Attention part 2: Practical implementation with optimization models

Pavan Appaiah N, technical consultant, Data and Analytics MiQ

See here for Attention part 1: Future currency of digital marketing?

Collecting and scaling attention data

Attention data is a human-centric measurement, obtained through the participation of opt-in panelists. Using webcams on various devices equipped with eye-tracking and facial coding software, these panelists view different ad variations allowing scoring based on gaze and duration.

As attention metrics gain traction, major players like Lumen and Adelaide have created their own attention models. Given the absence of a standardized definition for attention, the methodologies employed by each attention vendor differ, yielding distinct attention scores tailored to their respective methodologies. These models are subject to ongoing refinement by vendors, which includes expanding panel diversity by incorporating new languages and adapting to emerging ad formats.

Yet, attention modeling is far from straightforward. It incorporates an array of metrics, including ad format, screen dimensions, view duration, scroll speed, ad clutter, page layout, domain type, and numerous other signals, all treated as independent variables. These variables are subsequently used in a multivariate scoring system to predict the likelihood of attention. The resulting extensive dataset is harnessed to train machine learning models, facilitating scaled attention measurement across the web.

Using attention for custom bidding

Attention scores are useful across programmatic campaign management, with a primary focus on day-zero planning and mid-flight campaign optimization. Custom bidding scripts automate mid-flight optimization bidding strategies, specifying what media, context, or audience an advertiser values most. Attention is the latest dimension in these strategies. Some vendors provide pre-calculated attention scores at the domain, creative, channel and device levels, while others require campaign impression logs for analysis. Attention scores at these levels are used to create custom bidding scripts by indexing the top segments and they can also be combined with campaign KPIs like CPM, CPA, CTR and CVR. We use multi-criteria decision algorithms like TOPSIS for the indexing of top segments.

Here’s how it looks in practice when the campaign goal is CPM:

1. CPM and attention scores are combined and websites delivering the highest impressions with the high attention scores at the lowest media cost are ranked higher.
2. If the minimum bid price is $0.80 CPM, all segments (websites) obtained from the above ranking step are scaled between e.g. 1 and 1.8 using scaling techniques like MinMaxScaler.
3. We assign bid modifiers so that high-performing domains are bid at higher rates e.g. $0.80 multiplied by 1.8 CPM.
4. The campaign is optimized to prioritize attention while taking the advertiser’s target CPM into account. This process is repeated at regular intervals for 10 days or until campaign completion to achieve peak performance.

Campaign insights

Data providers like Adelaide and Lumen can integrate with a DSP to track campaigns and generate attention scores for each impression delivered. These scores are based on factors like ad format, screen real estate, view duration, scroll speed, ad clutter, page geometry, and domain. With impression-level attention data, we can uncover campaign insights for performance strategies, temporal trends, geographical performance, and contextual analysis, providing valuable insights into the optimal times, locations and contexts for maximizing outcomes.

Campaign performance case study

Optimizing for attention led to a significant improvement in campaign KPIs for a European transport company. We saw a 28% reduction in CPA and a higher total CVR than we achieved with the best-performing campaign strategies or DSP strategies. Additionally, optimizing for attention reduced the CPM by 32%, whilst delivering the same number of impressions as other line items and campaign strategies. At MiQ, we optimize for attention alongside KPIs like CPA, CPM, CTR, and CVR, but we also heavily use it as an actionable metric for brand lift and improving upper-funnel metrics.

Attention beyond display: YouTube, CTV, and ESG ?

YouTube
YouTube ad measurement uses methods like eye-tracking and contextual tracking, considering player interaction, pixel presence, scroll behavior, location, and time. Trained attention models predict YouTube ad scores for campaign optimization. Analyzing these scores helps fine-tune targeting strategies for various line items.

CTV (Connected TV)
CTV is the digital world’s fastest-growing platform, with a projected $34.49 billion spend by 2025. Although more expensive for advertisers, CTV offers less ad clutter, resulting in higher average attention scores. Data is captured through ACR (Automatic Content Recognition) tools, combining facial recognition, eye tracking, and room presence detection to generate attention scores for various ads. Over time, these models can derive attention scores from ACR data alone, improving brand inventory selection.

ESG (Environmental, Social, and Governance)
Optimising for attention can slash carbon emissions from digital campaigns. Playground XYZ found that 40% of online ads go unseen despite being MRC-viewable (MRC viewable display and video ads are when 50% of pixels are in view for 1 and 2 seconds respectively). With attention models, we can exclude low-attention domains, cutting campaign emissions by up to 63%. Attention scores empower marketers to balance performance and carbon footprint strategically.

Conclusion

Attention is a promising but still nascent metric that holds immense potential for campaign optimization using machine learning and advanced analytics. It’s a robust metric with diverse applications, making it uniquely suited to cross-channel expense comparisons and ROAS justification. Attention closely mirrors genuine audience behavior, representing the likelihood of being seen, and resulting in real engagement and positive business outcomes.

To learn more, watch MiQ’s attention webinar. Our panel of experts discussed industry updates on attention, its correlation to brand lift metrics and the need to make brand campaigns more dynamic and actionable.

Pavan is a technical consultant in MiQ’s Bengaluru-based capabilities team. Outside of work, you’ll find him playing something — whether that’s chess, cricket or FIFA!