← All topics

AI & technology

AI in research workflow

25 indexed claims · 13 guests · peaked 2025-Q3

AI in research workflow

Overview

The AI-in-research-workflow conversation across these podcast episodes has moved past the adoption question into a messier, more productive middle ground: not whether to use AI, but how much of yourself to hand over to it. Guests largely agree that AI earns its keep on high-volume, low-ambiguity tasks — open-end coding, transcript synthesis, literature review, stakeholder translation — while remaining genuinely uncertain about its role in the empirical core of research design and data collection. The dominant tension runs between the efficiency optimists, who see conversational analytics collapsing dashboards into dialogue and AI-enabled qualitative platforms supplanting surveys, and the intentionality advocates, who caution that speed is the wrong optimization target and that researchers who outsource the doing also risk outsourcing the learning. The editorial center of gravity lands on deliberate adoption: evaluate every workflow stage, automate the autopilot tasks, and guard the interpretive ones.

From the corpus

  • AI use in research falls along a three-level intervention spectrum: no intervention (client query → automated answer; enables retrospective ROI tracking on tracking studies), moderate intervention (current state — AI gets you almost there on specific business cases), and full engagement (seismic shifts like tariffs or the iPhone where AI has no context and humans must step in). — Dave Ritter, Ep. 16 · cl-dave-ritter-015
  • Counterintuitively, LLMs excel at querying unstructured data (PDFs, transcripts, open ends) but struggle with structured data; properly querying structured data on the fly is a tremendously harder technical challenge than reading verbatim, and remains the holy grail for the next generation of systems. — Shanon Adams, Ep. 49 · cl-50th-004
  • AI has not produced compelling use cases in pharma primary research and survey design; NLP-based open-end coding dates to the 1970s, AI-designed surveys miss clinical nuance, and the productivity gains in AI-assisted data categorization still require human supervision of the output. — Shawn McKenna, Ep. 9 · cl-shawn-mckenna-008
  • AI should not know your customer data better than you do; it is a companion and brainstorming partner, not an authority — researchers must remain the experts on their own data and use AI specifically and precisely, like Amelia Bedelia, to avoid misattributions and errors. — Louisea Hudson, Ep. 15 · cl-louisea-hudson-016
  • The right use of AI for researchers is to unblock — getting unstuck on a categorization problem, generating starter responses, accelerating secondary research — not to outsource the thinking; the wrong use is plug-and-play homework, and the responses are easy to spot. — Don DeVeaux, Ep. 18 · cl-don-deveaux-008
  • AI's biggest legitimate wins in research are not in data collection but in pre-fieldwork (literature review, prior-work synthesis) and post-fieldwork (persuasion of stakeholders) — the squeeze is on the bookends, not on the empirical center. — David Evans, Ep. 39 · cl-david-evans-006
  • Researchers should evaluate every stage of the UX/research workflow — exploration, planning, preparation, recruitment, execution, reporting — and intentionally choose where to leverage AI rather than defaulting to or rejecting it wholesale. — Tanya Pinto, Ep. 5 · cl-tanya-pinto-006
  • AI is changing insights work in two distinct categories: handling chores (volume tasks like transcript reading and open-end coding that LLMs natively do well), and enabling work that was previously impossible without a team of specialists. — Shanon Adams, Ep. 49 · cl-50th-002

Contributing guests

Episodes in AI in research workflow

See AI & technology mapped across all quarters →