NAD Reminds AI Advertisers: Correlation Is Not Causation

A recent NAD decision involving SafelyYou, an AI-enabled fall detection and monitoring platform for assisted living and memory care facilities, highlights an important theme in AI advertising: proving that a technology is associated with positive outcomes is not the same as proving that the technology itself causes those outcomes. 

NAD reviewed a variety of claims about the platform, including claims about fall detection accuracy, response times, false alarms, reductions in falls and emergency room visits, and cost savings. NAD found that many of the claims—including claims that staff are alerted in seconds, that the system generates very few false alarms, and that communities using the product experienced significant reductions in falls—were supported.

The more interesting issue involved the claim that the product was ​“proven to reduce falls, risk, and costs while elevating care.” SafelyYou submitted extensive evidence to support the claim and NAD agreed that the evidence showed an association between use of the system and improved outcomes. However, the evidence did not establish that the AI technology itself independently produced those results. Caregivers used information generated by the system to adjust care plans and interventions, and the studies did not isolate the AI from those human actions. Because the record did not establish causation, NAD recommended that the claim be modified or discontinued. 

The decision also demonstrates NAD’s continued focus on disclosures for AI performance claims. Although NAD found support for the claim that falls are detected with more than 99% accuracy, NAD recommended that SafelyYou disclose that the claim was based on a single study.

Advertisers can often establish that their products are associated with positive outcomes, but proving that a product independently causes those outcomes is a higher bar. Before describing AI as ​“proven” to deliver a result, advertisers should consider whether they can isolate the technology’s contribution from human actions and other variables. If they can’t, they will likely need to modify the claim so that it more accurately reflects their substantiation.