Machine-generated creative from Pencil
Direct-to-consumer advertisers pumping tens to hundreds of thousands of dollars into their Facebook ads each month know the power of a good creative message.
The best ad creatives drive ROAS (return on ad spend) that can be orders of magnitude higher than a failed message.
But how can busy, short-staffed marketing departments test new creative messages constantly? Tweaking messages involves both inspiration and grunt work, which often prevents marketers from testing as often or as extensively as they want to.
Enter Pencil, which launched in November after two years of technical development and with $4 million in pre-Series A venture capital funding.
Pencil intelligently pairs up a brand’s taglines, images and videos to create dozens of new messages a marketer can choose to test. The AI involved can sometimes make strange choices, but just as often creates “aha” moments for brands looking for new ways to market their products.
Case in point: A toothpaste brand testing Pencil found that AI-formulated creative messages that focused on the mouth, instead of a white smile, ended up being its top-performing creatives. And a dog food brand found that ads mentioning the taste of its product – despite the fact that humans don’t eat it – were top performers.
Going against conventional wisdom, Pencil’s AI might produce a message where the copy runs over a model’s face, Pencil CEO Will Hanschell posited.
“There is a naivete that we encourage,” according to Hanschell. “Any creative would say no to having text in front of the face. But we know people are drawn to eye contact. Does that make you more or less likely to read the text? You might find that an ad like this actually performs better.”
How the tech works
Pencil’s process works like this: A brand uploads their brand assets, like photo shoots and videos, their color scheme and taglines. Then the AI will ingest all that information, and spit out dozens of possible creative messages, along with scores for how well it thinks each message will perform. Brands are advised to pick three to five new ad creatives to test each week.
Marketers then export the machine-generated ad creatives, usually to Facebook. (Right now, Pencil is most closely integrated with Facebook, in part because of its strong APIs, Hanschell said.) When Facebook’s API returns data about how the ads performed, Pencil’s AI can learn how well the different ad creatives work each week and get smarter.
The AI also learns from what ads the marketer picks. So if the marketer keeps testing bold-colored backgrounds, for example, Pencil will surface more schemes from that palette.
Building Pencil’s tech required solving lots of small problems, Hanschell said – for example, reformatting a creative so it doesn’t cut off faces in an Instagram story, editing video or generating ad copy. The machine had to handle it all.
Performance: man vs. machine
Pencil’s AI picks include hits, but also misses.
In a test of 14 brands from August to December, on average brands doubled their ROAS brand baselines. But results were unpredictable. Two brands never found an ad that beat their brand baseline. And of the remaining 12 marketers in the test, a couple only saw 4% and 6% gains, while the two top performers boosted their ROAS by 334% and 576%, respectively.
“This is a hits business. It’s about finding winners, and putting your money behind that one,” Hanschell said. Even the brands that couldn’t beat their results might eventually land on some high-performing ads, he said. Since Pencil’s goal is to make the creative process ten times less work, brands can afford to do more aggressive testing of their ads.
Pencil also keeps its AI tech siloed at the moment, so learnings from one brand don’t enrich another. But that stance may change. “Some clients are up for pooling their data so they get better predictions,” Hanschell said.
Making Pencil’s tech ethical
For all the promise of AI, the technology is so new it can be hard to fathom some of its potentially negative effects. Pencil published its own AI ethics policy when it formed.
To put those ethics into practice, Pencil developed a misogyny detector to nix ads that disparage women. It also filters out ads with negative sentiment – “even though they’re incredibly effective and quite standard” in non-AI ads, Hanschell noted – because it doesn’t want to prey on peoples’ vulnerabilities to sell products.
Pencil doesn’t want its tech to put people out of work, which is the second part of its ethics policy. Instead, it wants to make mid-size marketing teams more productive, and do “upstream work,” like creative strategy, Hanschell said.
Pencil’s tech is also part of a shift that’s happening as people use more predictive, AI-driven software, he said: “The next generation will be much more familiar with working with software that has ideas of its own.”