Artificial intelligence (AI) is transforming the production and delivery of advertising campaigns. But while it can automate tasks and generate more creative assets, it also raises an essential question: how can we benefit from its capabilities without unnecessarily increasing the environmental footprint of advertising?

Clearly, generating ten versions of a banner only to test two of them uses resources, including for versions that will never be shown. To understand the relationship between AI and the carbon footprint of advertising, we therefore need to consider both production emissions and emissions from ad delivery.

In an opinion piece published by Minted, Vincent Villaret, CEO of Impact Plus, suggests assessing the usefulness of these tools against the resources they consume. A team might therefore ask whether it really needs to produce more variants or whether it could achieve more by choosing the formats and placements it tests more carefully.

Why can AI increase the carbon footprint of advertising?

From producing creative assets to delivering ads, a digital campaign relies on creative tools, media buying platforms and infrastructure that processes and transmits data. When these services are purchased from suppliers, the associated emissions may fall under the advertiser’s Scope 3, which covers indirect emissions across its value chain.

AI models also require computing resources, both for training and for use. At campaign level, the resources required depend in particular on the model selected and how often it is used.

The risk arises when generative AI is used to multiply the number of banner or video variants to test. Producing more creative assets guarantees neither greater attention nor better media results. It can, however, increase computing requirements, even if some variants are never shown. Once the ads are live, the volume of data transmitted varies according to factors including file size and the number of impressions.

Which uses of AI should be tested to reduce emissions?

The relationship between AI and the carbon footprint of advertising depends on how the technology is used. In his opinion piece, Vincent Villaret suggests three ways to limit emissions, from selecting placements to designing adverts.

1. Reduce wasted media through predictive analytics

Predictive analytics can help identify placements that generate high emissions relative to the attention they attract. It can also be used to examine certain sites known as Made for Advertising (MFA), which are primarily designed to generate advertising revenue. Before excluding a placement, its emissions and media results should be assessed.

The aim is to reach the intended audience with fewer unnecessary impressions. To assess the impact of changing placements, it is useful to compare emissions alongside a campaign-relevant metric, such as cost per completed view.

2. Choose AI models suited to the task

For a well-defined task, such as classifying creative assets by their message, a specialised model may be sufficient and require fewer computing resources than a large general-purpose model. This approach, known as “algorithmic frugality”, means choosing a tool proportionate to the need. Comparisons are most useful when they examine the same task, with comparable output quality, while accounting for the number of requests needed.

3. Design lighter advertising creatives

Reducing the file size of a banner or video can cut the amount of data transferred if the version ultimately delivered benefits from that optimisation. When AI is involved in creative production, the brief can set file-size and resolution limits from the outset. Testing two versions in the same placement over an identical period then makes it possible to compare delivery-related emissions, visual quality and the number of completed views.

AI and the carbon footprint of advertising: which metrics should you track?

To determine whether an adjustment reduces emissions, it is useful to compare a baseline version with an optimised version over an identical measurement scope. Three metrics can inform this comparison:

• Total emissions, to check whether they fall across the measured scope.
• Emissions per 1,000 impressions, to compare the carbon intensity of ad delivery when volumes differ.
• A media performance metric, such as cost per conversion, to check whether the change preserves the intended outcome.

Impact Plus helps advertisers and agencies bring these data together to identify the formats and placements where they can test emissions reductions.

Towards AI that supports more responsible advertising

The relationship between AI and the carbon footprint of advertising should be assessed at campaign level. If optimisation reduces emissions from ad delivery, it is still important to check its upstream impact, particularly the computing resources used to produce creative assets. Comparing these factors with media results helps determine whether a particular use should be adopted more widely.

Vincent Villaret proposes a clear shift in perspective: the advertising industry would benefit from assessing its tools according to their real usefulness, resource consumption and contribution to more effective campaigns. By combining carbon measurement, creative optimisation and media choices, resource efficiency can become a criterion for managing campaigns, alongside performance.

Source: Vincent Villaret, “How AI can help tackle the carbon footprint of advertising”, Minted.‍