Supercharge Ad Creation with PredictiveDesign℠

Powered by Havas Helia

Our team is data-driven. Predicting consumer behavior allows us to target the right customers with the right offers. To achieve this, audiences are segmented and personified, and each uniquely identified audience receives personalized content — content that resonates. Additionally, we design in-market tests, measure campaign results, and adjust strategies and tactics to continuously improve performance against KPIs.

But we wanted to go further. To take our data-driven approach to the next level and drive even better creative performance, we partnered with Havas New York to innovate a new approach for leveraging data and insights. Before testing creative concepts in-market, we are able to analyze an audience’s receptivity to different ads through consumer test panels; the ads are then broken down to figure out which characteristics drive the most positive response. We call this new methodology PredictiveDesign℠.

How does it work?

Two inputs are needed for PredictiveDesign to work:

1. A sample of at least 20 ads with different concepts to provide a variety
2. Each ad’s success metric obtained from consumer testing, such as favorability rating or purchase consideration

PredictiveDesign℠ then breaks down four dimensions of ad characteristics to see how they are associated with the success metric, giving us insights into which ad elements drive performance, either positively or negatively.

1. Keywords — meaningful terms used in the ad’s script that may or may not appear across the sample of ads. We extract these using text analysis powered by IBM Watson.
2. Content Elements — components of an ad, such as setting, music, callouts/banners, use of branding, etc., that describe an ad’s composition.
3. Talent Elements — variables that describe the actor(s) appearing in the ad, such as the number of speaking roles, demographic characteristics (gender, age, ethnicity, etc.), and other features.
4. Emotional Scoring —Unstructured text is converted to structured data points including the ad’s sentiment and intensity across the five principal human emotions: joy, fear, anger, sadness, and disgust. IBM Watson is used again to evaluate the ad’s script for these more nuanced insights.

What do we learn?

Through PredictiveDesign℠, we can identify the ad characteristics that are proven to have a significant correlation with the metric of interest. These characteristics will be a subset of the four ad components and help us understand which ad traits drive better — and worse — receptivity among consumers.

Figure 1: Relationship of ad traits to consumer receptivity

At the same time, we can also gauge the power these ad elements have to predict the metric of interest. To what extent is the variation in that metric explained by the ad’s measurable components? Only if ad components explain movements in the metric of interest effectively will we be able to use PredictiveDesign℠ to estimate the performance of other ad concepts.

Figure 2: In this example, the model predicts consumer response very accurately, with only 5% of the variation in the KPI being unexplained.

How can our clients benefit?

Testing ad concepts in a controlled environment before launching them allows us to efficiently assess performance, only implementing the ads that are most likely to resonate with consumers. PredictiveDesign℠ takes this efficiency a step further, allowing us to capture an ad’s components through its script and run them through PredictiveDesign℠ to get a forecast of ad performance.

PredictiveDesign℠ can also help guide ad creation by informing strategic and creative teams about what does — and doesn’t — resonate with target audiences. Such insights can help influence the creative development process as it evolves. The program can also estimate potential performance without formal testing and provide guidance on the value of creative elements.

While PredictiveDesign℠ is impressive, it can’t replace the creative team. For one, it’s no good at small talk in the elevator. And on top of that, it can’t actually build creative spots from scratch or come up with witty ideas. Nor can it explain why creative elements matter in consumers’ minds or give insights into creative elements that have not yet been tested. This is why it’s important to incorporate new ads into the sample to refine the PredictiveDesign℠ model over time so that it can dynamically and effectively support creative decisions.

Ready to get started with PredictiveDesign℠? Get in touch with us at hello@havashelia.com.