Following our recent AdTech & Ops Summit, we caught up with Shirofune CEO and Founder Mitsunaga Kikuchi for his take on AI, automation, and the future of ad ops.
Shirofune started in Japan and has now successfully expanded globally. What did you learn in those early years that you couldn’t have anticipated, and how did it shape what your product looks like today?
When I first started the company, I believed that if we built an excellent product, people would naturally start using it. In reality, however, I learned that making people aware of a good product and helping them understand its value can be even more difficult than building the product itself.
We also initially designed parts of the product in a way that competed with advances in machine learning and Google’s advertising optimization algorithms. In hindsight, trying to build a product that resisted technological progress was a mistake.
What matters most is accurately identifying what new technologies are good at and where they still fall short. In areas where they excel, we should embrace and benefit from that progress. In areas where they are less effective, we should continue developing our own proprietary technology. That philosophy of combining the strengths of emerging technologies with our own differentiated capabilities to increase the overall value of the product, is now deeply reflected in how Shirofune is built today.
AI-powered ad automation has gone from a niche concept to a boardroom priority over the past couple of years. From your vantage point building in this space now for many years, what’s actually new and what’s largely just hype?
The originality of AI, and LLMs in particular, lies in what fundamentally distinguishes them from conventional technologies: their generality and flexibility. Because of their generality, LLMs can handle a wide range of tasks, including ambiguous ones. For example, you can ask an agentic AI system to consolidate multiple campaigns within the same advertising platform, which is something conventional software systems find difficult to handle.
Their flexibility also allows them to take context into account. For example, they can generate a report based on highly specific, personalized instructions about how you want the information to be presented and made easy to understand. Similarly, LLMs can adapt to the context of search criteria and generate personalized advertising content for a particular audience. These new capabilities may make it tempting to use LLMs to automate everything. However, there is a trade-off. When you take advantage of their generality and flexibility, you sacrifice accuracy and reproducibility.
Whether you are creating presentation materials or writing copy, giving an LLM broad and flexible instructions will not produce an output that is exactly what you intended 100% of the time. A very common process is to adjust the instructions several times, gradually move the output closer to what you want, and then make the final edits manually until you consider the result perfect. Even when exactly the same input is provided, the output may differ each time. This is not simply a limitation in the current performance of LLMs. It is an inherent characteristic of the technology. As the technology improves, the amount of manual correction required will decrease, but it is unlikely to disappear entirely.
Real-time campaign budget management, for example, allows almost no room for error and requires an extremely high level of accuracy and reproducibility. If spending exceeds the allocated budget, it can create a financial problem. It is therefore easy to understand why applying a technology with inherent limitations in accuracy and reproducibility to this type of task would be a poor fit.
By contrast, there is no single correct answer in areas such as media planning or analyzing the factors behind campaign performance. Even when the output is not perfectly accurate, it is relatively easy for a human to review and correct it. These are the kinds of areas where LLMs are much more likely to deliver value. The idea that autonomous AI agents can be trusted with every type of task is an overstatement. In practice, LLMs are most effective for tasks where they remain useful even if accuracy and reproducibility are not perfect.
I believe the key to future competitiveness will be identifying and selecting those types of tasks carefully. Another important approach is to break down tasks that require high accuracy and reproducibility, isolate the components that can still function even with some degree of variation, and apply LLMs specifically to those parts.
The companies that can make this distinction effectively will be the ones that extract the greatest value from LLMs.
Shirofune operates across markedly different markets, from Japan and the US to Australia and beyond. Where are the biggest philosophical differences in how advertisers think about automation and AI, and does that ever pull your product offerings in different directions?
Fundamentally, I do not believe there are major differences across markets.
Our initial hypothesis was that advertisers in Japan would be more particular about granular manual adjustments, while advertisers in Western markets would be more results-oriented and therefore more willing to leave the detailed execution to automation.
However, after working closely with ad managers in the United States and Australia to implement automation, we found that they are also highly sensitive to even the smallest details of how the system behaves. They care deeply about each individual optimization decision and how it contributes to improving performance.
This has shown us that one point is consistent across markets: automation will not be accepted unless the underlying algorithm is highly sophisticated and capable of meeting the standards of experienced advertising professionals.
There are, however, some regional differences when it comes to reporting. In Japan and other Asian markets, clients tend to expect highly granular reporting, including breakdowns by keyword, time of day, and other detailed dimensions. In Western markets, there is a greater tendency to prefer a clear executive summary that highlights the most important metrics and insights.
As a result, while the core automation technology remains consistent globally, our reporting features have developed with some regional differences.
Australia’s digital advertising market has some distinct characteristics, with strong agency cultures, a relatively concentrated media landscape, and a regulator that’s paying much closer attention to data. How does Shirofune navigate these nuances, and what opportunity do you see here that others might be missing?
In terms of its strong agency culture and relatively concentrated media landscape, Australia actually has a great deal in common with Japan, our home market.
Our business began by building a product that agencies could genuinely trust and adopt. Rather than simply offering a collection of convenient tools, we have focused on connecting each product capability into a core operating system for the agency’s overall workflow.
What differentiates us is our commitment to measurable productivity gains. We do not stop at making work more convenient; we measure actual productivity and focus on delivering tangible outcomes, including doubling operational productivity in practice.
Tell us about what Shirofune is working on right now in terms of products & features that you’re most excited about?
In March this year, we launched a feature called Creative Timeline. It brings each round of creative testing together in a single chronological timeline, while AI analyzes the results of each round and recommends the direction for the next round of creative development.
We are now expanding this feature so that AI can generate the actual creatives to be uploaded for the next round. The goal is for Shirofune to orchestrate the entire creative optimization process – from analyzing campaign results and developing the next hypothesis, through creative production and deployment into existing ad groups, and then back to performance analysis.
We plan to release a beta version within the next one to two months.
This is particularly important in social advertising, where the ability to produce a high volume of relevant creatives at the right time has become one of the most important drivers of performance. I am very excited about the opportunity to introduce an entirely new workflow and user experience in this area, and to fundamentally transform how creative optimization is managed.
One criticism of AI-driven ad platforms is that they can become a black box, persuading advertisers to hand over the keys and potentially lose the instinct and craft that made their work so effective. How do you think about the balance between automation and keeping humans genuinely in the loop?
I believe this is an extremely important issue. It is something Shirofune has cared deeply about since its earliest days.
When it comes to humans and machines working together, relying on humans alone or machines alone is almost always the wrong approach. The most important thing for maximizing advertising performance is to build workflows that combine the respective strengths of people and technology.
That means breaking the workflow down into smaller components, identifying where humans are strongest and where machines are strongest, and assigning each part accordingly. From a product design perspective, it is especially important to determine which decisions should remain with people, and how the tool should support those decisions in order to maximize human judgment and expertise.
A tool that simply claims AI will do everything in place of people is unlikely to be accepted by users, and I believe it will ultimately deliver weaker performance as well.
For a media planner, performance marketer, or agency strategist trying to future-proof their career right now, what skills or mindsets do you think will matter most over the next 12-18 months?
I believe two things will be especially important: first, how closely professionals can align their work with solving advertisers’ broader business challenges; and second, how effectively they can use AI-driven technologies to redesign traditional workflows.
On the first point, as AI and technology increasingly provide and execute standard best practices, the value of simply carrying out established advertising operations will decline. What will become more valuable is the ability to question whether the current KPIs and operating methods are truly the best way to support the advertiser’s business growth.
Professionals will need to look at advertising from the advertiser’s perspective, ask whether the current approach is genuinely helping the business achieve its maximum long-term growth potential, and continually update that approach to move closer to that goal.
On the second point, the ability to integrate AI appropriately into workflows will create several-fold differences in the value that different people can produce within the same amount of time.
Even more valuable will be the ability to go beyond improving individual productivity and design entirely new team-wide workflows that use AI to deliver better outcomes for advertisers.
If you could change one thing about how our industry as a whole approaches AI adoption what would it be?
If I could change one thing, it would be the way AI is sometimes treated as an all-powerful force and discussed primarily as something that will replace people.
At this stage, I see AI not as an independent substitute for humans, but as an extremely powerful tool for extending human capabilities, much like computers or the internet.
There are certainly cases where the industry focuses too heavily on the question of how much human work AI will replace. But I believe the more important question is how we can combine human judgment with AI to produce better decisions and stronger outcomes than either could achieve alone.
Where do you see AI-powered advertising in two years? Is there a version of that future that worries you?
I think it is fair to say that AI is already embedded in almost every part of the advertising process today. The change we are likely to see over the next two years is not simply that AI will be used in more areas, but that the capabilities that currently exist in separate parts of the workflow will become much more deeply connected.
AI is already being used across analysis, bidding, budget management, reporting, creative generation, and many other areas. In many cases, however, these capabilities remain fragmented. People still have to connect the tools, interpret the outputs, and move the process forward from one stage to the next.
Over the next two years, I expect the entire process from performance analysis and the development of the next hypothesis, through creative production and deployment, and then back to the next round of analysis—to become more integrated as one continuous learning cycle. As a result, people should be freed from more repetitive work and able to focus more of their time on understanding advertisers’ broader business challenges and making more strategic and creative decisions.
At the same time, there is also a version of that future that concerns me. As AI makes it possible to produce every kind of advertising output quickly and at scale, the industry could become flooded with what is often called “AI slop”, work created without sufficient intent, hypothesis, or validation.
Mass-producing similar-looking creative, plausible-sounding analysis, generic optimization recommendations, superficial media plans, or weakly supported operational decisions will not lead to better performance. Without a deep understanding of who we are trying to reach, what we are trying to communicate, and why, as well as genuine human care for every result and every expression, the advertising industry risks giving up the ideas and creativity that it has long considered central to its value.
I am also concerned that overestimating the capabilities of AI could lead us to undervalue the role of people. If handing everything over to AI is treated as the most advanced approach, we risk weakening the human expertise that has always driven strong performance: the ability to form hypotheses, notice when something does not feel right, understand the context of the market and the customer, and care deeply about every detail of the outcome.
The future I hope to see is not one in which humans are removed from the advertising process. It is one in which AI creates more possibilities, enabling people to contribute more ideas and creativity, make the best decisions with a real sense of responsibility, and pursue a higher standard of quality.
The best future is not a world in which AI simply mass-produces safe, generic, and undifferentiated outputs. It is a world in which AI strengthens human insight, creativity, and commitment to performance more than ever before.