The Field Dispatch

Asma Qureshi's slide put it in one line: "AI is now inside your sample, not only your workflow." Qureshi ▸ slide Her worked example: a participant types five rough words and has AI polish them; the survey platform summarises the response; a researcher clusters the themes with AI; an executive summarises the summary — her name for the result was an efficient game of telephone. Qureshi ▸ ≈12:43 Her point was not that the data is worthless, but that nobody can say where the mediation entered the chain, what it changed, or who is accountable for the interpretation.

Rebecca Klee brought the numbers. She ran roughly a hundred AI-moderated interviews a week for a year, then audited more than a thousand sessions across ten rounds, scoring session conduct, question quality, probing behaviour and research coverage. Klee ▸ slide Question quality held up. Conduct did not: script loops, internal-state leaks, unacknowledged freezes and broken endings, with poor ratings between 2% and 41% and recurring after clean rounds. After the platform added structured objectives and depth controls, full coverage of research objectives rose from 26–38% to 69–91% per round.

Her follow-up interviews — fourteen past participants, interviewed by a human — found who absorbs those failures: ten described coping behaviours, seven took responsibility for keeping the session moving. "The participant may be doing the repair work." Klee ▸ slide One participant: "I'm very careful about what I say because I don't want to accidentally set it off … down some tangent." Klee ▸ slide Coverage metrics record such sessions as complete.

Klee's scoping for where AI moderation fits: recurring or time-critical research, broader qualitative comparison across markets and languages, hard-to-schedule participants, everyday lower-stakes topics. What it cannot do: observe behaviour, notice hesitation, or investigate the gap between what someone says and what they do. The judgement moves upstream — into the goal, the context, the boundaries and the moderation guidance — with a downstream duty to review what happened. "The interview may be automated. The judgement isn't." Klee ▸ slide

Qureshi's institutional answer is a written AI boundary in three bands: assist (transcription, translation, repository retrieval, draft summaries), govern (AI-assisted coding, synthesis, AI moderators, synthetic data for exploration — human judgement documented), and do not substitute (synthetic personas for real participants, AI-generated quotes as evidence, AI-only discovery for consequential decisions). Qureshi ▸ slide "The future is bounded use, not blanket use." On her show of hands, few teams have this written down; her argument was that individual judgement does not survive staff turnover, procurement or scrutiny.

Three more talks marked the same boundary. Weiyan Chee: "AI is an accelerator … not a surrogate for human analytical judgment" — set methodological standards before automated synthesis. Chee ▸ slide Melissa Voderberg: computer vision can find patterns in ethnographic footage a researcher would never have time to watch, but "the researcher should still do the sense-making." Voderberg ▸ slide William Yanko: "Measurement tells you how many. It cannot tell you why it made sense." Yanko ▸ slide

Asma Qureshi opened with the scenario: somewhere in the organisation, a number from your research is being quoted — in a board paper, a funding deck, a case study you have never seen. Months later someone asks "Where did that number come from?" — not which slide, not which project; the evidence. Qureshi ▸ slide Her name for what follows is research archaeology: Slack threads, repository tags, half-remembered project names, and a file called FINAL_v7_REALLY_FINAL.pptx. Qureshi ▸ slide "Our evidence often travels further than our documentation."

She put figures on the stakes. Under the design and distribution obligations, ASIC has issued more than eighty stop orders, and flawed customer questionnaires were identified as a basis for several interim ones. "Research quality can become operational risk." Qureshi ▸ slide Her gloss: a poorly designed question is not always just a poor question; sometimes it is a weak link in a decision system.

Her framework is the Evidence Ladder — episodic (persuasive stories), repeatable (consistent protocols), auditable (provenance and decision logs), contestable (people affected can understand the questions and seek review) — offered as triage rather than a maturity race: a marketing concept test and a hardship-eligibility journey do not carry the same consequences. Qureshi ▸ slide "The real leadership question is: what does this evidence need to survive?" Her smallest move: add one field to the repository record — "Which decision did this inform?" — plus one signal to monitor after launch. Qureshi ▸ slide

Anna Lee Anda made the commercial version of the argument. Her case study: a workshop run jointly with the go-to-market team for Zendesk's largest APAC customer — the relationship was saved, and the approach was then scaled into training for around 2,000 GTM staff globally. Anda ▸ slide Her practice behind it: know what the organisation measures and frame findings beside it, sourcing the metrics from board decks, analytics definitions, product requirement documents, support and churn data, sales calls and product marketing's competitor reports.

Ania Mastalerz: "Research is only influential when it is trusted, remembered, and acted on." Mastalerz ▸ ≈18:50 Her conclusion about the readout: Mastalerz ▸ ≈12:21 stop optimising for the final report and treat every interaction — the hallway chat, the Slack message, the check-in — as part of a continuous narrative.

Weiyan Chee covered the operational side: "Do not depend on an individual. Single points of failure ruin research maturity" Chee ▸ slide and "Don't let context leave with people." Chee ▸ slide From her experience of both kinds of organisation: the ones that document recover faster after a reorg.

Ania Mastalerz grew up in a family of diplomats, and organised her talk around three principles from that world. Soft power: anchor to business realities, and treat confidence as buildable — "It's something that comes from doing hard things." Mastalerz ▸ ≈00:50 Coalition building: map stakeholders on a power–interest grid — champions, supporters, challengers, bystanders — and look beyond product and design for allies. Her example: Atlassian's customer support team, drowning in tickets caused by usability issues; partnering with them produced estimated support-cost savings the research team could point to. Framing narratives: "Research is a collection of facts. Narrative is the story that gives those facts meaning." Mastalerz ▸ slide

Inside the second principle sat nemawashi — Japanese, literally preparing the roots before transplanting a tree; in an office, "the meeting before the meeting." Mastalerz ▸ slide Quiet one-on-ones before a formal review, so stakeholders see the idea, critique it and shape it first. Her framing example rested on a stated mechanism — threats narrow thinking, opportunities expand it — so "we lack basic accessibility features our customers need" becomes "Inclusive design expands our Total Addressable Market (TAM) by 15%." Mastalerz ▸ slide

Anna Lee Anda's version was structural. Opening slide: "Follow the question, not the org chart." Anda ▸ slide Her framework has four levers Anda ▸ slide: openness — she listens to recorded sales calls; proximity — sit in the go-to-market weekly sync; exposure — ask direct stakeholders to forward research to people you don't know exist; initiative — bring your own questions rather than waiting for a brief. With one caveat from the talk: none of it at the expense of your primary stakeholders.

William Yanko read the brief itself as data. Nearly every research brief he has been handed in twelve years arrived with the answer already written into it — move the button, cut the steps, add the feature. Citing Clifford Geertz on common sense as a cultural system: "A research brief tells you what the organisation has stopped arguing about." Yanko ▸ slide "Why won't they adopt it?" has already decided the tool is fine and the user is deficient; "How do we simplify this?" has already decided the barrier is effort. His practice: "Answer the research brief you were given. Then hand them back the questions they should be asking." Yanko ▸ slide In academic research, he noted, discovering the question was wrong counts as a result; in industry it is treated as going off brief.

Weiyan Chee's contribution was governance design: build around what you protect, and design each process for three groups — the people who decide, the people who follow, and the tools they use Chee ▸ slide — asking of each whether it survives a reorganisation. She opened by asking the audience to vote on which scenario would cause the most chaos — layoffs, the repository vanishing, budget halved, team doubling by acquisition — then: "It's not what will cause the most chaos, but it's about how you want to survive through it." Chee ▸ ≈03:52

Bhaven Chauhan and Chanel Lee reported what this costs in practice. Their stakeholders' objections: it's too complex, it's a lot of work, the interviews are too long, it's too risky to take people on live trains, we needed the insights yesterday. They reviewed the approach, kept it, and secured their advisory committee's backing. Their own summary: "it took a lot more defence than we expected."

Bhaven Chauhan and Chanel Lee are service designers on Transport for NSW's Future Fleet program, replacing Sydney's suburban trains. To write customer experience requirements, they travelled with people with a range of permanent disabilities on live services — Central to Hornsby on a Waratah train, back on a Mariyung. Twelve participants across six sessions over two weeks; sixteen people interviewed by the end. Each session ran close to three hours.

The pushback is on their slide: "Participants will get tired" · "Participants will get bored" · "Participants will lose attention" · "There'll be nothing left to say" · "All you need is 40 minutes… 1 hour max." Chauhan & Lee ▸ slide They reviewed the plan, kept it, and secured the Accessible Transport Advisory Committee's backing. The length was the method: "We actively listened and observed. We allowed them to think about their response, and we'd sit in silence. Let time pass, told them that we could come back to a topic if they needed. This gave them the space that they needed to open up." Chauhan & Lee ▸ ≈23:57

On what that produced: some participants started reserved, keeping physical distance, answering in a few words — until the researchers admitted their own nerves. "By the time we got to Hornsby, they were sitting right by our side. Sharing everything unfiltered." Chauhan & Lee ▸ ≈24:41 Their anchor, quoted from The Conversation: lived experience "is not simply a personal story. It is knowledge developed through ongoing learning that emerges from engagement with systems, institutions and the realities of everyday life." Chauhan & Lee ▸ slide And on the compliance-first alternative: "it gives us a necessary baseline, but it doesn't tell us whether something is genuinely usable or ultimately dignified."

Chandra Harrison's study for Air New Zealand began with the client's own words: "We don't know enough about disabled travellers' experience — need data and stories to convince the business that it is important." Harrison ▸ slide Three phases: an accessible survey, ethnographic research with twelve disabled people flying between Aotearoa and Australia, and an expert validation workshop. The survey was planned for about 300 quality responses; it drew around 3,000.

The research instrument itself was accessibility-tested: platform checked with screen readers, alternate formats, plain language, pictograms, phone support for completing the survey. A New Zealand Sign Language participant recorded her diary entries as NZSL video, translated afterwards. Diary prompts were positioned after the stressful moments the survey had identified — "take a breath", then tell us. Her reasoning: "We're an accessibility consultancy doing accessibility research. The research needs to be accessible."

Her recommendation on team composition: "Inclusive ethnographic research in practice should include a disabled member in the team – ideally as the lead." Harrison ▸ slide The precedent she gave: "in New Zealand, if we're doing any research with Maori people, we actually have to have a Maori researcher on the team."

"The human stories were the most important." Kylee, a power-chair user and disability vlogger, was transferred to her aircraft seat by hoist — a procedure she had done many times and the gate staff had not, so she talked them through it. Her account of the trip: "we don't get to travel together too often because I'm not usually well enough. But this is an opportunity of a lifetime for me and Mike to hang out together and all I want to do is share a meal and a glass of wine with him on the flight." Dan, a guide dog handler, was told after a rescheduled flight that his dog could not travel: "I felt like a piece of baggage." Harrison ▸ ≈29:27

Weiyan Chee stated the same position as operations. Her priority order is fixed — participant, researcher, company — and her permanent list is short: ethics, privacy, consent, legal. "Do not waiver in your ethics." Chee ▸ slide Her example: a mystery-shopper study she took to her ethics team, who asked whether lying to a participant was consistent with how participants should be treated; she dropped it. Rebecca Klee's session-quality test belongs beside it: "A robust session needs all three of these. Research coverage, a reasonable participant experience, and evidence that we can trust." Klee ▸ ≈21:59