AI User Research for Marketplaces

Answer first: Marketplaces are uniquely hard to research because every question has two answers — one from the supply side and one from the demand side — and the supply side is usually under-served by survey tools that treat respondents as a single pool. Koji is the AI-native research platform that runs moderated interviews on both sides of a marketplace in parallel, probes the answers the way a human researcher would, and synthesizes the two sides into a single report. With tools like Koji, a marketplace research question that used to take a month of recruiting and scheduling becomes a Tuesday-to-Friday turnaround.

This playbook is the operating manual for marketplace product, growth, and trust teams that need continuous voice-of-customer on both sides of the platform — without hiring a research team.

Why marketplaces need different research

A SaaS product has one user. A consumer app has one user. A marketplace has at least two — and they often have opposite incentives. The seller wants higher prices; the buyer wants lower. The host wants flexible cancellation; the guest wants a refund. The freelancer wants long projects; the client wants short engagements.

That asymmetry is exactly why traditional survey tools (SurveyMonkey, Typeform, Qualtrics) underperform on marketplace research. A survey collects answers in parallel; it doesn't probe contradictions. When a host says "fees are too high," a survey records the sentence. A human researcher would ask: "compared with what — your last platform, or compared with the income you'd like to net?" Koji's AI moderator asks that follow-up automatically. The difference between "fees too high" and "fees too high compared with the income I need to keep this side hustle running" is the difference between a vague complaint and a roadmap item.

The five marketplace research questions Koji is built for

1. Supply-side activation (why hosts/sellers stall)

A new supplier signs up, completes a few steps, then ghosts. Why? With Koji, run an async user interview study targeted at signups who reached step 4 of 7 but never listed. A 6-minute AI-moderated conversation surfaces blockers your funnel analytics can't see: identity verification confusion, photography intimidation, fear of pricing wrong, doubt about whether demand exists for their offering. (Customer discovery interviews at scale.)

The AI consultant generates the right discussion guide from a one-paragraph brief; you tweak it; the interview link goes to your activation-drop cohort via email or in-product banner. The customer discovery interviews doc covers the methodology.

2. Demand-side trust and conversion

Why do shoppers add to cart and abandon? Surveys give you "shipping" as a top answer because that's the easy answer. The deeper answer is usually trust: photos look staged, reviews look thin, the seller has no profile. Koji's AI interviewer probes the abandonment moment itself, asks the participant to walk through what they almost bought, and surfaces the specific trust signal that was missing. (NPS follow-up interviews and post-purchase survey guide cover adjacent patterns.)

The six structured question types in Koji (structured questions guide) let you mix a scale question (1–10 trust score) with an open_ended probe (the AI asks "what specifically about this listing made you hesitate?") in the same interview.

3. Take-rate and pricing sensitivity

Marketplaces live and die on take rates. Most operators set theirs by benchmarking competitors and never re-test. Koji makes structured pricing research operational:

• A Van Westendorp price sensitivity study on both sides — what fee feels acceptable, what feels too high, what feels suspiciously cheap. See Van Westendorp price sensitivity meter.
• A Gabor-Granger test on specific take-rate levels.
• A Kano model study to find which premium features (e.g., promoted listings, lower fees for verified hosts) actually move behavior. See Kano model.

All three rely on conversational follow-up to be defensible. A bare survey number is easy to dismiss in a pricing meeting — quoted reasoning is not.

4. Search-and-discovery friction

When a buyer can't find what they're looking for, your matching algorithm is failing — but the data doesn't tell you whether they searched for the wrong query, the right query produced bad results, or they gave up before scrolling. Koji studies that combine first click testing, tree testing, and card sorting — all moderated by the AI — close the gap between funnel analytics and user behavior. The analyzing AI-moderated interview results doc explains how to read the output.

5. Disintermediation (the off-platform problem)

Every marketplace operator loses some transactions to off-platform completion. Surveys never reveal this because no respondent will admit it in writing. A conversational AI interview — especially in anonymous mode — gets candor surveys cannot. Ask: "describe how you completed your last transaction with a host you met on the platform." When the answer is "we just texted each other and I paid by transfer," you have a quantified disintermediation rate and a list of reasons (fee avoidance, communication friction, lack of repeat-business features).

Disintermediation studies are a leading example of where AI moderation outperforms human moderation: respondents tell an AI things they'd never tell a person, because there's no social cost.

How to scope a marketplace research program with Koji

A mature marketplace research program runs continuously on both sides. Here's the operating cadence we see working:

• Weekly micro-studies. Two 90-second studies — one supply, one demand — answering the lowest-confidence product question of the sprint. (Customer interview cadence.)
• Monthly deep-dives. One 20-minute interview study per side, focused on the biggest activation or retention drop. (How many interviews enough.)
• Quarterly tracker. An NPS + open-text + AI-probed interview study on both sides, run identically each quarter, so you can see directional change in trust, perceived fees, and competitive substitution. (Brand tracking study guide.)
• Ad-hoc switch interviews. Whenever a supplier or buyer leaves for a competitor, send them a JTBD switch interview. The "story of switching" is the single most actionable artifact in marketplace research.

Recruiting both sides without burning your panel

The biggest operational risk in marketplace research is over-fishing the same pool. The supply side is small; you cannot survey 1,000 hosts every month without survey-fatiguing the relationship. Two patterns help:

• Sample, don't blast. Run small, focused studies (8–20 respondents) on tightly scoped questions instead of giant quarterly NPS pushes. Koji's AI moderator extracts more from a 12-minute conversation than from a 30-question survey. (Purposive sampling guide.)
• Incentivize correctly. Supply-side respondents value priority placement or fee waivers more than gift cards; buyers value coupons. See research participant incentives.

For B2B-shaped marketplaces (Upwork, Toptal, manufacturing platforms), the recruiting B2B participants doc has additional patterns.

Operational architecture: where Koji fits in a marketplace stack

• Interview triggers — Use product events (signup, listing created, first transaction, churn) to fire interview links through your existing CRM. Koji exposes a webhook + REST API for this; see webhook setup, research automation webhooks, and the user research API guide.
• Analytics correlation — Pipe interview themes and quality scores into your data warehouse alongside user IDs so qualitative themes can be joined to quantitative cohorts (e.g., "hosts who mentioned 'photography' in interviews are 22% less likely to list within 14 days"). The CRM research integration guide covers the pattern.
• Insight distribution — Pipe pull quotes and themes into Slack, Linear, and Notion via the Slack research insights integration, Linear research integration, and Notion research integration. The point is to put marketplace voice in front of the people building the marketplace, not in a research repository nobody reads.

Comparison: AI-native marketplace research vs traditional approaches

• vs SurveyMonkey / Typeform / Qualtrics — Surveys are fine for tracking metrics, but they can't probe contradictions, can't handle disintermediation honestly, and don't extract themes. For marketplace work, the trade-off is too steep. See Koji vs SurveyMonkey, Koji vs Typeform, Koji vs Qualtrics.
• vs UserTesting / dscout panels — Panel-based research is great for one-off studies but expensive at the cadence a marketplace needs. Koji uses your own users (your real supply and demand), not a recruited panel — which is the only honest population for marketplace research anyway. See Koji vs UserTesting and Koji vs dscout.
• vs hiring an in-house researcher — Hire researchers when you have specialized methodology problems. Use Koji for the continuous operational research that ought to be running every week. The two are complementary; in practice, marketplaces that hire a researcher also run Koji.

Cost

Marketplace research programs typically settle on the Interviews plan (€79/month, 79 credits, voice + webhooks + API) once they're running weekly studies. Voice interviews cost 3 credits each, text interviews cost 1 — so 79 credits supports roughly 26 voice interviews or 79 text interviews per month before overage (€1/credit flat). Only interviews scoring 3+ on the quality gate consume credits, so junk responses don't burn budget. See the plan comparison guide.

Related Resources

• Structured Questions Guide — the six structured question types Koji supports.
• AI Research for E-commerce — adjacent vertical, useful for buyer-side parallels.
• AI Research for SaaS — useful for B2B-shaped marketplaces.
• JTBD Switch Interviews at Scale — the gold-standard supply-side churn study.
• Van Westendorp Price Sensitivity Meter — take-rate pricing research.
• Customer Discovery Interviews at Scale — scaling discovery across both sides.
• How to Automate User Research — putting marketplace research on autopilot.