Zendesk + Koji: Trigger AI Interviews from Support Tickets and Sync Insights Back to Every Conversation

Answer first: A Zendesk ticket tells you a customer had a problem; it rarely tells you why the problem mattered, what they expected, or whether they are now a flight risk. Koji closes that gap. The integration is bidirectional: (1) a Zendesk trigger + webhook fires a Koji AI-moderated interview when a ticket is solved, a CSAT rating comes in low, or a specific tag is applied, and (2) once the interview is analyzed, your worker posts a concise insight summary back onto the ticket (or the requester's user record) as an internal note and tags — sentiment, top themes, quality score, and a transcript link. No native Zendesk app is required: Zendesk's trigger and webhook system calls Koji's headless API on the way in, and a small scheduled worker polling Koji's read endpoint handles the way back, because Koji has no outbound webhooks yet. A Zapier path exists for no-code teams. A first version is live in well under an hour.

The result is a support org that does not just resolve tickets — it systematically learns from them, at a scale no manual QA or callback program can match.

Why connect Zendesk and Koji

Support conversations are the largest, most honest stream of customer signal most companies own — and the least mined. The ticket text captures the symptom ("export is broken") but almost never the stakes ("I nearly cancelled because I present this to my CEO every Monday"). Traditional follow-up options are weak: a one-question CSAT survey with no depth, or an agent-initiated callback that does not scale past a handful a week.

Koji's AI moderator interviews the customer conversationally, asks adaptive follow-up questions, and works in voice or text across 28 languages, so you can follow up on hundreds of tickets a week without adding headcount. With platforms like Koji, every resolved ticket can become a two-minute "what really happened" interview, and every low CSAT score can trigger an immediate, structured root-cause conversation instead of a lost customer.

What flows in each direction

Zendesk → Koji (trigger interviews from ticket events)

• A Zendesk trigger fires on an event you choose (ticket solved, CSAT = bad, tag churn-risk added, priority = urgent escalated).
• The trigger's webhook calls Koji's Start Interview endpoint with the requester's email as respondent_email and the ticket ID as a custom field.
• Koji returns a personalized interview link. Koji does not send email, so Zendesk or your own mailer delivers it, or the customer meets it in an embed. Koji then moderates the interview and analyzes it automatically.

Koji → Zendesk (insights land back on the ticket)

• Koji has no outbound webhooks, so a scheduled worker polls GET /api/v1/interviews/{id} and acts when status is completed and analysis is present.
• The worker calls the Zendesk API to add an internal note to the original ticket (or the user) with the top themes, sentiment, quality score, and transcript URL, and to apply tags like insight:pricing-confusion or sentiment:frustrated.
• Agents and CX leads see the "why" attached to the very ticket that prompted it.

Ticket events worth interviewing

• Solved tickets — a lightweight "did we actually fix the underlying job?" check.
• Low CSAT / detractor ratings — an immediate, structured root-cause interview while it is fresh (see churn interview questions).
• Churn-risk or cancellation tags — route straight to a cancel-flow exit interview.
• Feature-request tickets — probe the underlying need, not just the asked-for feature.
• Repeated escalations — understand the cumulative frustration a single ticket hides.

Setup: three paths

1. Native Zendesk trigger + webhook (recommended). In Zendesk Admin, create a webhook pointing at Koji's Start Interview endpoint (with your API token), then a trigger whose conditions match the event and whose action calls that webhook with the requester email. For the return leg, run a tiny scheduled worker that polls Koji and calls the Zendesk API. There is no Koji webhook to register. Clean, and fully native on the Zendesk side.

2. No-code with Zapier. Use the Zendesk Zapier trigger ("New Ticket" / "Ticket Solved") with a Webhooks by Zapier Custom Request step calling Koji's Start Interview endpoint, then a second scheduled Zap that polls Koji's read endpoint and runs Zendesk's "Add Comment" action. There is no Koji Zapier app and no Koji webhook to trigger on. Zero code — see Connect Koji to Zapier.

3. Headless API from your own middleware. If you already run a support-automation service, call Koji's headless API directly and poll the read endpoint from your own scheduler; see webhook setup for the patterns that replace callbacks.

Design the interview so it maps to your taxonomy

Koji's structured questions make the returned data align with how your CX team already thinks. A study can mix six types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — so a single_choice "which best describes the root cause?" maps directly to your ticket-reason taxonomy, while open_ended probes capture the story. See the structured questions guide. Every quantitative answer aggregates into a real-time report you can watch fill in as interviews complete.

Example: turning CSAT detractors into a save motion

1. Zendesk trigger: CSAT rating = bad → webhook to Koji.
2. Koji sends a 3-minute interview: one single_choice root-cause question, two open_ended probes, one scale "how likely are you to keep using us?"
3. Once the analysis lands, your worker reads it back, tags the ticket sentiment:at-risk, posts the summary as an internal note, and alerts the account owner in Slack.
4. CX now has a ranked, themed view of why CSAT dips — and a triggered save conversation for every detractor.

Identity and privacy

Match on the requester's email (respondent_email) or pass the Zendesk user ID as respondent_id; the same key comes back when you read the interview, so notes land on the right ticket. For sensitive topics, run the study in anonymous mode — Koji collects no email and you skip the Zendesk write-back, keeping raw transcripts out of the helpdesk.

Plans and limits

The headless API and MCP connector are available on every plan, including a free account, and outbound webhooks are not available yet. The Zapier and CSV-import paths work everywhere too. Interviews start as low as €1 per qualified interview, and €3 per qualified voice interview.

The insight note that lands on the ticket

When an interview is analyzed and your worker writes it back, the internal note it posts on the ticket is designed to be skimmable by a busy agent. A typical note includes the overall sentiment, the two or three ranked themes, the answer to each structured question, the interview quality score, and a link to the full transcript. Tags such as insight:onboarding-gap and sentiment:at-risk are applied to the ticket and the requester, so you can build Zendesk views and reports around them — for example, a view of every "at-risk" customer surfaced by a follow-up interview this week.

Common pitfalls (and how to avoid them)

• Triggering on every ticket. Volume kills goodwill. Gate interviews to the moments that carry real signal — bad CSAT, cancellation tags, complex solved issues — and cap frequency per requester.
• Asking what the ticket already answered. Do not re-ask the reported issue. Use the interview to probe stakes, expectations, and alternatives considered, which the ticket never captures.
• Letting insights die in a note. Route the highest-severity findings (negative sentiment plus a high quality score) to Slack or a weekly CX review so they drive action, not just documentation.
• Ignoring language. Koji moderates in 28 languages; set the study language from the ticket's locale so international customers get a native-language interview.

From tickets to a continuous voice-of-customer program

Handled this way, Zendesk stops being only a cost center and becomes your largest research panel. Every week, solved tickets and CSAT detractors feed a themed, quantified stream of root-cause insight — the kind of signal product and CS leaders usually pay agencies for. Because Koji analyzes each interview automatically and aggregates structured answers into a live report, the CX team gets a standing "why customers contact us" dashboard without running a single manual study, and every new theme is traceable back to the exact ticket and transcript that produced it.

Related Resources

• Intercom + Koji
• Research Automation
• Structured Questions in AI Interviews
• Churn Interview Questions
• Customer Success Interview Guide
• Understanding Themes & Patterns