Business challenge

A word’s meaning depends on context, but many traditional content screening tools fail to recognize this—causing a lot of ad-friendly material to be banned unnecessarily. How can publishers avoid this?

Transformation

Using IBM Watson® and IBM Cloud™ solutions, Reach created Mantis—a tool that can understand text and images in context and at scale to see if articles are suitable for advertisements.

Results

Protects

advertisers from placing ads next to sensitive or inappropriate content

Unlocks

a huge amount of new ad inventory for publishers by eliminating the need for blanket blacklisting

Helps

editorial teams gather insight about whether content is ad-friendly before publication

Business challenge story

Spreading the word

When advertisers invest time, effort and resources in a unique advertising campaign, the last thing they want is for their brand to appear next to content that detracts from their message. An ill-placed ad can be seriously harmful—tarnishing brand reputation, deterring customers and, ultimately, damaging the bottom line.

Publishers are acutely aware of the negative impact of inappropriate ad placements and they are proactively applying content screening tools to identify both ad-friendly and inappropriate content at scale. While these solutions work to safeguard advertisers’ brand identity, their lack of precision presents a fresh challenge for publishers.

Terry Hornsby, Digital Solutions Director at Reach, explains: “Current methods for identifying unsuitable content rely on spotting keywords in articles and URLs. For example, any article with words associated with disaster, tragedy and other equally negative concepts will be deemed inappropriate for carrying advertisements.

“However, using keywords to identify brand-safe articles is not an ideal approach. Keyword-based tools use a simplistic rule-based method and generally lack an appreciation of context. As a result, up to 80 percent of media that is actually brand-safe gets marked unsafe. For example, lifestyle articles that refer to ‘fashion shoots’ or sports columns that describe ‘shooting a winning goal’ get incorrectly blacklisted due to the violent connotations of the verb ‘shoot’.”

The brute-force approach of existing tools means that a large proportion of perfectly good, brand-safe ad inventory is mistakenly earmarked as inappropriate—reducing potential ad revenue for publishers. Equally, as these tools are unable to analyze images, content that is considered brand-safe may actually be unsafe when viewed in context with the supporting picture. Traditional approaches to scoring brand-safety are also relatively fixed, whereas the digital world in which brands operate is highly dynamic. 

Reach set out to take a smarter approach. The company wanted to develop a solution that could understand the context in which words, images and concepts are used, then make swift decisions at enormous scale on advert suitability.

Hornsby continues: “We knew that we could reduce the margin of error for bad ad placement and, at the same time, avoid the unnecessary blacklisting of ad-friendly articles. With more precise tools for understanding the content of billions of articles a month, advertisers and media buyers could also tailor ad inventory screening processes to the unique brand identities and dynamic sensitivities of each client.”  

Transformation story

Innovating with AI

Inspired by a presentation from IBM on the transformative potential of AI and machine learning, Reach recognized that these technologies could enable a better way of screening articles for ad-friendliness. 

“We were blown away by the power of IBM Watson and the advanced API-based solutions that IBM brought to the table,” says Hornsby. “It was clear that IBM had the cutting-edge solutions and expertise to help us develop a game-changing digital brand sensitivity solution for publishers and advertisers.”

Reach put together a detailed business case and product plan for Mantis—its innovative content screening tool—and worked with IBM consultants in a garage environment to build a minimum viable product in the IBM Cloud. Together, IBM and Reach used the Natural Language Processing and Visual Recognition features of IBM Watson to build a model that could automatically read and understand the context of articles and images.

To help Mantis differentiate between safe and unsafe content, Reach created machine learning models based on industry-standard taxonomies of blacklisted URLs and keywords. The company then added its own proprietary catalogue of techniques for identifying ad-friendly content. In combination, the AI and machine learning features of IBM Watson enable the solution to classify content by its suitability for ad placement quickly and accurately. Critically, the solution also provides the rationale for each of its decisions, helping human reviewers tweak and improve the decision-making process.

“Throughout the project IBM has been highly supportive—they understand and value what we are trying to achieve and are lending their industry-leading expertise in AI to help us succeed,” says Hornsby. “We feel that we have developed a true strategic partnership with IBM on this project.”

A team of media experts at Reach is currently testing and enhancing its machine learning models, submitting thousands of articles to check the consistency and accuracy of decisions. The company has also integrated the solution with its internal content management system (CMS)—providing further input to refine the machine learning models.

Screening content before it goes to print enables us to make minor edits that could mean the difference between an article being suitable for advertisements or not. This means we further increase the amount of advertising space that we can offer to clients and buyers.

Terry Hornsby, Digital Solutions Director, Reach plc

Results story

Pages of opportunity

Reach believes that Mantis can turn the tide on inappropriate ad placement, helping publishers, advertisers and media buyers unlock more opportunities for profit.

Hornsby explains: “IBM Watson gives Mantis the ability to interpret the context in which words and images are used. This means we can determine which articles are ad-friendly with much greater accuracy at high volume—and without introducing delays in publication. Ultimately, IBM Watson will help us and our partners in the media industry to stop ads appearing next to inappropriate articles and eliminate the need to ban content based on keywords alone.

“The image recognition capabilities of IBM Watson enable us to gain greater insight into the context of articles, and also help us promote safe visual content on the internet—making it a better place for everyone.”

Each user of Mantis will have the ability to input their own blacklisted keywords, images and concepts. Providing a separate instance of the solution for each publisher, advertiser and media buyer ensures they can maintain confidentiality while quickly and easily finding the most appropriate ad space.

Hornsby adds: “After running media content that had been blacklisted by current best-of-breed tools, we were pleased to see that IBM Watson was able to identify content that had been incorrectly deemed unsafe for advertising. The solution also explained why the content was, in fact, safe. Bringing a contextual understanding of content into the decisioning process will be revolutionary for the publishing industry, as it eliminates the need for blanket bans of content containing words with multiple context-dependent meanings. This, in turn, will help publishers to increase their ad inventory and boost revenue.”

By integrating its digital brand sensitivity solution with its CMS, Reach can provide feedback to editorial teams before content is finalized for publication. 

“Connecting our CMS with our new solution powered by IBM Watson has given us the ability to pre-screen articles and other content for ad-friendliness before the final publication stage,” explains Hornsby. “Screening content before it goes to print enables us to make minor edits that could mean the difference between an article being suitable for advertisements or not. This means we further increase the amount of advertising space that we can offer to clients and buyers. Equally, picture editors can ensure that their choice of accompanying image does not change the suitability of the content.”

Alison Davis, IBM Industry Lead for Telco, Media and Entertainment, adds: “By bringing Mantis to market, IBM and Reach aim to improve the advertising landscape for both publishers and brands. It is only by deploying AI with IBM Watson, that we can help Reach detect brand safety and sensitivity issues in real time and help prove how not only safe, but effective content that’s blacklisted today, can be for advertisers. It’s a win-win and we believe that Mantis will scale across the industry in the UK and globally.”

Hornsby concludes: “We’re incredibly excited to bring Mantis to market to help improve the advertising landscape for everyone—publishers, advertisers, media buyers and, most importantly, readers.”  

We’re incredibly excited to bring Mantis to market to help improve the advertising landscape for everyone.

Terry Hornsby, Digital Solutions Director, Reach plc

About Reach plc

Reach plc is the largest commercial national and regional news publisher in the UK, producing and distributing content through newspapers, magazines and digital platforms. The company publishes content for influential and iconic brands such as the Daily Mirror, Daily Express, Sunday People, Daily Record, Daily Star, OK! and market leading regional titles including the Manchester Evening News, Liverpool Echo, Birmingham Mail and Bristol Post.

Take the next step

To learn more about IBM Watson, please contact your IBM representative or IBM Business Partner, or visit the following website: ibm.com/watson

View more client stories or learn more about IBM Watson Platform