Accenture tries AI-built adverts to eliminate the pre-roll

Accenture is developing a new advertising solution in hopes of helping brands battle ad-blocking. The solution works by essentially inserting subtle advertising in the form of product placement into video assets. But what’s new is that the company claims the platform is able to do this on the fly, automatically, and in conjunction with a programmatic ad exchange. The product is being developed by Accenture’s agency arm, Accenture Interactive (AI).

“Our main focus is on integrating our experience marketing content with commerce for global clients, for brands that range from BMW to Marriott, Unilever,” said Alex Naressi, managing director for R&D at Accenture Interactive. “We’re at the intersection of human trends, new business models and technology.”

The new advertising solution, Naressi said, is the result of two trends: the rise of ad-blocking on digital video platforms, and the desperate need for content owners to find new, non-interruptive ways to monetize content. “We assumed that people would not tolerate interruptions anymore, so were prompted to imagine a future where interruptions wouldn’t exist,” Naressi said. “One way for brands to still be able to create impressions is to do something that’s been around for quite a while,” he said, referring to product placement within content. “The idea is to bring some flexibility and personalization to product placement inside video content,” he said.

The solution would work essentially like native advertising for video. Instead of periodic ad spaces within the video file, the technology overlays subtle product placement advertising within the video itself. The specific brands or products that are inserted would be tailored to the audience, whether that be for large audiences watching broadcast content, households watching on connected TVs, or to specific viewers watching on mobile phones and other personal connected devices.

Back in around 2004, Massive Inc and Double Fusion both attacked this space not inside video, but in video games. More recently we have come across Mirriad in the UK doing something similar in video. The in-game systems failed perhaps because they had to identify ad positions in games as they were being built, perhaps two years ahead of revenue. Mirriad, which launched in 2013 shows video on its web site where someone is clearly drinking coca cola, and in the next version shown in Asia, the same person is drinking a local brand. Sounds like Accenture is building something similar.

The platform claims to use artificial intelligence to complete the product placement automatically, using machine learning to run simulated eye tracking programs that follows where the action is on the screen and create what Accenture calls “attention heat maps.”

Naressi claims the technology is able to not only overlay brand names and logos onto things like generic looking airplanes or clocks, but it can also virtually place objects onto surfaces that weren’t part of the initial scene.

“We built all of the logic that allows us to identify the hotspots where we can place content, specific surfaces, posters, billboards, planes, tables where we can put objects,” he said. “You could watch a reality TV show with some wine in the kitchen, we can actually place mobile phones on the table, we can change appliances in the kitchen, and do that dynamically on the fly,” Naressi said.

The first step is to identify where within a given shot or scene can the platform place a product. The system works for both existing and brand new content. “For existing content, we use a combination of AI and human moderation to scan videos and find hotspots,” he said. “For new content, the [content creators] have the ability to place markers where they want placements to appear.”

The heat map indicates where on the screen the viewers’ eyes are likely to be, based on what’s going on in the scene and its setting. “We developed a way of doing that automatically,” Naressi said. “It’s not 100% accurate, but it’s good enough to get an initial sense of a value of attention for a zone onscreen.”

By determining what surfaces can be used as product spots, and where the viewer is looking, the AI solution then inserts a product placement into the video. Naressi said the platform is able to execute in some challenging environments: in one demo, the platform is able to place a company name on the side of airplane, which is moving frame by frame, and is also partially obstructed by a tree at one point. It can also insert text onto surfaces while respecting orientation changes and depths of field, he said.

The workflow would look something like this: the content owner uploads the video to the platform, and AI’s solution begins what Naressi called “placement enablement” – mapping out where the ads can go.

“New content can be made placement-ready using special markers; for existing content, we perform an automated deep scene analysis using AI,” Naressi said. “We discover potential placement placeholders (hotspots) and then manually validate those placeholders and fine-tune visual effects to blend placements into the content in a photorealistic way. The solution uses machine learning to gradually ease the visual fine-tuning. We call this step placement placeholder optimization.”

Next, the solution determines the pricing of the ad spots. “We pre-score the placeholders depending on their proximity to the centers of attention,” he said. “We use a form of AI to simulate human viewing behavior and determine attention heat maps for any kind of content automatically.”

As an advertising solution, Naressi said product placements that appear closer to the on-screen action would be valued more than product placements that occur, for example, in a corner of the scene where nothing is happening. The platform would then take this into consideration to inform pricing of the particular product spots. It can also use semantic analysis to ensure contextual relevance and brand safety for the marketing message, he said.

The actual ad decisioning and placement occurs next, which is where Accenture’s platform would need to interface with a marketplace or exchange to connect with marketers. And we’d assume this advertising solution would use very specific ad formats. So far, Accenture has not partnered with any third party digital advertising marketplaces or demand-side platforms.
After the ad bidding occurs, the platform performs the actual rendering of the product placement and then distributes that video file to the end user.

A potential bottleneck will be the video player. “For a truly personalized dynamic placement that is processed on the fly, a specific player will be required,” Naressi said. “That being said, the solution could also pre-render variants of videos (e.g. one per country or customer segment) and use existing video platform players, however, this approach is more limited in terms of personalization.”

And the final piece of the advertising solution is verification. “We can track which placements occur and for how long those are being displayed,” Naressi said. And while there’s no way to tell whether or not the viewer actually noticed the product while watching the video, Naressi said the attention heat maps can also act as an imperfect predictor of impact of the ad. But if these are dropped in on the fly from a remote server, we’re not sure how they avoid the attention of an Ad blocker?

The solution was conceived as an alternative to advertising interruptions, but it’s easy to see how brands and marketers might want to use this technology in conjunction with traditional ad spots within a video to deliver a one-two advertising punch to viewers. Naressi said the solution could also implement clickable ads as a supplement to the product placement.

While the idea of rendering personalized product placements within a video file on the fly is certainly impressive, the solution is still in the very early stages of development, and as such, a lot of the solution is unproven – particularly the business case for it, and especially its effectiveness in terms of brand lift.

Accenture unveiled is new ad solution at Cannes earlier this summer. The platform has a number of patents pending, and the company hasn’t yet integrated the solution with any digital advertising platforms, nor does it yet have any customers. “Our objective over the next month is to prove the content from a business standpoint and explore different ways we can bring value to the advertisers and content producers,” Naressi said.

The biggest question Accenture will have to grapple with is whether product placement delivers better ROI for brands than a more traditional and obtrusive pre-roll or mid-roll advertisement? And of course, the other big hurdle will be proving to content owners that all that technology sitting between the platform and the viewer doesn’t in any way impede or hinder the viewing experience. Better to MPEG stich the ads in, at the server level, before the video is sent across a network. When streaming video online, even the widespread (and less technologically challenging) dynamic ad insertion platforms frequently experience hiccups in performance which leads to buffer wheels and stilted viewing. But because the ads are placed before the video starts, the biggest challenge will be keeping the video start time. Research has shown online viewers tend to abandon videos after only two or three seconds of buffering.