Contextual Ads: Winning in a 5G World
5G offers a unique opportunity for advertisers to serve contextual advertisements in ways never seen before.
Like each previous generation of wireless networks, 5G is set to fundamentally change the way we connect and do business.
A snapshot of the key markets as of Q1 2021 shows South Korea outpacing most 5G providers with median download speeds of more than 400 Mbps, followed by Taiwan (360 Mbps), Saudi Arabia (317 Mbps), and the UAE (285 Mbps). Many early adopters and countries with large 5G programs, including China, Germany, and the US, are building the requisite infrastructure to break into the top-ten list. At this rate, 5G will enable faster data streaming than ever before, paving the way for rapid digital advancements.
In advertising, this surge of data will enable deeper insights into customer behaviors and motivations, allowing companies to develop targeted, hyper-personalized ads at scale — but just migrating to 5G is not enough to enable these enhancements. To get this perfect mix of customer experience and media monetization, advertisers will need a technology framework that harnesses various aspects of 5G, such as small cells and network slicing, to deliver relevant content in real-time with zero latency and lag-free advertising.
Contextual Video Ads Set to Gain
A recent study shows that 86% of businesses used videos as their core marketing strategy in 2021 compared to 61% in 2016. A report by Ericsson estimates videos will account for 77% of mobile data traffic by 2025 versus 66% currently.
To drive this growth, brands need a scalable workflow that will allow them to:
- Redesign data repositories to publish content and optimize it with analytics on the edge.
- Optimize bidding processes by leveraging the enhanced speed and data processing offered by 5G.
- Customize media preferences and boost engagement rates/monetization.
- Meet regulatory guidelines by protecting content distributors from piracy/legal issues.
Successful digital marketing also requires an end-to-end strategy encompassing behavioral and contextual analysis of user data.
Behavioral data analysis uses sophisticated technology to analyze user data — personal identifiable information (PII), location, demographics, device graphs, multi-touch attribution — and provide an accurate description of who a consumer is. However, behavioral targeting has taken a big hit due to broader data-privacy trends in the post-GDPR era:
No more cookies
With GDPR and CCPA in place, Google Chrome will phase out third-party cookies by late-2023. Firefox blocks them by default. Google anticipates revenue for the top 500 ad publishers to reduce by 52% on average.
With the launch of iOS 14 in September 2020, apps must request permission from Apple device users before tracking their actions across third-party apps and websites. Apps can no longer access the device’s Identifier for Advertising (IDFA), which advertisers use to understand user behavior within and across apps. Today, close to 35% of iOS users have Apple’s Limit Ad Tracking function turned on; this is only expected to increase further.
As regulations on third-party data continue to increase, brands need to adapt by increasing their use of contextual analysis for targeting. Contextual analysis tracks the type of content individuals view in real-time to predict what they might be interested in given their present frame of mind. Someone live-streaming a football game may see advertisements for sports memorabilia or promotional offers for game tickets. Contextually aligned ads are more engaging, making the experience more memorable and driving up purchase intent.
Leveraging 5G with AI for Contextual Ads: A Look at the Architecture
The diagram below illustrates a multi-layered architecture that combines natural language processing with computer-vision technology to create relevant, brand-safe ads.
Section 0: Device Edge Components
- API discovery consists of all the contextual factors that are triggered to drive a hyper-localized ad experience (e.g. weather, traffic, location). The campaign management engine (downstream) is expected to automate ad-spend notifications to match fluctuations in these parameters. For instance, advertisers can refer to the weather API to promote a local ice cream shop more on hot days.
- Enrollment assigns a unique identifier to every connected device that is part of this experience. This is a three-step process:
- Onboard devices to a device-management server.
- Assign users and contextual parameters (defined at the API discovery layer) to each device for policy-driven ad targeting.
- Deploy deep-learning models to the enrolled devices, each model specialized for a use case. With today’s technology on cellular devices, the models can process videos at 2–4 frames per second. As end-user devices and hardware technologies evolve, processing to the edge will likely increase.
Section 1: On-premise Edge Components
End-user devices typically do not have the capabilities to handle the entire processing logic of the necessary framework. On-premise edge is a shared infrastructure that allows user devices to vertically scale their compute capabilities, adding a layer outside the boundaries of the device that can execute the necessary functions. On-premise edge has the advantage of proximity to the end-user device, reducing latency for processing. Such an infrastructure can also prevail as a fallback for devices with older hardware.
For advertisers, on-premise edge refers to nodes responsible for aggregating, storing, and processing data from 5G small cells set up at on-site locations such as shopping centers, offices, hotels, or large living spaces.
- Content-streaming server delivers both linear and live-streaming video content to consumers.
- Business logic is where the bulk of the media decoding and processing is expected to happen. As users consume media, a neural network extracts patterns from input data samples. Running in real-time, this processing engine performs the following actions:
i. Parse the text, metadata, and image captions of the media using natural language processing — using an ontology-based framework, the system can enhance information retrieval and resolve any natural-language ambiguities.
ii. Parse the main visuals of the media using computer vision — this can include making sense of the logo, face, emotion, or activity in the analyzed visuals. Images would be parsed into visual patterns to effectively disambiguate low-level segmentation cues such as shadows and low lighting.
iii. Combine the textual and visual analysis, and key contextual parameters assigned for each device, into one cohesive report that can be sent to the ad mapping engine.
By atomizing and distributing training and inferencing workloads on the edge, businesses can achieve objectives like guaranteed URLLC (ultra-reliable low-latency communication) with mobile access networks delivering 99.999% of packets within 1ms or better (as defined by 3GPP). This can ensure dynamic video ads that change their storylines based on a viewer’s responses. Advertisers can also localize content distribution rights to adhere to local regulations. “Edge computing enhances the distributed cloud paradigm,” says Nida Sahar Rafee, founder of Nife Labs. “Think of making the most relevant decisions closer to the end-user — at the network edge.”
Section 2: Network Edge Components
Network edge, which includes the campaign management engine, is a shared infrastructure that manages multiple on-premise edge networks (both physical and virtual), usually operating for a city or town. A single network edge can house multiple on-premise networks spread across the country, each with its topology and business rules. For instance, Brooklyn may house a network edge with multiple on-premise edges across the county mapped to it, and the network edges for Brooklyn may have a centralized data center cloud component representing New York.
The network edge houses the infrastructure to localize ad recommendations and ad intelligence for geography. Components include:
- Segmentation and win-rate estimation: A continuous feedback module that analyzes historical data (ads served, impressions bid, impressions won against campaign budget constraints) and estimates probable win rate for potential ads that can lead to impressions.
- Ad mapping: Analyzes information received from the business logic layer — using advanced AI/machine learning (ML), this information is reconciled against a brand’s campaign criteria to map contextual segments relevant to the targeted audience.
- Device ad sync: Ensures that data and preferences are synced across devices — with the right approach to modularize device enrollment, advertisers can ensure that settings are synced to every participating device in a specified location.
With full 5G capabilities, time from media processing to ad notification is expected to drop from 2–3 seconds (average 4G LTE data transmission rates) to an instant (real-time). This is particularly relevant in cases of location-based advertising (LBA). Say a customer is watching a movie clip wherein the character is eating a burger from a fast-food chain. If the customer is subscribed to that brand’s LBA program, they can receive a mobile notification about the nearest store offering a discount on the burger displayed on the screen. 5G networks will also provide the infrastructure to scale up to wider geographical areas without compromising on latency.
Karthick Loganathan, CTO of Bipolar Factory, says that, despite being flooded with user data, advertisers still struggle to utilize that data effectively and in real-time. “5G’s network slicing capabilities would move the ad-tech industry closer to achieving a 100% fill-rate without compromising eCPM.”
Section 3: Datacenter Cloud Components
- Master model optimizer imports training data from multiple edge instances (each locality, building, or office can have an edge instance running) to optimize a central repository of ML models. These models are optimized:
For space — with conservative topology transformation.
For maintainability — through a unified code pathway.
To be hardware-agnostic — through dynamically loaded plug-ins.
Once the models are optimized centrally, a feedback loop ensures that the models on the edge are continuously updated with the latest annotations. This improves the accuracy of the inference engine while optimizing ad spends.
What is in it for the stakeholders?
- Advertisers and media owners
Real-time insights from the content analysis provide advertisers with more precise data sets, enabling better targeting for more relevant ad content to viewers on every single impression. Media owners, on the other hand, can use dynamic feeds to better monetize their video content. With a dynamic feed, watching an adventure video can trigger a GoPro ad for one demographic and a vacation package for another. Such strategies can reduce the cost per view and increase click-through rates while fully complying with brand safety guidelines and local regulations.
A distributed, cloud-based architecture can open multiple revenue streams to service providers. As the full functionality of 5G becomes mainstream, service providers can set up virtual network slices and dynamic service-level agreements (SLAs) that correspond to slice utilization, latency, or backhaul congestion. Through the right mix of B2B and B2C ecosystem participants, service providers can also offer a pay-as-you-go OpEx spend, fueling broader adoption among enterprises.
- For consumers
A global survey by Ericsson discovered that 16% of consumers are interested in 5G-fixed wireless access (FWA) offering to replace or supplement their existing home broadband. Similarly, consumers say they are willing to pay a 10% premium for a 5G plan, and early adopters as much as 20%.
5G can end up delivering much more immersive ad visuals to consumers than what is available today. Network providers estimate 5G will load pages, apps, and ads with less than 10ms latency even after processing ad tags, header bidding, verification, attribution, and servers. These speeds will allow brands to provide a more personalized, seamless overall consumer experience. AI-based gesturing is already a part of high-end smartphones. 5G infrastructure will allow more points of tactile feedback to be processed in real-time, offering greater personalization to consumers, whether at their homes or on the go.
Enablers in the Journey Ahead
- Video Codecs
Most video codecs (AVC, HEVC, VP8, VP9) today can reduce the bitrate of uncompressed video to less than 1% of the original rate without any noticeable visual quality degradations. Next-generation developments including versatile video coding (VVC) are expected to improve compression efficiency, enable low-delay video coding, and enhance support for immersive video, guaranteeing required performance levels for media services over 5G networks.
- 5G-enabled Smartphones
Mobile vendors like Apple, Samsung, Sony, and Motorola began launching their 5G-enabled flagship models in the first quarter of 2020. With players such as Qualcomm and MediaTek offering performance-optimized 5G-ready chipsets, 2022 could be the year when mobile vendors extend their offerings to the mid-tier cost segment. IDC expects 5G smartphone shipments to grow to 69% of the global volume in 2025 (accelerated by the success of Apple’s fully 5G iPhone 12 lineup).
- Commercial 5G Launches
The most extensive 5G commercial launches have been in the US, China, South Korea, and Switzerland, mainly in larger cities. A quick look at the global 5G map shows 84.9 thousand commercially available 5G deployments, meaning a network is present and consumer devices are available for use. As the global economy gradually re-opens, spectrum auctions will likely increase, leading to an uptick in 5G subscriptions, extended network coverage, and greater speeds.
“5G and mobile edge computing (MEC) are impressive technologies in their own right,” says Jason Leigh, research manager of 5G and mobile services at IDC, “but achieving their full potential requires thoughtful strategic integration. Combined, 5G and MEC supercharge the ability to gather and analyze consumer behavioral and preferential data to deliver personalized, actionable contextual advertising.”
Data is worthless without connected intelligence. The convergence of three transformative technologies — 5G, AI, and edge computing — offers application providers the means necessary to build the next generation of distributed applications. Adopting a new paradigm of highly accurate, probabilistic data models and 5G can ensure on-the-go inferencing of a consumer’s current frame of mind and interest while guaranteeing incremental revenues for broadcasters, operators, and D2C providers. It is only a matter of time. Arriving at a critical mass of technology and infrastructural capabilities guarantees that the entire media ecosystem’s long-term health stands to gain.