Background and objective
Cyber Communications Inc (CCI), owned by Dentsu Group, faced a serious challenge. Campaigns it was delivering for 500 ad agencies across 1500 websites in Japan were affected by fraud, brand safety violations and lack of viewability.
“In the Japanese market, increasing caution relating to ad fraud, and ensuring the increase in brand safety and viewability are becoming hugely important," said Akio Niizawa, Dentsu CCI's representative director and president.
At the same time, first-generation ad verification providers whose role it is to tackle these challenges were found to be highly ineffective.
In the first instance, existing solutions were “post-bid”. This saw verification partners only measuring problems after an ad had been served, with detection reports only issued after-the-fact. In the case of ad fraud, this required a lengthy and complex process seeking to convincing publishers to provide reimbursement for fraudulent impressions, while for brand safety violations — the damage was already done.
The second type of verification, “pre-bid", involved verification partners scraping ad-inventory and pooling "safe" inventory into a pre-bid segment, then targeting this safe segment to avoid negative exposure. However, CCI found this hurt scale on campaigns — the non-real time aspect of scraping and indexing effectively excluded the newest and most popular inventory from the pre-bid segment. This measure also failed to account for the fact that URLs often change, so a page that was previously classified as safe, might not be safe at a later date.
“The most common means of blocking bad placements is to rely on past data via a demand-side platform, but the gold standard is a real-time adjustment,” said CCI division manager Ad Platform Business, Shigehiro Ando.
CCI decided they needed to implement real-time ad verification, and selected cybersecurity company Cheq to execute this, due to its ability to block threats in a matter of milliseconds using AI and cybersecurity.
Cheq deployed 700 proprietary parameters, including honey pots (bot traps), advanced OS fingerprinting and deep learning algorithms to block fraudulent requests, and trained its AI engine with tens of millions of documents to understand Japanese language, the nuances of Japanese cultural and brand sensitivities. It established 39 categories of brand safety in Japanese.
In contrast to blunt keywords-based decisions, Cheq's AI claims it can recognise a concept however it is phrased or wherever it appears and block brands from appearing next to unsuitable or toxic content with 95% accuracy.
The cyber firm also worked with CCI to implement header tags across tens of publishers in the CCI network, which the software analyses on an ongoing basis.
CCI hired marketing research company Macromill, which tracked 3395 consumers who were served ads on publisher sites. The study compared the experience of consumers seeing ads with Cheq’s AI brand safety and ad fraud-prevention enabled, against sites without this ad verification.
The results showed that click-through rates increased by up to 250% when Cheq was in place, and ads were served next to safe content, compared to publishers without this protection.
Overall in this period Cheq analysed 2.6 billion ad requests for CCI. Of these ad requests, 7% were blocked for fraud (183 million impressions) before any ad was served or money paid. Meanwhile, across 2.4 billion ad requests, nearly one quarter (24%), were blocked on the grounds of brand safety (totaling 576 million ad requests).
Cheq stopped brands from appearing next to potentially disastrous articles, including executions of women for affairs, incest, and animal cruelty.
It also increased ad serving opportunities to sites that were previously off-limits, including Goo, a Japanese web portal that was formerly blacklisted but was brought back into media plans after Cheq showed that 80% of its inventory was in fact safe.