How our optimization engine made YouTube video ads a 5x more profitable sales channel for 10 Brands

It’s always an unbelievable feeling for a product owner to see what you have been working on delivering exactly what you have intended it to do.

While building VideoRunRun, we had many convictions and the support of several brands, during the last two years, who understood the business potential of YouTube but were confronted with the operational reality that makes that goal tricky to master.

With initial learnings so encouraging, we intend to make this publication a quarterly one with the additional learnings of the coming campaigns.

Why look at YouTube as a sales channel?

Users are spending more and more time on digital and OTT platforms. The trend started years back, and brands are starting to grasp the importance of Youtube as a business driver, not without a few struggles though.

But are brands equal in leveraging the opportunity when it’s about solving a complex and balanced formula of creative, audience, call to actions, landing pages, and tools and know-how that take us more comfortably towards the branding route?

Well, finding the right balance to move the sales needle using YouTube video ads has been an experiment we’ve been conducting in the last four years. It’s what the VideoRunRun beta platform stemmed from

So, from Q4 2019 and early Q1 2020, we ran ten campaigns for brands of multiple industries (direct to consumer, retail, entertainment, tourism, services) selling product between 30€ and 500€ per cart.

The goal was to check and confirm the best approach to deliver direct outcomes using muti-creative testing and continuous optimizations of the media plan for a specific business goal.

As our goal was to highly leverage the business contribution potential of YouTube as an acquisition channel, we ran the campaign to mainly non-exposed users who were not clients (90% of our reach), and did so in two phases.

Here’s what we’ve learned.

First major learning – Creative rules all.

Phase 1 – Pre-testing of the video to validate the best possible option

In order to get to a stage where we’re confident to scale our beta-clients campaigns, we needed to determine the best possible combinations between :

+ Videos (between 3 to 12 tested videos)
+ Landing Pages (between 1 to 3 tested redirections)
+ Call to Actions (between 1 to 5 tested for every creative)
+ Audiences (more than 20 tested every time from top to bottom funnel)
+ Devices (always mobile vs desktop and even TV)
= that equates to testing between 3 000 to 8 000 sub campaigns combining all these variables.

There is the one you snack on your mobile, the one you consume on your desktop and what you experience on your connected TV. All three devices coming with their own media consumption context and setting.

You also get to address audiences, and with signals coming from most of the google ecosystem products, YouTube’s targeting and audience segmentation is incredibly granular.

Finding the combination matrix (creative asset, audience, device, CTA, landing page, time of day…) is what we have put our mind on for this first phase, by putting these against sub-campaigns each other using for two weeks and, VideoRunRun’s scoring algorithm determined the best combination and the following learnings.

“Creative accounts for 47% of your campaign performance, exceeding brand, targeting and context combined.”

• Creative is key. And performance can vary of up to 475% between the best and the worst combinations. We knew that the creative was a massively important factor in delivering brand recall, intent and sales. Nielsen Catalina ran a study that took it even further, making creative the largest contributor in driving sales.
  We confirmed the same on YouTube.
• Performing audiences aren’t necessarily the ones you would have suspected. We confirmed that scale doesn’t necessarily come from suspected (bottom of the funnel) potential buyers, and that’s the beauty of a totally agnostic approach. By testing the 3 levels of the funnel audiences, we could generate 45% of the sales from top of the funnel audiences and another 45% from middle of the funnel ones, reducing bottom of the funnel (non-scalable) audiences to only 10%.

Without surprise, with an average performance variance of 475% between the worst and best combination options, looking at 8000 sub-campaign in a completely neutral unbiased way helped identify “winners” and pockets of growth impossible to reach without process automation.

One thing that is counter intuitive during this first phase especially for brands who are used to Facebook or Instagram, was to accept NOT to optimize anything and gather precious data and insights. In order to act upon to scale effectively, later.

Second major learning – On-going profitable scale is possible.

Phase 2 – Outcome-based continuous optimization of “always-on” creative combinations

We entered phase 2 with up to 8000 combinations per campaign, and two weeks worth of performance data, with the top deciles outperforming bottom ones.

Thanks to the VideoRunRun optimization algorithm, not only the target cost-per-acquisition (CPA) was reached within 2 weeks, but the platform managed to reach CPAs that were much lower, at a pace that surprised us.

“For the Le Slip Français campaign on Q4 2019, 60% of the incremental sales came from middle of funnel audiences and 40% from top of the funnel -highly scalable- ones”

• Top funnel scale is there if you know how to get it. Scaling is allowed when we turbocharge delivery on a. the top two deciles, exhausting available scale within our client’s target CPA, and b. on the top two to four deciles, opening more audience sets when we were below the client’s target CPA.
• CPAs optimization are a marathon, not a sprint. Out of the ten campaigns, 8 decided to continue as an on-going campaign letting the algorithm pursue it’s optimization and at lower but still effective pace.

For these height brands, within the following two months the CPAs declined even further with an average of -75% vs. start of campaign CPAs.

Conclusion

Running phase 1 and phase 2 led to unlock pockets of scale that couldn’t have been on our radar, or at a painful unsustainable almost hourly review and optimization of our delivery combinations, what we have done is just create a feedback loop to gain maximum knowledge in phase 1, that we used wisely in phase 2, a phase that unlocked the ability to sell at large volumes at top funnel non-clients.

With an average in ROI of 5x and 1000 incremental sales after only three months of campaigning and scaling, the results are very encouraging.