洞察未知：人性化AI（第三部分）——视觉AI与AI智能体如何变革产品测试

SEEING THE UNSEEN Humanizing AI, part threeHow Vision AI and AI agents aretransforming product testing Dr. Nikolai ReynoldsColin Ho, Ph.DJinHo Cho Seeing the Unseen, which marks thethird edition in ourHumanizing AIseries,reveals how business, marketing, insight,and product development leaders canleverage HI+AI to see moments in productexperiences, not memories – at speed,scale, and unparalleled levels of depth. The rise of generative AI has triggeredtectonic shifts in product development. Today, manufacturers are shrinking theirproduct development cycles from years tomonths, chasing a first-mover advantage Speed in product development matters.Meanwhile, sustainable growth remainsrooted inspeed to superiority. Without aclear understanding of the total productexperience, products can fall short of At Ipsos, we champion the unique blend of HumanIntelligence (HI) and Artificial Intelligence (AI) topropel innovation and deliver impactful, human- Our Human Intelligence stems from our expertisein prompt engineering, data science, and ourunique, high quality data sets – which embedscreativity, curiosity, ethics, and rigor into ourAI solutions, powered by our Ipsos Facto Gen AIplatform. Our clients benefit from insights that #IpsosHiAi By reinventing traditionalin-home usage tests (IHUTs)with video, observationaltechniques, Vision AI and AIagents, we can surface hiddenhuman truths and identify the Limitations in legacy IHUTs This progression was notmerely technological;it reflected a growingunderstanding that The in-home usage test (IHUT), in whichconsumers evaluate products in theirhomes, has long been a cornerstone ofproduct testing. Unlike central-locationtests (CLT), where consumers testproducts in facilities in controlled andstandardized ways, IHUTs let consumersuse and evaluate products in the contextof daily life. Products are shipped toor placed with participants, used for a (i.e, consciously tracking the productperformance several times during theusage). At Ipsos, we recognized the need Since 2012, Ipsos has systematicallybuilt targeted AI capabilities for Automated analysis and codingof social media linked to survey- Predictive analytics to simulateoptimal product profiles usingsensory, product analytical and However, in recent years, the marketresearch industry has hit an inflectionpoint. Expectations have soared,compressing product development cyclesfrom years to months, leaving less time toevaluate products against all requirementsto get them right from the start. Theexplosion of direct-to-consumer brands,coupled with social media amplificationof consumer voices, means that productfailures become viral moments, damagingbrands within hours. At the same time,small and local brands can harness socialmedia to amplify their reach with ease, Large language models (LLMs)to generate innovations andimprovements in products,packaging, and concepts, and This progression was not merelytechnological; it reflected a growingunderstanding that traditional research That said, IHUTs’ reliance on post-usageself-reported data is their Achilles’heel. While testing occurs in real-worldcontext, the retrospective survey formatmisses moment-to-moment productexperiences, and those moment-to- The transformative potential of video evaluations of longer-term events (e.g.,using a product), are primarily grounded inintense momentsand thelast experiencein the event rather than the overall average •The emergence ofAI agents, whichcan simulate specialized evaluation Instead of relying on imperfect memoryabout a product’s performance, wecan ask participants to press “record”on their product experiences. A videoof the product journey, going fromunboxing to repeated use and storage,provides unfiltered observational data.Videos reveal hesitations, small delight videos at scale and in granular detailusing Vision AI, which is a type of AI thatcan see, and is leveraged to analyzevisual stimuli, like pictures and videos.We should note that Vision AI is notnew. In addition to analyzing videos ofproduct experiences with Vision AI, weleverage the same technology to identifythe products consumers use and buyin their daily lives, from their product •Increasing consumer opennessand comfort towardsself- Ultimately, retrospective survey questionsdo not capture the sequences of eventsand micro-experiences that shape finalopinions, and recollections are vulnerableto bias and memory error. On that basis, With these three advances, we’vere-imagined product testing to capturereal-world product usage, groundedin video with Gen AI based analysis, tosurface hidden truths using AI agents.The payoff? Richer and more actionable Three converging advances in technologyin 2023-2024 enabled opportunities for Video in market research is not new;the revolution is our ability to analyze •The maturation ofdiffusion AImodels coupled with LLMs, which Focus is key: Nine observational dimensions Videos are rich in information. In fact,each v