Honeybook gets 3.5X VIP paying customers with LTV predictions
Highlights
“Not only have we brought more paying accounts with the same media spend, we brought accounts with higher value. We brought more annual plans (compared to monthly plans). We also acquired teams with a higher number of members for each account”
The Challenge
The Pilot
2. Google: Both groups were optimized for tROAS. The control used a proxy goal (free trial), while the experiment group used Voyantis’ prediction of the likelihood of becoming a paying and activated user (using the payments service).
3. Meta: The control group optimized for tCPA (free trial) while the test group optimized for the likelihood of becoming an activated user (process payments), based on Voyantis’ models.
4. On Meta, Honeybook could further refine targeting to the customer percentiles - for example, the top 50% of the VIP segment and only the top 10% of the non-VIP segment.
“Partnering with Voyantis allowed us to acquire higher-value users at a lower cost. Voyantis predictions are being used as a key decision metric in our day-to-day decision-making”.
How does it work
Prediction Model Generation
Voyantis' AI engine learned the unique nature of Honeybook's users, and the value they will generate over time, differentiated between Honeybook's distinct segments (VIP and non-VIP), and created a custom prediction model tailored to Honeybook.
Privacy First
Voyantis created accurate prediction models using Honeybook's engagement, transactions, and attribution data, all strictly anonymized. We never used personally identifiable information (PII) for predictions, ensuring our clients' data is secure.
Ad Network Optimization
We customize the prediction feed for each platform based on their unique network algorithm setup. For Google, we optimize for value-based bidding and send values through the Server's API. On the Meta feed, we used CAPI to optimize for custom events of a 20/50/80% likelihood of converting to an activated user.
Accuracy Verification
The model is constantly tested and adjusted when needed to provide the most accurate predictions (93% accuracy and continuous improvements.
The Impact
The results of the pilot clearly indicate the advantage of using value predictions for improving campaigns outcomes.
Scale increase
3.5X more high-value VIP customers: Not only did Voyantis help Honybook score more VIP paying customers that also use the paying service, Honeybook was able to reduce its CAC by 50%.
Longer term impact
Google’s Ad engine preferred Voyantis’ targets: after Voyantis predictions were fed back to Google, the engine shifted to optimized almost only according to Voyantis events (95% of the ad spend).
Scalable effect
Highly accurate predictions: the predictions that Voyantis generated were 14X more accurate than the proxy that was used by Honeybook with 93% accuracy when compared to actual results.