March 2026 ARVIC

Across four talks, the March 2026 ARVIC conference returned repeatedly to a single tension: AI is accelerating nearly every part of the research and design process, but the parts that matter most, interpretation, direction-setting, and trust in the data, still depend entirely on human skill. Practitioners from insurance, streaming, cybersecurity, and academia brought concrete field experience to that tension, and the result was a conference less about AI hype and more about where researchers must hold the line.

AI Compresses the Process; It Does Not Replace Judgment

A consistent thread across the UX and qualitative research talks was that established frameworks, including the double diamond and traditional discovery phases, remain valuable, but they are being squeezed. Practitioners described AI as most useful for efficiency work: coding themes, searching repositories, rapid prototyping, and running studies faster and cheaper. What it cannot do is decide which problem is worth solving or catch its own errors.

• AI hallucinations and misinterpretations make human discernment a mandatory skill, not optional.
• Abductive reasoning, drawing multiple plausible inferences from observations, is a core human capability that large language models cannot replicate.
• Qualitative research provides directionality and focus that faster, AI-assisted product development currently lacks.
• Junior practitioners face a specific risk of over-relying on AI output without the foundational skill to recognize when it is wrong.
• Researchers should move toward giving recommendations and opinions, not just reporting findings.

The Data Quality Crisis Runs Deeper Than AI

One talk delivered an urgent warning that sits underneath all AI-assisted research: the participant data going into studies is increasingly compromised. Panel fraud is not new, but AI has made it easier and cheaper for bad actors to generate convincing fake responses and fraudulent profile photos. The structural cause is compensation that is far below minimum wage, which drives genuine participants away and attracts fraud.

• No single detection technique is sufficient. Effective fraud defense requires combining digital fingerprinting, automated screening, reverse image search, and other methods.
• Low-cost sample exchanges circulate bad traffic repeatedly with little accountability. The field needs a return to building and maintaining high-quality proprietary panels.
• Over-screening carries its own cost: legitimate respondents can be incorrectly disqualified, so fraud prevention must be balanced against participant experience.
• The industry has not meaningfully improved sample quality even after high-profile legal action against a panel provider for fabricating results.

Discipline Boundaries Are Blurring and That Demands New Skills

Two talks approached from different angles the same underlying shift: the clean separations between UX research, consumer insights, and even engineering are dissolving. Digital products generate constant brand-product interactions, self-service platforms let non-researchers run studies independently, and AI rewards practitioners who can move across methods and functions. The researcher role is shifting toward enabling, operationalizing, and storytelling rather than pure execution.

• Most research methods, including tracking, segmentation, interviews, journey mapping, and ethnographies, can be applied to both UX research and consumer insights. The framing differs more than the tools.
• Practitioners who combine generalist fluency with deep specialization will have a structural advantage as lines blur.
• Merging disciplines carries real risks including diluted expertise and stakeholder confusion, so convergence should be approached deliberately rather than by default.
• Skills to develop now: mixed-methods expertise, AI and data literacy, business acumen, and stakeholder management.
• Knowledge repositories matter less as standalone platforms and more as insights pushed into wherever stakeholders already work, whether Slack, Jira, or AI-assisted search.

The Stakes Are Highest Earlier in the Pipeline

Across the conference, the strongest case for human-led research was not about validating finished products but about shaping what gets built in the first place. When AI enables faster product development, the window for strategic influence shrinks unless researchers are embedded earlier, uncovering problem spaces, mental models, and differentiation opportunities before solutions are locked in. Techniques like contextual observation, co-creation with boundary concepts, and immersion were highlighted as particularly suited to emerging technology contexts where the problem space itself is still undefined.

Market pressure is driving wild-west AI adoption across industries before standards are established. Frameworks provide a stabilizing anchor during that chaos.