OPINION16 February 2024

One size doesn't fit all: Cultural sensitivity and AI avatars

AI Asia Pacific CX Europe Inclusion Latin America Middle East and Africa North America Opinion People Technology UK

Much research with synthetic AI avatars has focused on Western markets. Alexandra Kuzmina examines the findings of a study examining consumers’ responses to and the cultural nuances in the depiction of avatars in Indonesia. 

Earth with data points

Have you ever caught yourself wondering about the future of market research? It is something that has been on my mind a lot lately, especially considering that traditional online surveys, once the backbone of consumer research, face a bit of a participation crisis. In response, MMR’s in-house innovation team, Nova, has taken on the challenge to understand whether synthetic artificial intelligence (AI) video avatars should be the key to future-proofing our industry.

As such, an AI video avatar is a synthetic doppelganger based on real individuals, crafted through text-to-video technology using generative AI and neural video synthesis. The result? A life-like researcher introduced at the outset of a survey, replicating realistic appearances and movements, injecting a fresh approach into consumer research engagement.

In a collaboration with Heineken earlier this year, we tested the effectiveness of AI-powered avatars in English-speaking markets. We found that using these avatars yielded a 44% increase in the word count in open ended responses. Moreover, the quality of the open-ended feedback shared was superior in comparison to more traditional methods, providing more depth and enabling actionable recommendations.

Indonesia experiment
This positive outcome led to the next question: is it possible to achieve similar results in non-English speaking markets? To answer this, MMR collaborated with Symrise, a global flavours and fragrances house, to test the effectiveness of this approach in Indonesia.

A quantitative online survey targeted at engaged 636 females in Indonesia to delve into their perceptions of new cooking condiment concepts. This focus on a female-only sample was intentional, as women represented the primary consumer demographic for these condiment concepts. Notably, 57% of these participants were daily hijab wearers, reflecting the cultural diversity of the target market.

The survey itself was structured into six distinct experimental cells. Four of these cells featured a video of an AI avatar, introducing the concept ideas, while the remaining two were control and validation groups.

The AI video avatars, all speaking Bahasa Indonesia, varied in representation – a ‘local’ male, a ‘local’ female, a ‘local’ female wearing a hijab, and a Western female. Our aim was to closely align the local avatars with the target audience. However, the choice of the avatars was more of a constraint rather than a preference, dictated by the limited selection of avatars specific to Indonesia available on the third-party avatar platform used for this project.

Culture and language
This experiment found that the avatars can help connect and engage with the audience, but to do this effectively they must be culturally and contextually appropriate.

As in the Heineken pilot, we viewed the depth of open-ended response as a good indication of respondent engagement. However, in reviewing the word count for the Indonesia experiment, we discovered that Bahasa Indonesia uses fewer words than most languages, often conveying the same information more directly and succinctly than English.

As a result, we only observed marginal differences in word count across the experimental cells. Most notably, the male avatar elicited the fewest words, signalling a potential misalignment with the female-centric nature of food preparation in Indonesia. Some even voiced their preference for a female avatar instead of a male one. 

Participant feedback provided nuanced insights into avatar perception. Notably, the hijab-wearing avatar was consistently rated more favourable than the local female avatar. This study confirms that avatars closely reflecting the audience’s identity and culture lead to a more engaged and positive experience, highlighting the critical impact of cultural resonance on research engagement.

Navigating diverse markets
The Indonesian market, with its diverse cultural nuances, underscores the saying: one size doesn’t fit all, emphasising that a personalised research experience goes beyond a universal approach. As technology continues to rapidly evolve, we anticipate a broader selection of culturally and contextually appropriate avatars.

And perhaps one day soon, participants will be able to choose the avatars they engage with, rather than researchers making this choice for them, fostering an even more personalised research experience. My team and I look forward to continuing to explore the promise of AI avatars in ensuring a more inclusive, culturally resonate, participant driven research future.

Alexandra Kuzmina is innovation associate director, NOVA tech innovation team, at MMR Research Worldwide