• by Laurie Sullivan  @lauriesullivan, Yesterday

Carbon footprints of advertising models lack access to clear emissions data and environmental impact when brands use generative artificial intelligence (GAI) for buying, serving, searching and purchasing.

Sustainability models for GAI had been built on assumptions rather than data, but that will change with a recent initiative announced by The Brandtech Group. The company has made an undisclosed “major” investment in Scope3, a company dedicated to decarbonizing advertising and AI.

The two companies jointly released a white paper examining the environmental impact of using generative AI to create ads and marketing assets. 

Brandtech Group also created a carbon calculator to help marketers understand the impact of GAI to help change their path to lessen emissions. 

“We started with websites pages and articles, but plan to expand into video and other formats that weren’t live when we began this project,” said Rebecca Sykes, Brandtech partner and head of emerging technology.

Users enter four data points, and it calculates consumption of electricity and emission, and translates the data into common tasks that people can relate with, such as driving half a mile or taking an airplane trip.

To estimate emissions, the team analyzed tasks, model size, model popularity, and more to combine with average training and runtimes to determine the impact for a specific action — real-life advertising and marketing scenarios.

For the study, the team chose to analyze emissions generated by automating common marketing tasks using GAI models -- the generation of a 1,000-character-long product page copy and the generation of 20 static creative assets for a marketing campaign, for example.

The calculator assesses the energy consumption (kWh) and GHG emissions (kgCO2e) of executions of a list of GAI tasks, deployed with a particular model version and type of output.

Pencil Pro, a Brandtech model aggregator for AI content generation, was used as a data source to understand typical workflows and to refine the hypotheses based on live cases.

Creating one product page with descriptions on a web or ecommerce site for a brand’s entire product inventory generated very little carbon impact, with a total of 4 gCO2e generated.

But for a company creating 1,000 products yearly, the impact of using GAI for these product descriptions is a few kgCO2e, roughly the impact of a meat-based meal.

For companies that are looking to automate further by creating multiple versions to match language or cultural specificities -- for example 25 variations -- the resulting impact is around 100 kgCO2e, equal to around 1,500 hours of video streaming.

For the biggest retailers, it is estimated that 300,000 new items could be released every year, published in 50 different versions. If GAI were used for the related product pages, it would result in over 50 tCO2e emitted. That is roughly 30 round trips from Paris to New York.

U.S. data centers consumed 3% of the nation’s power in 2022. By 2030, this figure is expected to rise to 8%, according to the white paper. This means those data centers will require about $50 billion in new electricity generation capacity, which is why companies like Google, Microsoft, Meta and others are making investments in alternative energy sources.

Goldman Sachs Research estimates a 160% increase in data center power requirements by 2030, up between 3% and 4% from between 1% and 2% in 2023. The firm also believes the social cost of these emissions between $125 billion and $140 billion based on today’s value.