[What You Need to Know Now] The Latest on DMP (Part 2)

How will the advertising and marketing industry, rapidly converging with technology, change in the future? A young planner at Dentsu Inc. working on digital marketing explains the latest insights centered on "Data × Marketing."

Continuing from Part 1, focusing on DMP implementation and operation, Katsuaki Sakai of Legoliss, Koichiro Kondo of Dentsu Inc., and Taku Arakawa of Dentsu Inc. discuss their frontline trial and error and future outlook.

※Part 1: "The Latest on Audience Data (Part 1) – Things You Can't Ask Now"
※Part 2: "The Latest on Audience Data (Part 2)"

The Creativity Needed to Bridge Advertising and Data

Arakawa: Moving forward, the breadth of data we handle will undoubtedly expand. This means we'll need people who can drive DMPs forward and make groundbreaking proposals using DMPs.

While data analysis existed in traditional marketing, understanding technology is especially crucial in ad tech areas like DMPs. I might be lucky having come from a technical background, but without a fundamental grasp of the technology, the scope of proposals naturally becomes limited. Keeping up with the constant stream of new technologies and services is also essential. Furthermore, to master these technologies, you must consistently make decisions based on data.

Even within the same advertising industry, I suspect this is an area that creative professionals often find difficult to grasp.

Kondo: When discussing ad tech, people who seem capable fundamentally understand the boundaries of what a given technology can and cannot do. Those who blindly believe new tech can do everything rarely succeed.

Sakai: When I first entered the ad industry, it was full of people asking, "What's an IP address?" But now, there are more young professionals strong in both data and technology. On the flip side, those who can read the numbers often tend not to look at the creative side. It's like the common talk of right-brain vs. left-brain types – people often assume you can't be good at both.

Kondo: It's disappointing when you have a data-savvy person who's segmented the audience perfectly, yet for some reason, there's only one creative piece and no differentiation at all.

Arakawa: So, are there really few people who truly understand both technology and marketing...?

Sakai: Data can be deceiving, right? Depending on how you slice the report, you can potentially arrive at completely different analysis results. That means even looking at numbers requires "sensibility." Unless you approach data with a hypothesis—knowing what you want to prove or discover—you risk being manipulated by the data.

Kondo: Analysis is precisely the kind of work where results vary significantly based on individual intuition and experience. How you evaluate or interpret a figure doubling, for instance, depends entirely on the lens you apply and the context leading up to that point.

Regarding DMPs, the same principle applies to segment creation. If you're marking visits to common, unremarkable pages as audience segments, you must factor in visit frequency. Conversely, for highly specialized pages, it might be better to isolate them into a separate segment.

Sakai: Even with third-party data seller-type DMPs, when creating data segments for sale, they initially list sites users visited using URLs or meta keywords, somewhat automated, from a perspective like "audience interested in ●●."

Even if you pull in a large chunk of data at that stage, there's often a manual process next to trim away the impurities. When you visually inspect the resulting list, you sometimes find URLs that don't fit the purpose, like "What is this URL?" For that reason, as someone who works with data daily, it's impossible to fully trust data generated solely by systems.

Kondo: It's the same with tools. In reality, the fundamental skills of a marketer and the ability to think about how to represent segments using web behavior history and other data are far more important.

What do integrated advertising agencies look for in a DMP?

Sakai: Regarding how to interpret data, I recall an experience from when I handled hotel marketing. During analysis to decide new locations, I was reviewing reservation lists. The hotel industry is unique—reservation data often doesn't match actual guest data. So, I thought I needed to see the actual guests firsthand. I parked my car in the hotel lot, stayed overnight, and observed the actual customer base (laughs).

And then, among the people I saw in the parking lot, there were actually some who clearly weren't telling the truth when they checked in. For example, the car's license plate number didn't match the address they wrote down, or you'd think, "Ah, this person is on a business trip and came with a rental car." There are things you only see when you go to the actual location.

That experience really drove home the point I mentioned earlier: you can't just trust data because "data lies."

Kondo: That sense that you won't understand unless you actually see or go somewhere is deeply ingrained in integrated advertising agencies. This is especially true for the creative department; it's standard practice to visit the store if the client's product is there, observe how it's being purchased, or actually try it out yourself.

Sakai: Based purely on my experience, internet-only agencies back then rarely visited physical locations where products were displayed or bought and tested products themselves. They also often assumed user surveys were biased while believing web behavioral data never lied. In reality, both surveys and behavioral data inevitably contain arbitrary responses or clicks.

This ties back to an earlier point: it's not about whether you're specialized or integrated. What's crucial is that analysts, when faced with biases revealed in data, have the ability to go to the actual site to verify, to hold things in their hands, to feel and think critically.

Regarding talent, we're increasingly consulted by not just brands but also ad agencies asking, "Teach us about DMP!" or inviting us to study sessions. In such settings, I often ask, "What do you think is the mission of those managing ads?" The typical response is something like, "It's all about boosting performance!"

When I then ask, "So what exactly is performance?", they usually pause and say, "Um..." I explain that since the job of an advertising agency is to create ads and ensure their message reaches consumers reliably, it's all about moving consumers. In other words, I believe the mission is "to engage with consumers with the goal of 'moving' them in mind, skillfully using data and technology to fine-tune the methods of contact every day."

Defining "consumer action" also becomes clearer when you break it down: ad clicks, brand awareness, product favorability, purchase intent.

If you blindly focus only on the final conversion metric, it narrows your operational scope. During that earlier study session, I also asked, "If you were to quantify 'emotional engagement,' what KPIs would you set?" Quantifying emotional engagement is actually difficult, but people who chase conversions day in and day out are so fixated on click-to-conversion rates that they simply don't think about such things.

Kondo: Well, first, you can refine impressions like viewability. There are also tons of areas you can tweak, like refining your target audience or eliminating ad fraud.

Sakai: Exactly . Just focusing on whether you're reaching the right audience effectively and tuning for reach instead of conversion can significantly boost performance.

Why Customize Data Instead of Platformizing It

Sakai: By the way, why doesn't Dentsu Inc. hold its own data?

Kondo: Because if we owned our own data, selling it might take priority over our clients' interests. I believe the solutions we should truly provide are gathering and proposing exactly what clients need each time. So, by not holding data ourselves and preparing what's necessary on a project-by-project basis, we can deliver the most optimal solutions.

That said, if we had truly proprietary assets that directly benefit clients, we should hold onto them. But honestly, we haven't found any magic data like that yet.

Sakai: I see your point. The landscape of what's needed in data marketing changes constantly. Legoliss doesn't have its own product either, but our scope has expanded significantly—from consulting and data visualization to building clients' core systems, not just handling the advertising part.

Precisely because the scope is broadening, there's a choice to deliberately not own products, or to choose to own them. I think this is where strategies diverge between companies.

Arakawa: The difference between a data provider and an advertising agency lies in their ability to identify the right tools for a client's specific challenges and to build them flexibly.

Regarding data ownership, what are your thoughts on third-party data? While it's an effective means to access targets unreachable by first-party data, it still faces accuracy challenges and inherently lacks uniqueness due to its third-party nature.

Sakai: As mentioned in the first part, while relying solely on cookie data based on web browsing history won't be perfect for third-party data at this stage, it's certain that we need to start challenging ourselves now, looking ahead to the era of data circulation. Moving forward, diverse companies will provide data in diverse ways. If what we call and use as third-party data is only cookie data, that's problematic. We're also entering an era of apps where cookie-based tracking of behavior is limited.

Right now, I think the best approach is to gather several types of third-party data and build our own customized database. This might be closer to what's called second-party data than third-party.

Kondo: If data value is evaluated by accuracy, volume, versatility, and uniqueness, the weakness of third-party data is that originality is often overemphasized, leading to underutilization. In that sense, exploring second-party data and unique ways to combine our own original data will become increasingly interesting.

Arakawa: From an advertising agency perspective, leveraging second-party data is relatively straightforward due to existing relationships with media partners. For example, we've combined multiple segments to identify individuals who hold negative views toward a specific branding campaign.

Just focusing on the "dislike" segment doesn't yield useful insights. But by focusing on the segment with high purchase intent, you can see specifically which aspects aren't being appreciated – you can hear the "voice" of the buying audience. This kind of fine-tuning is becoming quite possible by combining several data points.

Kondo: That's right. Another recent trend is when someone says, "We need a segment like this," and we realize we have to create proprietary data for it. Such requests are gradually increasing.

Sakai: That's precisely why we're entering an era where swapping out cartridges, like with game consoles, becomes crucial.

Having lots of cartridges is more fun and helps you get better at games, right? But if the core DMP isn't built, having a wealth of cartridges is meaningless. How well we can build data that fits us perfectly will be key to DMP success going forward. I'm looking forward to seeing what Dentsu Inc. does next. Thank you for your time today.