Fully Optimized and Automated Ecosystem of Facebook

Kalle Tiihonen May 20 2015 9 AM | 5 min read

Ever wonder how optimization algorithms such as oCPM or lookalikes work? The first part of our series described Facebook’s budget pacing algorithm. Now we reveal the secrets of the oCPM bidding and how to use automation in general for competitive advantage in the ecosystem.

Facebook's Latest Algorithms at Your Service

Did you know that Facebook® knows people better than anyone else? The amount and quality of data is their competitive advantage: they don’t just know your location, age and mobile phone model but also when you fall in love. Both lookalike and oCPM algorithms utilize approximately 100,000 different attributes per person to find the most relevant Facebook users for your marketing campaigns.

Using Facebook’s Lookalike Audiences, lets you find new people who are likely to be interested in your business. It helps you reach people who are similar to your current customers, page fans, mobile app users or website visitors. If you want to build a lookalike from your CRM data, the algorithm requires emails, phone numbers, Facebook user IDs or mobile advertiser IDs.

elvis lookalike

Elvis lookalike from Norway

From CPM to oCPM – Towards Next Generation Bidding

Optimized CPM is the most innovative bidding type invented by Facebook’s engineers. Like the lookalike algorithm, it identifies the people from your broad target audience who are most likely to take an action as a result of seeing your ad – e.g. purchase a product at your web store. However, oCPM needs click and conversion data in order to identify the right people from a large audience.

How the oCPM Algorithm Works

As in all of the Facebook bidding methods, your oCPM bid is converted into a comparative metric called effective Cost Per Mille (eCPM) before the auction. The eCPM can be calculated from oCPM with the following formula: your maximum oCPM bid * estimated conversion rate (eCVR) + quality score. Please note that the quality score is a broader metric than the relevance score.

In the beginning, the oCPM gives the ad some impressions to calculate estimated conversion rates (eCVRs) for everyone. Along with the eCVRs, the algorithm utilizes about 100,000 attributes Facebook knows about your audience. The algorithm predicts the individual conversion rates based on different attributes – such as age, likes, and mobile phone model. For example, young single people using the latest iPhone could have higher predicted probability to convert. Thus, the eCPM bid is more competitive for the people with the higher estimated conversion rates. The eCVRs and eCPMs will be updated as the algorithm collects more conversion data.

The first day of your campaign does matter: achieving high click-through and conversion rates are crucial success factors. The more click and conversion data the campaign gets, the smarter the oCPM algorithm becomes. Please note that the oCPM uses data from the last 24 hours, so the algorithm needs new conversion data every day. In addition, oCPM for actions requires a conversion rate (CVR) of 0,5 % or more with a 1 day click-through window in order to work effectively. Having large target audiences and high bids help to attain this. If you have trouble getting enough conversions, you can try to change the action (e.g. from a web store purchase to a landing page or something else upper at the conversion funnel).

The Optimized Ad Format: Multi-Product Ads

Showcase 3-10 images and links within a single ad unit to direct people to specific locations on your website or mobile app. This feature will optimize the performance of your ad by selecting and ordering the best-performing images and links automatically. Using the new ad format, Multi-Product Ads for Mobile App Installs, to promote its app, our customer Foodpanda increased click-through rates by 180% and reduced their cost per install by 39%.

foodpanda ad

Say No to Manual Campaign Creation

With Automated Ads you can automate ad creation with a product feed, both for prospecting and retargeting campaigns. The feed updates your ads in real-time, and each ad is automatically archived when the offer or product is not available anymore, based on stock information from the feed. New additions and offers in your product feed are automatically added to your campaign as new ads. As our customer Mark de Louw from TravelBird said, “With Automated Ads, we cut the time spent in campaign creation by 75%, saving 15 hours every week." Have a chat with us to learn more. We always offer a 14-day free trial and no minimum contract period.

Predictive Budget Allocation by beta

Our data scientists rebuilt the budget allocation algorithm so that the CPA (cost per action) will be automatically minimized. The optimization goal can be chosen while creating the campaign. The feature continuously reallocates the campaign budget between the ad sets such that the expected number of future conversions is maximized and the CPA level is minimized.

The intelligence behind the feature is based on Bayesian multi-armed bandit method. If all ad sets have plenty of conversions, most of the budget will be allocated to the ad set having the lowest CPA. If some ad sets have so far only few conversions and impressions, we cannot be yet confident about their true conversion rates. In this case, those ad sets get more budget even though their current CPA wouldn’t be so low. The exact proportion is calculated by using Bayesian posterior estimation. To guarantee continuous exploration, a small proportion of the campaign budget is allocated evenly between the ad sets. In the future we will include the predicted lifetime value (LTV) and revenue estimates to maximize the ROI for the predictive budget allocation.

What else would you like to know about the ecosystem, optimization or best practices in the next blog post? Please, comment below and we will fulfill your wishes.

Facebook® is a registered trademark of Facebook, Inc.

   Retail Re-UP: Adjusting to the Next Normal of Social Advertising CHECK IT NOW
Kalle Tiihonen

Read Next