“All models are wrong, but some are useful,” is a famous quote from British statistician George Box, dating all the way back to 1976. In today’s world, where AI tools are automating tasks like never before, it’s even more important to remember that while AI technologies are helpful, they have limitations.
Even as AI and robotic machines grow stronger - with the rise of chatGPT and other AI writing systems seemingly threatening to take over even creative positions- there are still limits to what this technology can do.
What does this mean for the world’s oldest form of advertising? When planning an outdoor advertising campaign strategy, data science, and technology are powerful tools. Still, these tools must be managed with the creativity and strategic approach of the invaluable human touch.
In this article, we’ll look at exactly what AI can and cannot do in OOH and show why a strategy that blends the human touch with advanced machine learning technologies is the best approach to creating engaging and successful outdoor campaigns.
“AI is not limitless. It can only recombine what it already knows,” Billups CTO Shawn Spooner said. “Anything for tech in Out-of-Home has to be built with that in mind.”
All AI machines are probability machines. If you can find the boundaries of their capabilities, you can break them. They are beholden to the data used to train them. Machine learning models are always different because of the randomization in how they are trained. Every model is different, even with the same codes and trained on the same data. Even when they behave the same way, it is for different reasons, and it can be challenging to know why.
“AI is great for telling what works, but not at saying why. In certain ways, they are like airplane black boxes. They need a human to interpret the information,” Spooner said.
But Spooner also said that there are certain things that AI and advanced machine learning excel at - anything that can be pattern matched.
“It’s really good for automation. You’re telling it what to do, and it can excel on its own.” Spooner said. “However, it struggles when there is a strategic element involved. What these systems are really doing is large-scale pattern matching. They can get arbitrarily complex and complicated in how they string facts together, but none of the models available are creative in the way humans are.”
With the randomization of elements, AI can be used to determine a creative, and eventually, it will get really good, but it’s not true creativity. The model is just doing permutations until it comes up with something decent.
“They can create, but it's not creativity - it’s a combination,” Spooner said.
Beyond that, all of these advanced machine learning models have trouble encoding long-term history sequences of relationships. As humans, we can understand long chains of causality, whereas AI models have finite time horizons. They can learn a lot but have limited capacity.
Another issue is that it takes a lot of CPU power to make things happen. Using these models is computationally expensive. To run a model as complicated as chatGPT takes a huge building and training process, as well as a lot of energy and expertise to build and maintain the system. To build AI for OOH, it would need a purpose that makes sense in the industry.
A considerable limitation of using AI in OOH is that vast amounts of quality data must be available. AI is only as good as the data that it is fed. But we don’t have that because no one has been collecting and storing the data digitally until more recently.
“For someone to make a robust AI model for Out-of-Home, they would need one clean data source, and it is a burden to bring that together in a meaningful way,” Spooner said.
Dmitry Semenov, Head of Innovation at Billups, said that while historical campaign data for the past 20 or 30 years is not available, much of that data is now being collected. While it’s too late for campaigns of yore to provide insight and analytics for ideal campaign scenarios, the future is moving towards this.
“Since we are collecting this data now for all of the campaigns, this information feeds into the AI and will help us with more accurate planning in the future,” said Semenov.
Can AI make creatives? While abundant AI software is available to make images, the quality of these images varies, and creativity is limited to what was created before. Without human intervention, an AI graphic could easily be a copy or imitation of fast creatives that have been fed into the model. This could be an issue if an AI-generated creative for a brand looks too similar to a competitor’s past creatives. Though AI can generate ideas, relying upon it for your creatives in either art or content is not a great idea.
“When it comes to brand strategy and real creativity in terms of how we use OOH and play around between the creatives and the environments, you need the human mind, intelligence, and creative spirit to create the art,” said Semenov. He explained that it wouldn’t necessarily be suitable for making Billboard embellishments and isn’t set up at this time to look at the interplay between a unit, its location, and the environment in which it sits to create the most influential creatives.
Where AI can help and help a great deal is with campaign planning. Machine learning works with modeled data, analyzes the data, and learns from the results.
“It’s continuously evolving,” Semenov said. “It can suggest the best locations to meet the goals, and it can increase the efficiency of the planning."
However, even then, there are limitations. "Obviously, we can ask ChatGPT about the goals and KPIs for an OOH campaign, and we would get a very general answer, but if you go deeper to specifics, I wouldn't trust it," Semenov said. "Things like brand strategy and brand safety require the human mind and years of experience."
Semenov said the best role for AI in OOH is in campaign planning, setting a budget, and predicting results.
At Billups, we do a lot of Machine Learning (ML) and Advanced Machine Learning. However, we do have a few prototypes that could be considered AI if we consider AI to be systems that interact with people directly.
We can look at how campaigns work and when campaigns burn out and need fresh creative. We can understand what’s in the physical environment affecting visibility. We can mimic the human visual system and give a good recommendation on where to put your creative, but we don’t have sentient systems.
We consider ourselves towards the forefront of OOH technology. We are one of the few, if not the only, OOH companies with seven patents approved and eight more pending. About three or four of those patents are truly deep ML. The rest have ML elements but are more mathematically based.
“We are always looking to improve our offerings to help us be ahead in the OOH measurement space,” Billups Sciences Director Chen Chen said. “We try to find things that the industry is not doing yet and find a way to do it.”
There are different stages in OOH advertising. At Billups, we believe AI and machine learning can be well-used in planning. A good example is our Recommendation Engine. Given a campaign budget and the desired demographics, our recommendation system can generate a suite of plans for our clients. What we are trying to improve on the recommended plans is to include the personal touch from individual planners, as they have the niche knowledge of our clients, the local markets, and the suppliers. By incorporating the feedback from the planners into these automated campaign plans, we can improve the system for future recommendations.
The Recommendation Engine takes information on the client’s budget and requirements, including demographics, to generate a basic plan. The planners then provide their input and feedback to tailor the campaign and improve the system for future recommendations.
“It really learns the individual’s habits and preferences and comes out with unique propositions for each client,” Chen Chen said. “In that way, we can really free the planner from part of the work. They can focus on other things that will make our clients happier.”
The feedback the recommendation engine receives from the planner on its outputs lets the system know what the planner likes and doesn’t like and what they would do differently in a campaign. Eventually, the system learns to make recommendations like an individual planner, learning both planners’ preferences and planner’s preferences by client.
“Planners may think they are just being replaced by the system, but that’s not the idea,” Chen Chen says. “The planners work really hard. On top of performing strategic planning, they need to maintain communications with our clients and suppliers. What AI and machine learning can help us achieve is to use the recommendation systems to do the heavy lifting of campaign planning, and the planners can direct their focus to providing better service for our clients.”
Here are some of the other research-oriented and machine-learning projects in which the Billups Sciences team invests its time:
.jpg?width=1200&height=900&name=SSI_VISUAL%20(1).jpg)
Example of SSI index form.
Sicong Chen, Billups Staff Scientist, says there’s much more to be done in the future regarding AI and machine learning in OOH. However, some capabilities will just never be replaced by AI.
“Talking with the machine or chatbot is incredibly human, but we all know that it's not the same as talking to a human,” Sicong Chen said.
Clients and planners value their relationships with one another. The machine cannot replace that part.
Even if one day the client could use AI to understand the market and make suggestions for campaigns-even if the machine provides perfect results, at the end of the day, it is just talking to the bot, and there is no relationship there.
“That part I don’t think will ever be replaced,” Sicong Chen said. “But planners can use all these tools so that they will spend less time building spreadsheets and plans and less time worrying about whether a unit’s price is fair.”
The importance of local knowledge and personal guidance will not go away with advanced technologies. Tools can help planners reduce manual work but will never replace the relationships between humans. There is more value when there are planners alongside the tools for consultation to help understand a proposed plan and why it will work.
Even in terms of personal preferences, Sicong Chen says, “I don’t like talking with the chatbot. I want to reach a real human.”
In terms of what Billups has in store for the future, we are working on delivering “live exposures” before the end of 2023. Clients who choose this service will be able to see how their campaign is performing within 2-3 days instead of waiting until post-campaign for results. Clients can watch their units daily to see trends in reach, exposure, and frequency change over time. The dashboard will show clients' results with just a few days of lag time, allowing them to track performance and make necessary adjustments to their campaigns, such as swapping out messaging or creative.
Billups blends art and science together to create campaigns that are better than using either strategy on its own to help you build effective campaigns efficiently and effectively. Contact us today to learn more about how we leverage our machine learning technologies combined with boots-on-the-ground experienced media planners to help you create impactful campaigns.
