{"site":{"name":"Koji","description":"AI-native customer research platform that helps teams conduct, analyze, and synthesize customer interviews at scale.","url":"https://www.koji.so","contentTypes":["blog","documentation"],"lastUpdated":"2026-11-21T02:05:14.164Z"},"content":[{"type":"documentation","id":"d2e0ef78-a098-4723-b2cd-06c314f06215","slug":"amplitude-research-integration","title":"Amplitude + Koji: Trigger AI Interviews from Product Analytics and Pipe Insights Back as Events","url":"https://www.koji.so/docs/amplitude-research-integration","summary":"A bidirectional integration pattern between Amplitude product analytics and Koji AI interviews. (1) Export an Amplitude behavioral cohort and trigger a Koji interview link to every user in it via Zapier, Amplitude Cohort Sync + webhook, or reverse ETL (Hightouch/Census). (2) Koji has no outbound webhooks yet, so a scheduled job polls Koji's read endpoint (or your product listens for the embed completion event) and writes themes, sentiment, quality score, and transcript URL back to Amplitude as a custom event (koji.interview_completed) plus user properties (last_interview_themes, last_interview_sentiment, last_interview_quality). Setup is 30 minutes via Zapier or under an hour via a polling job plus the Amplitude HTTP API. This unlocks Amplitude charts that segment any metric by interview theme — e.g., \"Day-30 retention for users whose last interview surfaced 'pricing confusion' is 14 points lower than cohort average\" — a join survey tools cannot enable because surveys don't extract themes natively. Quality gating (only 3+ scored interviews write back) keeps Amplitude user properties clean. Recommended for product, growth, and retention teams that want to close the quant/qual gap inside Amplitude.","content":"# Amplitude + Koji: Closing the Loop Between Quant and Qual\n\n**Answer first:** Amplitude tells you *what* happened in your product. Koji's AI-moderated interviews tell you *why*. The native integration pattern between the two is bidirectional: (1) export an Amplitude behavioral cohort and trigger a Koji interview link to every user in it, and (2) pipe each completed Koji interview's themes, sentiment, and quality score back into Amplitude as a custom event (`koji.interview_completed`) and user properties. End-to-end this takes about 30 minutes to wire up via webhooks + the Amplitude HTTP API, or under an hour with no code via Zapier. With tools like Koji, the gap between \"users dropped off at step 4\" and \"here are the seven sentences they said about it\" closes in the same dashboard.\n\nIf you've ever wanted to click on an Amplitude funnel drop-off and read the actual transcripts of users who fell out, this is that integration.\n\n## Why combine Amplitude and Koji\n\nAmplitude is excellent at quantitative behavior. It cannot tell you why a user did what they did — only what they did. Most product teams patch that gap by:\n\n- Running quarterly NPS surveys (low resolution, low candor).\n- DMing power users on LinkedIn (low scale, biased sample).\n- Booking five user interviews per quarter (too slow to keep up with sprint cadence).\n\nKoji solves the scale problem — AI-moderated interviews mean you can run dozens per week without scheduling — but the bigger unlock is **targeting**. You don't want to interview \"users.\" You want to interview \"users who completed onboarding, used the app for three sessions, then never came back\" — a behavioral cohort Amplitude already knows how to define.\n\nWith the integration, that cohort becomes an interview audience automatically.\n\n## What flows in each direction\n\n### Amplitude → Koji (trigger interviews from behavior)\n\n- A user enters an Amplitude cohort (defined by any combination of events and properties).\n- A scheduled cohort export or a real-time webhook sends their user ID and email to your forwarder.\n- The forwarder builds that user's personal interview link: your study link with their Amplitude user ID added as `?rid=` ([personalized interview links](/docs/personalized-interview-links)). If the interview runs inside your own product instead, start it through the headless API with the user ID as the respondent's `external_id` ([starting interviews via API](/docs/starting-interviews-via-api)).\n- The invite reaches the user by email. Send the link from your own email tool, or import the cohort into Koji's Participants page as an audience (map the Amplitude user ID to the ID column) and email that audience from the study's Self-recruit tab. Koji gives each person their own link and shows who received it, opened it and took part ([managing research participants](/docs/managing-research-participants)).\n\n### Koji → Amplitude (pipe insights back as events + user properties)\n\nWhen each Koji interview is complete and analysed, the read endpoint returns themes, sentiment, quality score, and a transcript URL. Koji has no outbound webhooks yet, so a scheduled job polls that endpoint and PATCHes Amplitude:\n\n- **As a custom event** — `koji.interview_completed` with properties: `theme_top_1`, `theme_top_2`, `sentiment`, `quality_score`, `transcript_url`.\n- **As user properties** — `last_interview_at`, `last_interview_sentiment`, `last_interview_themes`, `last_interview_quality`.\n\nOnce that's in place, you can build Amplitude charts that segment any metric by interview theme — e.g., \"Day-30 retention for users whose last interview surfaced 'pricing confusion' is 14 points lower than the cohort average.\" This is the kind of join survey tools cannot enable because surveys don't produce themes natively.\n\n## Step 1 — Decide the cohorts you'll interview\n\nThe most productive Amplitude → Koji pairings are behavioral cohorts where the *why* is most valuable:\n\n- **Onboarding drop-off** — Users who hit step N then went silent for 7 days.\n- **Aha-moment skippers** — Users who never reached the activation event.\n- **Power-user expansion candidates** — Users who hit the top decile of usage but stayed on the free tier.\n- **Reactivated churners** — Users who churned and came back; what changed?\n- **Feature-adoption laggards** — Users who used a feature once and never again.\n\nDefine each one in Amplitude as a saved cohort. The [feature adoption research](/docs/feature-adoption-research) and [customer retention research](/docs/customer-retention-research) docs cover the methodology side.\n\n## Step 2 — Trigger Koji interviews from Amplitude\n\nThere are three integration paths, in order of operational maturity:\n\n### Path A: Zapier or Make (no code, ~30 min)\n\n1. **Trigger:** Amplitude cohort entry (via Amplitude Cohort Sync to a destination tool, or via a scheduled cohort export to webhook).\n2. **Action:** a Formatter or Code step that builds each user's interview link by adding their Amplitude user ID to your study link as `?rid=`.\n3. **Action:** your email tool sends the personalized link. Or skip steps 2 and 3: export the cohort as a CSV, import it into Koji's Participants as an audience, and email it from the study's Self-recruit tab.\n\n### Path B: Amplitude Cohort Sync + Koji headless API (~1 hour)\n\nIf you're on Amplitude Audiences (or any cohort sync feature), pipe the cohort to a webhook destination. A small serverless function receives the cohort payload, iterates the user list, and either sends each user their `?rid=` interview link or, for interviews that run inside your own product, calls Koji's [start interview API](/docs/starting-interviews-via-api) with the Amplitude user ID as `external_id`. Koji doesn't skip a user you send twice on this path, so keep a cooldown property in Amplitude (for example `last_interview_at`) and leave recent interviewees out of the cohort.\n\n### Path C: Reverse ETL (Hightouch, Census) (production-grade)\n\nFor a high-volume program, use a reverse-ETL tool to mirror an Amplitude cohort into Koji on a schedule. This is the cleanest pattern for orgs that already run reverse ETL for other destinations.\n\nFor any path, the [user research API guide](/docs/user-research-api-guide), [API authentication](/docs/api-authentication) doc, and [managing API keys](/docs/managing-api-keys) doc cover the credential basics.\n\n## Step 3 — Pipe Koji results back into Amplitude\n\nKoji has no outbound webhooks yet, so run a small scheduled job that polls Koji's read endpoint for newly completed interviews, or let your product listen for the embed completion event (the options are covered in [webhook setup](/docs/webhook-setup)). Your job:\n\n1. **Reads the interviews that finished since the last run.**\n2. **Posts to Amplitude HTTP API v2** with two payloads:\n   - An event: `event_type = \"koji.interview_completed\"` plus event properties.\n   - An identify call: setting `last_interview_sentiment`, `last_interview_themes`, `last_interview_quality`, and `last_interview_at` as user properties.\n3. **Handles anonymous interviews** — when the interview has no `external_id`, skip the Amplitude write (there's no `user_id` to attach to).\n\nThe write completes in <100ms p95 and shows up in Amplitude within ~2 minutes (subject to Amplitude's standard ingestion latency).\n\n## What you can build in Amplitude once the data lands\n\n- **Funnel × theme segmentation** — Take an onboarding funnel and segment by `last_interview_themes`. The drop-off step where the \"pricing confusion\" theme is 3x represented is the actual problem.\n- **Retention curves by sentiment** — Are users whose last Koji interview was `negative` more likely to churn? You no longer have to guess.\n- **Cohort triggers from sentiment** — Build a workflow where any user whose `last_interview_sentiment = negative` enters a save-the-account cohort that triggers CSM outreach.\n- **A/B test annotation** — When you ship a change, drop a Koji interview link to the control + treatment cohorts and pipe the qualitative read back into Amplitude alongside the quant lift. You stop relying on instinct to explain \"why did treatment win?\"\n\n## Comparison: Koji + Amplitude vs survey-tool + Amplitude\n\nAmplitude integrates with most survey tools (SurveyMonkey, Typeform, Qualtrics). Why is the Koji integration meaningfully different?\n\n- **Themes are first-class.** Surveys deposit a numeric score and a raw text blob. Koji deposits themes — Amplitude can group on them directly. With surveys, you'd have to run an analysis pipeline to extract themes before you could segment on them. ([Understanding themes and patterns](/docs/understanding-themes-patterns).)\n- **Probing closes the \"I don't know\" answers.** When a Typeform open-text says \"it's fine,\" that's the end of the data. When a Koji interview gets \"it's fine,\" the AI follows up and frequently extracts the real reason. The Amplitude property ends up populated with substance.\n- **Quality gating.** Koji scores each conversation 1–5 and only writes back quality 3+ to keep your Amplitude properties clean. Surveys cannot filter this way — every junk response lands as a \"user property\" and contaminates your cohorts.\n\nIf you're evaluating the broader category, the [Koji vs Amplitude](https://www.koji.so/blog/koji-vs-amplitude-2026) comparison covers when each tool is the primary system — they're complementary, not competitive.\n\n## Plan requirements and cost\n\nThe headless API, API keys and the MCP connector are included on the Interviews and Enterprise plans, and the write-back to Amplitude reads results through the API, so the full loop needs one of those. Building personal interview links works on every plan, and Koji's own recruitment email comes with every paid plan: Insights includes 20 emails a month, Interviews 50 and Enterprise 1,000, with extra emails at €10 per 1,000 plus VAT. Outbound webhooks are not available yet, so the write-back polls the read endpoint or uses the embed completion event. Text interviews cost 1 credit and voice interviews 3 (voice is on Interviews and Enterprise), and conversations that don't pass Koji's quality check aren't charged. See the [plan comparison guide](/docs/plan-comparison-guide).\n\nA typical \"Amplitude-triggered\" research program of three weekly studies with 10 respondents each uses around 30 credits a week in text and around 90 a week in voice. The Interviews plan includes 79 credits a month, and you can buy credit packs when a busy month runs over.\n\n## Anonymous mode, consent, and compliance\n\nFor regulated industries (fintech, health), see [GDPR-compliant AI user research](/docs/gdpr-compliant-ai-user-research) and [anonymizing customer interview data](/docs/anonymizing-customer-interview-data). When Koji runs in anonymous mode, no email is collected and your Amplitude write should be skipped — there's no user to attach properties to. This is enforced by your forwarder, not by Koji.\n\n## Related Resources\n\n- [Structured Questions Guide](/docs/structured-questions-guide) — the six structured question types that power the events.\n- [Webhook Setup](/docs/webhook-setup): how to get interview results out of Koji today, since outbound webhooks are not available yet.\n- [Research Automation Webhooks](/docs/research-automation-webhooks): idempotency, retries, and replay patterns for your polling job.\n- [User Research API Guide](/docs/user-research-api-guide) — programmatic access to studies and interviews.\n- [HubSpot Research Integration](/docs/hubspot-research-integration) — the same pattern, mapped to HubSpot CRM.\n- [Salesforce Research Integration](/docs/salesforce-research-integration) — the same pattern, mapped to Salesforce.\n- [Zapier Research Automation](/docs/zapier-research-automation) — no-code path for any of these tools.\n- [Feature Adoption Research](/docs/feature-adoption-research) — the methodology side of the cohorts you'll target.","category":"API Reference","lastModified":"2026-10-23T04:10:08.999984+00:00","metaTitle":"Amplitude + Koji: Trigger AI Interviews from Product Analytics","metaDescription":"Trigger Koji AI interviews from Amplitude behavioral cohorts and pipe interview themes, sentiment, and quality scores back into Amplitude as events and user properties.","keywords":["amplitude integration","amplitude research integration","amplitude user research","amplitude cohort interviews","product analytics qualitative","close the loop product analytics","trigger interviews from analytics","amplitude qual integration"],"aiSummary":"A bidirectional integration pattern between Amplitude product analytics and Koji AI interviews. (1) Export an Amplitude behavioral cohort and trigger a Koji interview link to every user in it via Zapier, Amplitude Cohort Sync + webhook, or reverse ETL (Hightouch/Census). (2) Koji has no outbound webhooks yet, so a scheduled job polls Koji's read endpoint (or your product listens for the embed completion event) and writes themes, sentiment, quality score, and transcript URL back to Amplitude as a custom event (koji.interview_completed) plus user properties (last_interview_themes, last_interview_sentiment, last_interview_quality). Setup is 30 minutes via Zapier or under an hour via a polling job plus the Amplitude HTTP API. This unlocks Amplitude charts that segment any metric by interview theme — e.g., \"Day-30 retention for users whose last interview surfaced 'pricing confusion' is 14 points lower than cohort average\" — a join survey tools cannot enable because surveys don't extract themes natively. Quality gating (only 3+ scored interviews write back) keeps Amplitude user properties clean. Recommended for product, growth, and retention teams that want to close the quant/qual gap inside Amplitude.","aiPrerequisites":["A Koji account with API access (available on every plan, including free)","An Amplitude account with cohort and HTTP API access","A small webhook forwarder (Vercel/Lambda/Cloudflare) OR a Zapier seat"],"aiLearningOutcomes":["Pick the right Amplitude cohorts to trigger Koji interviews on","Wire Amplitude → Koji via Zapier, Cohort Sync, or reverse ETL","Wire Koji back to Amplitude by polling the read endpoint and posting to the Amplitude HTTP API","Poll Koji's read endpoint safely before writing to Amplitude","Build Amplitude charts that segment metrics by interview theme and sentiment","Handle anonymous-mode interviews safely"],"aiDifficulty":"intermediate","aiEstimatedTime":"20 minutes"}],"pagination":{"total":1,"returned":1,"offset":0}}