How to Run 100 Customer Interviews in a Week (2026)

Qualitati Research Team · 2026-05-13 · 13 min read

Why this matters in 2026

"Talk to more customers" has been the most repeated piece of product advice for a decade. The reality has always been the opposite: most product teams talk to far fewer customers than they should because each interview costs ~3 researcher-hours all-in, and a single senior researcher caps out around 8-12 quality interviews per week. AI moderation has changed the math. In 2026 it is genuinely possible for a small team to run 100 customer interviews in five working days — without burning out the research function and without sacrificing methodological rigor — if the plan is right. This post is the playbook.

What you'll get: a five-day execution plan, the recruitment math (the rate-limiting step), the four mistakes that kill high-velocity programs, and a checklist of the protocol decisions that need to be made before day one.

The 100-in-a-week target — is it real?

Yes, but with caveats. A small product team with a defined protocol, a willing participant pool, and an AI moderator can realistically complete 100 30-minute interviews in five working days. The hard limit in practice is rarely the moderator — it is recruitment. Without a pre-warmed participant pool, getting 100 qualified participants into completed interviews in a week requires either (a) an established panel relationship, (b) a customer base you can email, or (c) a paid panel service with same-day fulfilment. We'll cover all three.

The five-day execution plan

Day −7 to Day −1: protocol + recruitment ramp

The week before execution is when the program is won or lost. Three workstreams run in parallel:

Protocol design (8-12 hours, senior researcher). Lock the interview outline: 6-10 questions, each with a clear purpose mapped to a research question. Include a 2-question screener at the top. Define what "complete" means (e.g., "answered all 6 core questions, audio ≥ 15 minutes"). Define the post-hoc quality criteria.
Recruitment ramp (variable, depends on channel). If using your customer base: send a recruitment email 5-7 days out with a screener link, schedule slots open immediately. Aim for 4× your target signup-to-completion ratio (i.e., 400 signups for 100 interviews). If using a paid panel: brief them on screener criteria 5 days out, confirm fulfilment capacity for the week.
Tooling setup (4-6 hours). Configure the AI moderator with your protocol, decide on text vs voice interviews (voice produces richer data but takes ~50% longer per participant; text is faster but loses tone), test the participant flow end-to-end with three internal pilots, and confirm consent + GDPR compliance language is in place.

Days 1-2: pilot wave + tune

Do not launch the full 100 immediately. Run 15-20 interviews on day one, review them on day two morning, tune the protocol if needed. This is the single most important step the 100-in-a-week playbook gets wrong — most teams launch everything Monday and discover by Wednesday that a question is unclear or the screener is letting through wrong-profile participants. By then you've wasted 50 interviews.

Things to look for in the pilot wave review:

Are participants understanding the questions as you intended them?
Is the AI moderator probing in useful ways or getting stuck?
Are responses producing the kind of data you'll actually be able to code?
Is the screener letting through the right participants?

Days 3-5: full execution

Launch the remaining 80-85 interviews. Monitor daily. Most AI moderators (Outset.ai, Strella, Listen Labs, Qualitati) can run multiple interviews simultaneously, so the throughput limit is recruitment fulfilment, not moderator capacity. Track completion rates daily and bump recruitment volume if you're trending below target.

Day 5 evening to Day +2: analysis

Run AI thematic analysis across all 100 transcripts (~5-15 minutes depending on platform). For Qualitati that's the ThemeLens map-reduce pipeline; other platforms have similar capabilities. Then spend Day +1 and Day +2 doing human review of the AI-surfaced themes, pulling representative quotes, and writing the findings deck. Total elapsed time from kick-off to recommendations: 7 working days.

The recruitment math (this is the bottleneck)

To complete 100 interviews you need to plan for a funnel like this:

Invitations sent: 800-1,200 (depending on relationship warmth — warmer base, fewer needed)
Screener completions: 250-400 (30-40% of invitations)
Qualified by screener: 150-250 (60-70% pass rate is healthy; if higher, your screener is too loose)
Schedule/start interview: 130-180 (some qualified participants drop)
Complete interview: 100 (15-25% mid-interview drop is normal)

The implication: if you have 5,000 customers in a database, you can probably hit 100 interviews. If you have 500, you cannot — you'll need a paid panel partner or a longer timeline.

Voice vs text interviews at scale

Voice produces richer data — prosody, hesitation, emotion — and tends to elicit longer, more reflective responses. It also takes participants ~50% longer to complete (because they have to actually speak) and produces audio that needs to be transcribed (modern AI moderators do this automatically). Text is faster, lower-friction for participants (especially in noisy environments or open-plan offices), and easier to analyse mechanically.

For high-velocity programs we recommend:

Use voice when: the research question is sensitive, narrative, or relies on tone/emotion. Examples: "How did you feel when [event] happened?", "Walk me through your decision process when [scenario]".
Use text when: the research question is task-oriented, evaluative, or asks for specifics. Examples: "Which of these three concepts would you pay for?", "List the steps you took to set up [feature]".
Mix both: screener can be text, deeper-dive questions voice. This is how Qualitati's mixed mode works by default.

The four mistakes that kill high-velocity programs

Mistake 1: Launching the full wave on day one

Covered above. Always pilot first. Always.

Mistake 2: Treating "interview completed" as "data usable"

Not every completed interview is good data. A useful step: define a quality rubric upfront (e.g., "all 6 core questions answered with at least 2 sentences each, no obvious bot-like responses, no contradictory profile signals"). Run every completed interview through the rubric — automated where possible, sample-based otherwise. Reject 5-10% as a healthy quality discipline. Modern platforms include automatic quality scoring (Qualitati uses a GABRIEL-based rating framework with use-case-specific attributes — see our for-teams page for details).

Mistake 3: Letting the AI moderator decide when an interview ends

Some platforms let the AI decide to wrap up when it judges that the protocol is complete. This can produce 6-minute "interviews" that technically covered all questions but missed depth. Set a minimum duration target (e.g., 12 minutes for a 30-minute protocol) and have the AI probe for depth if responses are too thin. Most platforms have a "probing depth" setting — use "moderate" or "deep" for high-velocity programs to avoid runaway shallow data.

Mistake 4: Skipping human review of the AI-surfaced themes

AI thematic analysis (ThemeLens, Outset's analysis pipeline, etc.) is fast and produces structured output. It also gets things wrong in interesting ways: it can overweight surface vocabulary, miss tacit themes that span multiple participants, and produce themes that are technically correct but strategically useless. A senior researcher should spend half a day reviewing the AI output, validating the themes against representative quotes, and rejecting or merging themes that don't hold up. This is non-negotiable for high-stakes findings.

The protocol checklist (do these before day one)

Define the 6-10 core questions, each mapped to a research question.
Write a 2-question screener that filters for the participant profile you need.
Decide voice vs text (or mixed).
Set the minimum duration target.
Set the probing depth (light / moderate / deep).
Choose single-model or dual-model AI moderator (dual-model with a supervisor catches more off-topic drift; recommended for high-velocity programs).
Define the post-interview quality rubric.
Confirm consent / GDPR language is in the participant flow.
Run 3 internal pilots end-to-end before any external recruitment.

What this actually costs

Order of magnitude for 100 30-minute voice interviews + full thematic analysis on a transparent-pricing platform (Qualitati pricing): 100 × 30 min × 5 credits/min = 15,000 credits for moderation, plus ~5,000 credits for analysis of the resulting transcripts. Total: ~20,000 credits, which depending on your plan is roughly $400-700 in platform cost.

Add recruitment: if using your own customer base, the cost is in the email sends and any incentive (

0-25 per completed interview is typical). For 100 completed interviews at $20 incentive each: $2,000. If using a paid panel partner, expect $25-75 per recruited participant depending on profile, so $2,500-7,500 for 100 interviews.

Total program cost: $2,500-8,000 to run 100 customer interviews end-to-end in a week. Compared to the traditional $20,000-45,000 for the same volume of human-moderated interviews over 6-10 weeks, the case for AI moderation in high-velocity programs is straightforward.

When NOT to do this

High-velocity AI-moderated programs are the wrong tool for: ethnographic / contextual research, sensitive disclosure topics (trauma, mental health, financial stress), highly ambiguous early-stage discovery (when you don't yet know what questions to ask), and vulnerable populations. See our AI Moderator vs Human Researcher post for a four-criteria decision framework.

Summary

Running 100 customer interviews in a week is a real capability in 2026, not a marketing claim. It requires: a tight protocol, a warmed recruitment funnel sized 8-10× the target, a 2-day pilot wave before full launch, an AI moderator configured for moderate-or-deep probing with a quality rubric, and a senior researcher's half-day on AI-output review at the end. Teams that ignore any one of these usually end up with 100 completed interviews and no actionable findings. Teams that do all five end up with a research practice that runs weekly instead of quarterly — and that's the actual point.