Austin Milne
Back to home

Research script generator

An AI-assisted workflow that turns a rough research idea into a rigorous, unbiased user interview guide. A preview of how it works below.

  1. 1. Research idea
    "Almost no trial agencies connect a second data source. We think they don't realize multi-source reporting exists."
  2. 2. Discovery conversation
    Clarifies the decision it feeds, who's being interviewed, and the honest hypothesis, before any questions get written.
  3. 3. Quality pass
    Flags that the hypothesis is leading: it assumes low awareness rather than testing for it.
    Redirects toward what agencies actually did with their first source, not what they'd say they'd do with a second.
  4. 4. Interview guide
    Structured, timed guide with objective, participant profile, must-ask questions, probes, and a bias check, ready to run.

This is a working preview of the tool's output shape, not a live model call.

Example prompt

I feed the assistant a research idea. It holds a short discovery conversation before writing a single question, and pushes back if the hypothesis is leading.

"Only about 12% of trial agencies connect a second data source, even though multi-source reporting is one of our strongest upgrade drivers. We think they just don't know it exists. I want to interview a handful of trial users to figure out why. Help me write the script."

Output screens

The assistant returns a guide I can run in a real call: the research framing, a quick bias check, and the timed question script, each as its own reviewable screen.

Research framing

Business question: Why don't more trial agencies connect a second data source?

Research question: What stops agencies from seeing enough value in a second source to connect one during trial?

Interview question: "Walk me through the last time you thought about adding another client or platform to your reports."

Quality flags

  • Leading hypothesis. "They don't know it exists" is assumed, not tested — start from behavior instead.
  • Hypothetical risk. "Would you connect a second source?" is weaker evidence than what they actually did with the first.
  • Missing context. Guide assumes single-client agencies too — probe for that before generalizing.

Interview Guide

Multi-Source Reporting Adoption — Trial Agencies

Research Objective

Understand what actually happens when a trial agency considers reporting on more than one client or platform — and where connecting a second data source stalls, if it does. Feeds a decision on whether to redesign the "add a source" flow or invest in awareness instead.

Participant Profile

  • Currently on trial or converted within the last 60 days.
  • Connected exactly one data source, and manages 2+ clients or platforms in their own work. (the group most likely to need a second source)
  • Mix of agencies that connected a second source late vs. never — the contrast is the useful part.

Interview Structure (45 min)

  • Introduction — 3 min
  • Background & current setup — 6 min
  • Current reporting workflow — 10 min
  • Recent experience: the first connection — 8 min
  • Problem exploration: a second client or platform — 10 min
  • Workarounds in use today — 5 min
  • Closing & priorities — 3 min

Background & Current Setup

Q: Tell me about your agency — how many clients or platforms do you currently report on, in any tool?

Probes: How is that split across tools today? Who else touches reporting on your team?

Research intent: establishes real reporting scope before asking about our product — separates "single-client agency" from "hasn't gotten to it yet."

Recent Experience: The First Connection

Q: Walk me through the last time you connected a data source — what were you trying to get done that day?

Probes: What did you expect to see afterward? Did anything take longer or feel less obvious than you expected?

Research intent: behavioral, specific, and recent — stronger evidence than "how do you usually set things up."

Problem Exploration: A Second Client or Platform

Q: Has reporting on another client or platform come up since then? What happened?

Probes: If it hasn't come up — why not, as far as you know? If it has — what made you connect it, or hold off?

Research intent: tests the "unaware" hypothesis directly by asking what happened, not whether they'd like the feature.

Q: How do you currently handle reporting for the clients or platforms that aren't in [product]?

Research intent: surfaces workarounds — spreadsheets, another tool, or simply not reporting on them — each implies a different fix.

Nice-to-Ask (only if time)

Q: If connecting a second source took the same time as the first, what would make you do it today versus later?

Hypothetical — flagged as weaker evidence, included only to sanity-check timing, not to drive the decision.

Closing

Q: Of everything we talked about, what would make the biggest difference to how you report today?

Quality Review

  • No question assumes the "unaware" hypothesis — awareness is inferred from what participants describe, not asked about directly.
  • Only one hypothetical question, clearly flagged as weaker evidence.
  • Workarounds section included so a "no problem" answer isn't mistaken for "no need."

Interviewer Guidance

  • If a participant contradicts the hypothesis (e.g. they knew about multi-source and chose not to use it), follow that thread — it's more valuable than steering back.
  • Let silence sit after the "walk me through" prompts; the second, unprompted detail is often the useful one.
  • If time runs short, drop Nice-to-Ask before any Must-Ask question above it.

Why I built this

I made this skill at AgencyAnalytics so designers and PMs could plan sharper interviews without waiting on a researcher to draft every guide. It pushes back on leading hypotheses the same way a good research partner would, before they turn into leading questions.

Back to home