Traditionally, there has been a tradeoff between having access to a daily-updated brand lift model, and having to wait until mid- or-post campaign to get a model update. The daily model jumps around so much, it’s not conducive to making informed decisions. On the other hand, having to wait weeks or even months for updated model data means missed opportunities for campaign optimization. Upwave now delivers the best of both worlds: daily model updates without erratic shifts in data through the Daily Automated Reads feature.
Using Daily Automated Reads, updated brand lift models are provided daily, paving the way for quick and nimble, data-backed ad campaign decisions.
No more brand lift model tradeoffs
Obtaining regularly updated control models to help inform campaign optimization strategy has traditionally been a tricky business. Advertisers only get updated control models mid- or post-campaign, which is far too late to make campaign adjustments. Innovations from Upwave’s Machine Learning (ML) team have removed that limitation.
- Traditional brand lift measurement: With traditional brand lift measurement, weights may be updated nightly, but control models (which select the features on which to weight) are only updated when mid-campaign reads are requested. The more time that goes by between updates to the control model, the more the baseline, and thus the brand lift read, will fluctuate.
- Upwave’s automated brand lift modeling: Using cutting-edge causal ML, Upwave can provide daily-updated control models (i.e., brand lift models). In contrast to the limitation of traditional methods that can result in erratic model behavior day-to-day, Upwave applies techniques to help avoid “overfitting” the model, thus keeping it “smooth” (see sidebar below). This smoothing technique means the model isn’t overreacting to new data, and empowers advertisers to make informed decisions about their campaigns at any time.
Upwave’s ML team began applying a technique called Bayesian Information Criterion (BIC) to smooth daily model changes. This upgrade enables advertisers to get the benefit of daily model improvements without the erratic impact to lift measurements that otherwise results from model changes. BIC is unique in that it balances predictive accuracy (the traditional criterion for feature selection in models) with model parsimony, thus limiting the number of model changes to those most likely to be applicable as more sample data is collected.
Building and training smart brand lift models
Each brand lift model has two parts: the features (e.g., income level, gender, age) and the weighting of the features—that is, how much a feature impacts the model. Upwave data scientists have begun building smart brand lift models that essentially “learn” as more data is collected. As the model learns what features and options seem to impact brand exposure the most, it automatically adjusts features and model weights to provide advertisers with better data. This data can then be used to adjust campaigns on-the-fly. Here’s an example of how it works:
Flower Fresh was running an ad campaign to help drive awareness of their challenger brand within floral arrangements. After 3 weeks, the brand noticed in the Upwave dashboard an early read that exposures on flowers-dot-com was driving awareness more than their contextual buy. Their agency account lead was skeptical, however, and told the brand to wait until the mid-campaign report read-out, as their previous brand lift partner waited until the read-out to update the control model used to calculate the baseline of the lift-measurement. Consequently, the 3-week read in the dashboard was based on an out-of-date model. The agency analyst spoke up and explained that Upwave actually updates brand lift models every night, and uses ML to avoid erratic overreactions to new data, so shifting budget from the contextual buy to flowers-dot-com was a reasonable optimization to make.
Find out more about the Daily Automated Reads feature to make better-informed decisions about your ad campaigns—when you want.
