It was difficult to find the next topic to write about since the possibilities are endless with AI and marketing. This post started out as a guide on how to use AI to make your own 3D influencer. Now, after Googling ‘ai influencers’ I jumped into a rabbit-hole of mind-f*$k.
Virtual influencers range from cartoon-like to uncanny valley to animalistic and have millions of followers. The GEICO Gecko, Duolingo Owl, and the DAVE Bear are examples of in-house virtual influencers.
There are already agencies that represent virtual influencers:
Did I miss the train completely or are we just getting started?
Virtual influencers and Virtual Youtubers (VTubers) are increasing in popularity as the younger generations spend more time online. If you spend any time on Twitch.tv you might notice VTubers on the rise.
Open World found that VTubers now represent nearly one-third of all livestream watchtime on YouTube, and that within the past 18 months, “we’ve seen a 400% increase in watchtime of English-speaking VTubers,” the company says. - source
VTubers are matching the demand and cranking out content. See below for VTubers’ video uploads and hours streamed over time.
According to this report by Influencer Marketing Hub, virtual influencers are trending up at a faster rate than ‘real’ influencers. I can’t confirm if their data was cherrypicked to prove a point but at face value this is an important trend to note.
The data suggests the market has an increasing appetite for virtual avatars on their screens. Look, I don’t get it either, but let’s pretend we’re David Attenborough and observe from afar. This is a judgement-free zone.
If you are in growth marketing, you know that an influencer’s follower count has little to no correlation to direct-response goals. Influencers that are easiest to work with who create performant UGC videos have fewer followers. Large influencers can have a positive lift on your brand recall though measurement is difficult.
Our feeds are inundated with UGC ads that all follow similar concepts:
3 Reasons/Signs Why
Before and After
Infomercial
How-to
Storytelling
Skits
Generative AI starts to unlock more storytelling possibilities:
“Before and after wearing a Canada Goose jacket on top of the Himalayas sitting on a rug”
“Comedy skit while drinking Athletic Greens in the middle of the Amazon rainforest riding an elephant”
“Storytelling while doing a space walk sipping a nice cup of coffee with a chimpanzee talking about the benefits of ZipRecruiter”
Practical things for Growth Marketing
leverage - verb. use (something) to maximum advantage.
Use text2image to create several ‘influencers’, text2speech to create variations of voices, text2image to change backgrounds. This allows you to test different influencers along with different voices.
Portrait animation software already allow you to upload an image and audio clip. The results are not perfect but the novelty to competitor’s basic UGC ads could make the difference between a successful campaign versus a dud. They will look better if the facial features are distinguishable (e.g. no hat brim right above eyebrows, no bed of flowers around face).
Text2speech options on the market are decent. I am patiently waiting for Google to release their Duplex AI. Remember their release from May 2018 where the AI called a restaurant to make a reservation?
For the VTuber/anime aesthetic, you can use VRoid.com to create your own avatar. I might write about that more in the future.
Should you test virtual UGC ads? Here are some pros and cons:
Pros:
You eliminate the week-long turnaround time with creative briefs and revision rounds for faster concepting and iterations.
Social media feeds are not used to seeing virtual UGC ads (yet), so performance is better.
More storytelling capability with environmental and wardrobe settings.
Cons:
Portrait animation results are not perfect.
Outputs are not human and watchers approach the uncanny valley.
Comment section might be full of confused people.
You could reframe this positively because more engagement → more algo love → lower CPM.
Creative Testing Framework
Here’s a framework you can follow to test virtual influencers on Meta. I personally use this framework for creative testing so you can follow it to test real influencers, scripts, or new concepts:
Set up a Meta campaign for creative testing with ad sets targeting a broad audience in your country (e.g. US) optimizing for your specific bottom-funnel KPI (e.g. purchase, install, lead, etc).
Each ad set will contain up to 4 new creatives grouped by hypothesis for testing. For example:
Ad Set 1 - Virtual Influencer test
Virtual Influencer 1, ‘3 Reasons Why’ script
Virtual Influencer 2, ‘3 Reasons Why’ script
Virtual Influencer 3, ‘3 Reasons Why’ script
Virtual Influencer 4, ‘3 Reasons Why’ script
Ad Set 2 - Influencer & Script test
Real Influencer 1, ‘Before & After’ script
Real Influencer 1, ‘Skit’ script
Virtual Influencer A, ‘Before & After’ script
Virtual Influencer A, ‘Skit’ script
Let each ad run until it gets at least 30 conversions. Ideally, you reach 50 conversions because that is when Meta’s algorithm exits the ‘learning phase’. However, you will have a good understanding of creative performance before then. You will likely have one ad that is taking up all the spend while the others are sitting around. This is a good indication the ad gets better engagement than the rest in the ad set and can scale.
Pause poor performing ads and copy winning ads that meet your performance goal into your business-as-usual (BAU) campaigns. You can continue running the ad in the creative testing campaign so there are no performance fluctuations in your ad account. It’s common a winning ad doesn’t get any spend when copied into your BAU campaigns.
Repeat steps 2-4 at desired tempo.
There are other creative testing frameworks out there that each come with their own pros and cons:
Creative test campaign in a cheap country, like India, with an upper funnel objective first
New creatives go straight into BAU campaigns
The main takeaway is to follow a framework. You will also need to construct your creatives in a way that allows you to measure performance of each element. I covered that briefly in this post (excuse the click-bait title):
Synthesia.io, D-id.com, Tokkingheads.com, and Movio.la already allow you to create video ads with preset 3D avatars and voices.
Do you think AI UGC ads will be a competitive advantage for DR campaigns?
i love the idea of AI unlocking more storytelling ability - it's like a greenscreen on steroids.
im not AS sold on V-UGC - simply cause the whole point of UGC is it's authentic-ish. the virtual aspect of this neuters that, no? i guess we'll have to see.