Can AI Transform How We Understand Consumer Values? Trinetra Research Heads to MIT

We're excited to announce that we will be presenting a poster summarising our ongoing research work on Large Language Models (LLMs) in ethnographic analysis at MIT's prestigious Business Implications of Generative AI Conference (BIG.AI@MIT) in Cambridge, Massachusetts on April 2, 2026.

The poster presentation, "Can LLMs Capture Expert Uncertainty? A Comparative Analysis of Value Alignment in Ethnographic Qualitative Research,” and authored by Arina Kostina and Marios Dikaiakos (University of Cyprus) with Alejandro Porcel and Tassos Stassopoulos (Trinetra).  It tackles a question central to Trinetra's investment methodology: Can artificial intelligence help us better understand the deep human values that drive consumer behaviour?

Why This Matters for Trinetra

At Trinetra, our investment edge comes from understanding people—not just data. Through our Fast and Collaborative Ethnography (FACE) framework, we conduct in-depth, semi-structured interviews with consumers in emerging markets to uncover the values, aspirations, and lived experiences that shape spending patterns and signal long-term trends.

This research collaboration with the University of Cyprus directly tests whether open-sourced LLMs can support this crucial work. Using 12 two-hour interview sessions with residents in China, the same type of ethnographic material we analyse at Trinetra, the study evaluated whether leading AI models could identify dominant human values as accurately as our expert team of anthropologists, and investment specialists.

Models tested: Qwen (Alibaba), DeepSeek, Llama (Meta), and Mistral.

Key Findings: Promise and Limitations

The results reveal both exciting potential and important boundaries:

The Good News:

• Near-expert performance: The best-performing model (Qwen 2.5) achieved accuracy in identifying the top three values from interviews that approaches the human expert ceiling.

• Ensemble advantage: Combining multiple AI models improved performance by 8-10 percentage points, suggesting that diverse AI perspectives, like diverse human experts, yield better insights.

• Sensitivity to nuance: Leading models showed uncertainty patterns that partially mirrored expert disagreement, indicating they can detect when values are genuinely ambiguous in the material.

The Reality Check:

• Prompt sensitivity: Model performance varied by up to 27 percentage points depending on how questions were framed, highlighting the need for rigorous, consistent methodology.

• Systematic biases: All models consistently over-assigned "Security" as a value compared to human experts, raising questions about whether this reflects genuine insight or training data bias.

• Ranking struggles: While models could identify relevant values, they struggled more with accurately ranking their relative importance.

What This Means for Investment Research

For Trinetra, these findings open new pathways to scale and strengthen our ethnographic approach:

1. Acceleration, not replacement: LLMs could dramatically speed up initial analysis of interview transcripts, allowing our expert team to focus on higher-level interpretation and investment implications

2. Quality assurance: AI analysis could serve as a systematic "second opinion," helping identify values or themes that might be overlooked in manual review.

3. Scalability: Over Trinetra’s 15 years of ethnographic research, we’ve found that the optimal interval between studies is six months; intervals under four months don’t allow sufficient time for deep traditional analysis. LLM assistance could enable us to process larger volumes of qualitative data without compromising depth and  potentially shorten the analysis cycle below the four-month threshold.

4. Methodological rigour: The research confirms that combining multiple models (ensemble methods) consistently outperforms any single model—paralleling our collaborative, multi-expert approach to ethnographic insight.

Critically, the research also reaffirms what we've always known: understanding human values requires human judgment. The systematic biases, prompt sensitivity, and lower performance on value ranking underscore that LLMs are tools to augment, not replace, expert ethnographic analysis.

The Broader Context

This work positions Trinetra at the intersection of two transformative trends:

• AI reshaping business operations and research workflows

• The growing recognition that qualitative, human-centred research provides irreplaceable investment insights.

Our findings suggest AI can help us gain those perspectives more efficiently, while expert judgment remains essential for true understanding.

The BIG.AI@MIT conference will bring together leading researchers, practitioners, and executives exploring generative AI's business applications. Our participation reflects Trinetra's commitment to thoughtfully integrating cutting-edge technology with our proven human-centred methodology.

Visit Us at MIT

If you're attending BIG.AI@MIT on April 2nd at the MIT Samberg Conference Center, we invite you to visit our poster presentation and discuss how AI might transform qualitative investment research, while preserving the human insight that makes it valuable.

For those who can't attend, we'll be sharing more insights from the conference and this research in upcoming communications.

Read the full abstract: A copy of the research abstract, "Can LLMs Capture Expert Uncertainty? A Comparative Analysis of Value Alignment in Ethnographic Qualitative Research," can be downloaded below.

Trinetra_BIG_AI_MIT.pdf

Download PDF • 483KB

Trinetra Investment Management disclaimer: This material has been prepared by Trinetra Investment Management LLP (“Trinetra”) for information purposes only and it is not intended to be relied upon as a forecast or research. While the information herein is considered to be correct, no representation or warranty can be given on the ultimate accuracy or completeness of such information.

Unless otherwise specified, all views expressed are those of Trinetra. These views reflect current economic and market conditions, which are subject to change.

This material, or any views or opinions expressed herein, does not amount to an investment advice nor does it constitute a solicitation or a recommendation to buy, sell or invest in any financial product, investment structure or instrument, to enter into or unwind any transaction or to participate in any particular trading strategy.