models to support that side of the business, which is an increasingly important area for many game developers (especially those focused on hybrid monetisation). Improving those models is a big focus for us this year.
“Traditionally, you needed large creative teams and studios to produce ads: running A/B tests manually, iterating, and optimising over time. Now, you can utilise AI to generate thousands of creative variations in a single day”
Phoena Pang
Geographically, we’re investing heavily in high-growth regions. We’ve recently opened a new office in Brazil to support South America, which is a growing, fascinating market. The audience is large, young, and highly engaged – the number of devices is also incredibly high (two to one!). We’re also seeking opportunities in regions like the Middle East and India, aiming to directly engage with developers in those regions.
Another major focus for me is how we leverage AI internally. We’ve already developed AI tools and agents within the company, and now the goal is to make sure our teams, particularly sales, are equipped to effectively use AI capabilities to deliver greater value to our partners.
Cast your mind forward, please. Will AI continue to disrupt things in the long term?
It’s quite difficult to predict what will happen over the next few years!
This year marked my 11th attendance at the Game Developers Conference (GDC), and over the years, we’ve seen various waves of innovation: AR/VR, the metaverse, blockchain, etc. But I think AI is a little bit different, because it is a technology that enhances everything. Those technologies were often seen as replacements or shifts in how games are delivered. On the other hand, AI can support everything.
AI is not here to replace games – it’s here to help make them better. What that means is that developers can spend less time on production effort and more time focusing on the game itself. Content is still king. Therefore, creating great experiences, building strong IP, and capturing players’ interest, these things won’t change.
Right now, the industry is still figuring out how to integrate AI into workflows. But in a few years, I think it will just become a fundamental tool that everyone uses. And when that happens, the focus will shift back to what really matters: making better games.
Further down the rabbit hole
Some useful news, views and links to keep you going until next time:
Head of live games at Finnish giant Supercell, Sara Bach, told PocketGamer.biz, “We believe AI can put superpowers in the hands of creative teams with great taste.” She rejected AI for its own sake but discussed how it can help studios better understand their players.
Screen actor Milla Jovovich, famous for The Fifth Element, with Ben Sigman of libre.org, has built a new AI memory system called MemPalace using Claude.
Netflix revealed a “smart erase” feature for video. Video Object and Interaction Deletion (VOID) landed last week as a free AI tool (hat tip Jason Bradbury, who described the move as Netflix “positioning themselves at the centre of how film and TV gets made in the AI era”).
Sony Interactive Entertainment acquired UK-based machine learning firm Cinemersive Labs, which will join Sony’s Visual Computing Group and contribute to efforts “to improve rendering techniques and visual fidelity across PlayStation titles”.
Recruitment outfit Games Factory Talents is running an AI & Investments In Games careers day in Helsinki on 23rd April 2026.
Sobering thought of the week: a report by Oumi and The New York Times found that Google’s AI Overviews are accurate about 91% of the time. Sounds impressive, but given the trillions of searches Google processes annually, that amounts to tens of millions of wrong answers every hour, “misinformation on a scale that may be virtually unprecedented in human history.”
Paranoia about how quickly AI can clone games is apparent in statements like that of indie darling Lucas Pope, developer of Papers, Please and Obra Dinn. “You don’t really talk about stuff when you’re working on it, because I don’t know that it’s going to get slurped up by AI,” he told the Mike & Rami Are Still Here podcast. Check it out here: