We sat down with the Facebook Disruptors team, the internal team that focuses on high growth, venture-backed startups, to press them on how they can make Facebook effective for B2B. They confirmed many of our hypotheses. Read on for our main takeaways.
1. Algorithms need a lot of data.
To allow Facebook to truly optimize a direct-response campaign, the Facebook team says you need to spend $50K/day for a week. Yikes!
The real truth is you need to get to 50 conversions per week per ad set, which requires a large audience and a big budget.
2. You can get stuck in learning mode forever
Since very few B2B companies hit that threshold, most B2B advertisers are perpetually stuck in Facebook’s learning phase every day of every week of every campaign.
3. Direct Response is not the right objective for smaller audiences
Awareness/traffic is often the right campaign objective if you have an audience <10K. The campaign will get more reach as the estimated action rate is higher. (For how to create high performing small B2B audiences, check this out.)
Remember bid x estimated action rate is how the auction works.
Direct-Response action rates are 0.2% for B2B vs 1–4% for ecomm. An ecomm advertiser will almost always win out against you in a DR auction.
4. For audiences <10K, don’t let Facebook optimize off of conversions
B2B advertisers should still track conversions from traffic/awareness campaigns, but not allow Facebook to use those conversions for optimization, otherwise you’ll constrain reach.
If you’re running traffic campaigns with small audiences it’s up to you to decipher what optimizations to make. Don’t count on the algorithm.
If your audience is big enough, by all means let Facebook’s algorithm do its work. Set up the correct conversion signal and watch the leads roll in.
Lookalikes off of 400 super high quality leads will oftentimes outperform a lookalike built off of 5000 disparate/low quality people.
The model is looking for similarities between users and sometimes it is easier to find more similarities with a smaller, higher quality dataset.