Zammad to Crisp migration requires custom API work — no native adapter exists. Extract via Zammad API, transform tickets to conversations, stage attachments, and load via Crisp API while managing daily quotas.
Migrating from Zammad to Crisp is a fully custom API engineering project with no native migration path — there is no built-in Zammad exporter that produces Crisp-compatible files, and no Zammad adapter exists in Crisp's official import tooling. The fundamental data model mismatch is significant: Zammad organizes data as discrete, closable Tickets containing typed Articles, while Crisp structures interactions as continuous Conversations containing Messages anchored to a Contact profile. Bridging this gap requires extracting all data from the Zammad REST API (or directly from PostgreSQL for self-hosted instances), transforming the ticket-article model into Crisp's conversation-message model, and loading the result via Crisp's REST API while managing daily quotas and rate limits. Several Zammad constructs — including Organizations, ticket priority, SLA timers, email threading headers, and CC/BCC fields — have no structural equivalent in Crisp and require custom handling or external archiving.
Read this first
Pair-specific gotchas that catch teams out. Each one has cost somebody a weekend.
TL;DR — Zammad to Crisp Migration
Zammad structures data as Tickets containing Articles. Crisp structures data as Conversations containing Messages. No native adapter exists. You must extract data via Zammad's API (or PostgreSQL for self-hosted instances), convert HTML emails to plain text or use Crisp's original message payload, stage attachments for upload, and map discrete tickets to Crisp's conversation model — all while managing Crisp's daily API quotas, implementing retry/backoff logic, and designing for idempotent re-runs.
Using expand=true on the ticket list can cause timeouts on large instances
If that happens, query /api/v1/ticket_articles/by_ticket/{ticket_id} for each ticket individually.
For large Zammad instances, attachment downloads are the extraction bottleneck
Each file is a separate HTTP request. Budget 2–5 seconds per attachment and parallelize with a thread pool (4–8 workers) to keep extraction time reasonable.
Estimate your total API calls before migrating
Crisp's import tool documentation estimates the formula as roughly (n × 5) + (n × m), where n is the number of conversations and m is the average messages per conversation. A 10,000-ticket Zammad instance with an average of 8 articles each would require ~130,000 API calls minimum. Use this to project how many days your daily quota will require and request an increase proactively.
Know the importer's built-in limitations
In the current source, text and note bodies are clipped at 2,000 characters, oversized inline data-URI images are stripped from original HTML, note messages are skipped on the Free plan, and extra participants are capped by plan. If your Zammad history includes long internal notes or large HTML email bodies, fork the importer or write your own loader. (raw.githubusercontent.com)
Crisp merges contacts by email address
If a Zammad user's email already exists in Crisp (e.g., from live chat activity during the migration window), the import will update the existing profile rather than creating a duplicate. This is usually desirable, but verify that merge behavior won't overwrite data you need to preserve.
Message ordering matters
Always sort articles by created_at before sending. Crisp conversations display messages in insertion order. If you load them out of sequence, the conversation history will be unreadable.
Block outbound emails during import
Crisp's official import repo recommends contacting support before bulk imports so outgoing emails can be temporarily suppressed. Without this, Crisp may send notifications to customers for every imported message. (github.com)
Imported history visibility
Crisp says imported conversations are visible to your team in the Inbox, but users do not automatically see past imported exchanges in the chatbox widget. If you need customer-facing transcript continuity, design that separately rather than discovering it after go-live. (docs.crisp.chat)
The runbook
Work top to bottom. Tick steps as you go — your progress is saved in this browser.
01 Discovery Establish why you are moving, what "done" means, and who signs off.
Objective A written scope with agreed success criteria, a named owner per workstream, and a budget approved by finance.
Keep these open
-
Pull the real numbers out of Zammad
Export counts for tickets (open and closed separately), contacts, organisations, attachments, macros, triggers, automations, views and SLA policies. Note the oldest ticket date — history depth drives the whole timeline. Estimating from memory is the single most common cause of a blown migration window.
Data Profiler Get real record counts instead of estimating from memory -
Decide what history actually moves
Agree a cut-off with the support lead: all history, last 24 months, or open tickets plus a read-only archive. Every extra year of closed tickets adds API time and cost without adding much agent value. Get this in writing — it is the decision people relitigate mid-cutover.
A "move everything" default is what turns a two-week migration into a two-month one.
COI & ROI Calculator Build the 36-month business case you will need for sign-off -
Confirm Crisp can hold your support model
Walk your current workflow through Crisp: multi-brand, business hours, SLA targets, CSAT, side conversations, public vs internal notes, and any channel you depend on (voice, chat, WhatsApp, social). List anything with no native equivalent — those are project risks, not configuration details.
-
Build the business case
Model licence delta, migration effort, agent retraining, and the cost of staying put (Cost of Inaction). Executives approve a number, not a plan, and you will be asked for it again at the go/no-go.
Helpdesk Migration Planner Turn ticket volume into a dated Zammad → Crisp timeline -
Name owners and set the go/no-go date
One named owner each for data, configuration, integrations, and agent enablement, plus a decision-maker who can call a rollback. Put the go/no-go meeting in calendars now, 48 hours before the freeze.
Zammad → Crisp specifics
- Moving away from self-hosting
- Self-hosted Zammad requires infrastructure management, security patching, and Elasticsearch maintenance. Teams that no longer want to manage a Ruby on Rails stack and PostgreSQL instances move to Crisp's fully managed SaaS.
- Chat-first support model
- Zammad is email-centric by design. Teams shifting to live chat, WhatsApp, Instagram DMs, and Messenger as primary channels find Crisp's unified inbox purpose-built for that workflow.
- Simpler tooling
- Crisp bundles live chat, CRM, knowledge base, chatbot builder, and campaigns in one platform. Teams consolidating away from Zammad + separate chat + separate CRM find that appealing.
- Per-agent pricing vs. flat-rate
- Zammad's hosted plans charge per agent; Crisp offers flat-rate workspace pricing. At scale (15+ agents), the cost structure differs materially. Check current pricing on both vendor sites before modeling TCO — published figures change frequently.
Don't move on until
- Record counts confirmed for tickets, contacts, organisations and macros
- Success criteria signed off by the support lead
- Freeze window provisionally booked with the business
02 Data Audit Find out what is actually in the data before you try to move it.
Objective A profiled, cleaned export with every quality defect either fixed at source or explicitly accepted.
Keep these open
-
Take a full Zammad export and profile it
Export to CSV or JSON and profile every file: row counts, null rates per column, distinct values, and type consistency. Compare row counts against the API totals from Discovery — a gap here means your export is silently truncated, usually by pagination.
Data Profiler Profile the Zammad export for nulls, outliers and type drift -
Validate file structure before anyone writes a transform
Check delimiters, quoting, encoding (expect UTF-8, watch for BOMs and Latin-1), duplicate headers, and embedded newlines in ticket bodies. Ticket descriptions with raw newlines and commas break naive CSV parsers and silently shift columns.
A single unescaped quote in one ticket body can shift every subsequent column without any error.
CSV Validator Catch broken headers and ragged rows in the raw export -
Inventory PII and set retention
Scan for emails, phone numbers, payment card fragments, national IDs and anything else regulated in ticket bodies and custom fields — support tickets are where customers paste things they should not. Decide what gets migrated, masked, or dropped, and record the legal basis.
Ticket bodies and attachments routinely contain card and ID data that never appears in a structured field.
PII & Compliance Scanner Find regulated fields before they land in a new system -
Quantify duplicates, orphans and dead references
Count duplicate contacts (same email, different casing), tickets whose requester no longer exists, organisations with no members, and attachments whose parent ticket is gone. Fix these in Zammad where you can — migrating them just moves the mess.
Data Cleaner Strip empty rows, stray whitespace and dead columns -
Clean and normalise the export
Trim whitespace, drop empty rows and columns, normalise casing on emails and tags, and standardise every timestamp to UTC ISO 8601. Timezone drift is invisible at load time and shows up weeks later as SLA reports nobody can reconcile.
-
Produce a masked copy for sandbox work
Generate a realistic but fake version of the export for testing and for any vendor who needs sample data. Loading real customer PII into a sandbox is a breach in most jurisdictions, and sandboxes are rarely covered by your DPA.
PII Masker Generate a safe copy for sandbox and vendor testing
Don't move on until
- Export parses cleanly with no ragged rows or encoding errors
- PII inventory complete and retention decisions recorded
- Duplicate and orphan records quantified and triaged
03 Field Mapping Turn two schemas into one signed-off mapping spec.
Objective A reviewed field-level mapping covering every object, with an explicit decision for every field that has no target.
Keep these open
-
Generate the first-pass Zammad → Crisp field map
Start from an automated match on both schemas, then review every row by hand. Automated matching gets the obvious 70% right and is confidently wrong on the rest — especially anything named "type", "status" or "custom_field_1".
Schema Mapper Opens pre-loaded with the Zammad → Crisp field pair -
Map status, priority and channel values, not just field names
Enumerate every value in each picklist on both sides and map them explicitly. Value-level mismatches are the defect class that survives all the way to production because the field itself mapped fine — a ticket that should be "Pending" arriving as "Open" reopens SLA clocks.
Statuses with no target equivalent (on-hold, pending-customer) need a policy decision, not a best guess.
-
Decide how custom fields land
Create the target custom fields first, matching type exactly (a dropdown mapped to free text can never be mapped back). Where Crisp has no equivalent, decide between a new custom field, a tag, or a note appended to the ticket body — and record which.
-
Resolve identity and threading
Decide how source IDs are preserved — most platforms will not let you set the primary key, so keep the original ID in a custom field. Without it, reconciliation becomes fuzzy matching and every future support question about an old ticket is unanswerable.
Losing the original ticket ID makes reconciliation and rollback effectively impossible.
-
Plan attachments, inline images and threading order
Confirm size limits, allowed MIME types, and whether inline images survive as attachments or need rehosting. Decide the comment ordering and author attribution rules: comments loaded out of order, or all attributed to the API user, destroy the conversation history agents rely on.
-
Freeze and sign off the mapping spec
Version the spec, walk the support lead through it row by row, and get explicit sign-off. Any change after this point goes through change control — mid-flight mapping edits are how partial loads happen.
Don't move on until
- Every source field is mapped, deliberately dropped, or parked in a custom field
- Status, priority and channel value maps agreed with the support lead
- Mapping spec version-controlled and signed off
04 Test Migration Prove the pipeline on a small, representative slice.
Objective A pilot load into a Crisp sandbox that reconciles cleanly and has been reviewed by real agents.
Keep these open
-
Stand up a Crisp sandbox that matches production config
Create the custom fields, groups, brands, business hours and SLA policies first. A pilot into a default sandbox tests nothing, because the failures you care about are all configuration mismatches.
-
Pick a deliberately nasty pilot sample
Take 500-1000 records chosen for difficulty, not convenience: the longest ticket threads, tickets with the most attachments, non-Latin character sets, merged and split tickets, deleted requesters, and every status value. A clean random sample proves only that easy records are easy.
-
Run the load with masked data and instrument everything
Log every API request and response with its source record ID. When 40 records fail out of 10,000 you need to know exactly which ones and why, without re-running the whole batch.
PII Masker Never load real customer PII into a sandbox -
Measure real throughput against the rate limit
Record achieved records-per-hour under Crisp's actual rate limits, including retries and backoff. Extrapolate to the full volume: if the maths says the full load exceeds your freeze window, you fix that now, not on cutover night.
Published rate limits are ceilings, not throughput. Assume real-world rates are meaningfully lower once retries and backoff are counted.
-
Reconcile the pilot and triage every failure
Diff source against target on record counts and field-level values. Every discrepancy gets a root cause and a fix — "probably fine" at pilot scale becomes thousands of broken records at full scale.
Migration Validation Tool Diff the pilot batch against source before scaling up -
Put real agents in front of the pilot data
Have two or three agents work sample tickets end to end in the sandbox. They find the things reconciliation cannot see: unreadable threading, missing context, macros that no longer make sense. Fix the mapping, then re-run.
Don't move on until
- Pilot batch reconciles to 100% on record counts
- Agents have reviewed sample tickets and confirmed they are workable
- Measured throughput extrapolates to a viable full-load window
05 Cutover Execute the switch inside a controlled, reversible window.
Objective All in-scope data live in Crisp, agents working in the new system, and a rollback path that stayed available throughout.
Keep these open
-
Pre-load history before the freeze
Load closed tickets and contacts days or weeks ahead while Zammad stays live. Only open tickets and the final delta need to move inside the freeze — this is the single biggest lever on window length.
Helpdesk Migration Planner Size the freeze window from Crisp's real API limits -
Publish the runbook with times, owners and abort criteria
A timed sequence: freeze start, final export, delta load, channel switch, smoke test, go/no-go, agent switch. Name who does each step and the explicit condition that triggers a rollback. Decide the abort criteria before the night, when nobody wants to be the one to call it.
-
Freeze Zammad and take the final delta
Stop new ticket creation, let agents finish in-flight replies, then export everything changed since the pre-load. Announce the freeze to the whole business, not just support — someone always tries to raise a ticket during it.
Tickets created during an unenforced freeze land in the old system and are the most common source of permanently lost data.
-
Load the delta and open tickets
Run the delta load, then reconcile counts before touching any channel. Do not repoint email until the delta has verified — an inbound ticket arriving mid-load is far harder to untangle than a few extra minutes of freeze.
Migration Validation Tool Confirm the final delta landed before you reopen -
Repoint channels and verify with live traffic
Switch email forwarding and MX or connector settings, update chat widgets and web forms, and re-authorise integrations. Then send real test tickets through every channel and confirm each lands, routes and triggers the right automation.
Email forwarding changes can take up to a full DNS TTL to propagate — check the TTL days in advance and lower it if needed.
Cron Expression Builder Schedule the delta syncs that run through the freeze -
Run the go/no-go and switch the agents
Walk the exit criteria with the decision-maker, call it explicitly, then move agents over with a named person on hand for the first few hours. Keep Zammad read-only rather than cancelled — cancelling the old contract on day one removes your only fallback.
Don't move on until
- Full historical load complete and counts matched
- Inbound channels repointed and verified with live test tickets
- Rollback decision point passed explicitly, not by default
06 Validation Prove the migration is complete, then close it out.
Objective Documented evidence that data, workflow and reporting all survived, and a signed acceptance.
Keep these open
-
Run the full reconciliation
Compare source and target on every object: total counts, counts by status, counts by group, attachment counts, and field-level spot checks on a random sample. Produce one report you can hand to an auditor.
Migration Validation Tool Reconcile Zammad and Crisp record-for-record -
Verify field completeness, not just record counts
Re-profile the loaded data and compare null rates per field against the source profile. Matching record counts with a field that silently arrived empty is the failure mode counts alone will never catch.
Data Profiler Prove field completeness held up through the load -
Rebuild reporting and compare against baselines
Recreate your core dashboards — volume, first response time, resolution time, CSAT — and compare to pre-migration figures for the same period. Explain every variance; a changed SLA calculation is a real finding, not a rounding error.
SLA and first-response metrics are usually recalculated from the loaded timestamps, so they will differ if any timestamp mapping was approximate.
-
Test the workflow layer end to end
Fire every trigger, automation, SLA escalation, macro and notification with a live ticket. Workflow does not migrate — it gets rebuilt — so it is untested until someone has actually watched it run.
-
Confirm compliance and produce the audit trail
Re-scan the loaded data for regulated fields, confirm retention and deletion policies are configured in Crisp, and file the evidence with your PII decisions from the audit phase.
PII & Compliance Scanner Produce the compliance evidence your auditor will ask for -
Sign off, then decommission on a schedule
Get written acceptance against the Discovery success criteria. Keep Zammad read-only for an agreed period (30-90 days is typical), take a final archive export, and only then cancel. Diarise the decommission date so it does not quietly renew.
-
Contact linkage
Verify conversations are linked to the correct customer email. Broken email mapping means orphaned conversations agents can't find by customer search.
Zammad → Crisp specifics
- Count match
- Total conversations in Crisp should equal total tickets extracted from Zammad (minus intentionally excluded ones like spam or test tickets).
- Message count per conversation
- Spot-check 20–30 conversations. Article count in Zammad should match message count in Crisp for each.
- Attachment integrity
- Download a sample of migrated attachments from Crisp and compare file sizes to Zammad originals.
- State accuracy
- Filter Crisp conversations by resolved and verify they correspond to closed tickets in Zammad.
- Timestamp ordering
- Open 10 migrated conversations and verify messages appear in correct chronological order.
Don't move on until
- Full reconciliation report attached to the project record
- Reporting baselines match pre-migration figures within agreed tolerance
- Formal acceptance signed and archive retention scheduled
Field mapping reference
The field-by-field mapping for each object. Use this as the starting point for your mapping spec.
Transform Zammad Tickets Crisp Conversations
| Zammad field | Crisp field | Notes |
|---|---|---|
| email (external) | text | Set from = user if customer, operator if agent |
| note (internal) | note | Requires Essentials plan or above |
| phone | text | Prefix with [Phone note] to preserve context |
| web / chat | text | Direct mapping |
| Article with attachments | file or text + separate file message | See attachment handling below |
Phase Duration
| Zammad field | Crisp field | Notes |
|---|---|---|
| API token setup + test extraction | 1 day | Verify permissions, test pagination |
| Full extraction to staging | 1–3 days | Depends on ticket volume and attachment count |
| Transform + mapping script development | 2–4 days | Including HTML sanitization, encoding, edge case handling |
| Load into Crisp (staging workspace) | 1–2 days | Test with a subset first; validate checkpoint behavior |
| Validation + reconciliation | 1 day | Automated + manual spot checks; review all failed records |
| Production load + delta sync | 1–2 days | Run during low-traffic hours |
| Total | 7–13 days | Engineering time, not calendar days |
Risk matrix
Per-object risk for this pair. Plan extra validation around anything marked high.
| Object | Risk | Notes |
|---|---|---|
| Tickets / Conversations | medium | Zammad tickets map conceptually to Crisp conversations, but state mappings are approximate — Zammad's five states (New, Open, Pending Reminder, Pending Close, Closed) must be reduced to Crisp's three (Unresolved, Pending, Resolved), which can cause status ambiguity on migrated records. |
| Articles / Messages | high | Zammad's typed Articles (email, phone, internal note, chat, social) must be individually mapped to Crisp's limited message types, and HTML article bodies must be converted to plain text or wrapped in Crisp's original payload, creating a high risk of content formatting loss. |
| Attachments | medium | Attachments are extractable via the Zammad API but require one HTTP request per file, and must be re-uploaded to Crisp as file messages or message attachments, introducing risk of incomplete transfers if the pipeline is not designed with retry logic. |
| Contacts / Users | low | Zammad User records map reasonably well to Crisp Contact/People profiles, and Crisp supports CSV import for contacts, making this the lowest-risk entity class in the migration. |
| Organizations | high | Zammad's standalone Organization objects with custom attributes and user memberships have no first-class equivalent in Crisp, and must be flattened into contact-level custom data fields, resulting in permanent structural data loss for organization-centric workflows. |
| Internal Notes | medium | Zammad internal articles (flagged internal: true) can be mapped to Crisp note-type messages, but this capability is not available on Crisp's Free plan, meaning internal note history may be unloadable depending on the target workspace subscription tier. |
| Tags / Segments | medium | Zammad's freeform tags on tickets map to Crisp's segments on conversations or contacts, but the scoping and behavioral differences between the two systems mean tag-driven workflows and routing rules must be manually re-validated after migration. |
| Custom Fields | medium | Zammad supports custom object attributes on tickets, users, organizations, and groups, while Crisp supports custom data keys only on contacts and conversation metadata, requiring field-by-field mapping decisions and potential data loss for ticket-level or group-level custom attributes. |
| SLA and Escalation Data | high | Zammad's native SLA timers, escalation thresholds, and breach event records have no structural equivalent in Crisp's data model and cannot be represented post-migration, requiring external archival before cutover to avoid permanent loss of compliance-relevant records. |
| CC / BCC Fields | medium | Crisp's conversation model does not natively support CC and BCC fields as Zammad's email-centric article model does, so this contextual data must be explicitly preserved in conversation metadata or custom fields during transformation or it will be silently discarded. |
The hard parts
What makes this specific migration difficult, beyond the mechanics.
No Native Migration Path
There is no built-in Zammad export format compatible with Crisp, and no Zammad adapter in Crisp's official crisp-import-conversations tool, requiring a fully custom API-to-API pipeline to be built from scratch.
Ticket-to-Conversation Model Transformation
Zammad's discrete, closable Ticket containing typed Articles must be structurally transformed into Crisp's continuous Conversation containing Messages, requiring a non-trivial mapping layer that preserves message directionality and type.
Organization Structure Flattening
Zammad's standalone Organization objects with their own attributes, user memberships, and routing relationships have no first-class equivalent in Crisp, and must be flattened into contact-level custom data fields, losing hierarchical structure.
Attachment Extraction Bottleneck
Each attachment binary in Zammad requires a separate HTTP request per file, making attachment extraction the primary throughput bottleneck and requiring parallelized download workers to keep migration runtime manageable.
Crisp API Quota Management
Crisp enforces daily API quotas on conversation and message creation, requiring the migration pipeline to implement rate limiting, retry/backoff logic, and idempotent re-run support to avoid data duplication or quota exhaustion.
Irretrievable SLA and Email Threading Data
Zammad's SLA timers, escalation breach events, and email Reply-To/References header chains have no structural equivalent in Crisp's data model, meaning this data must be archived externally before cutover or it is permanently lost.
Tools used in this playbook
All free, all run entirely in your browser — nothing is uploaded.
FAQ
Can I export Zammad tickets directly into Crisp?
No. There is no native import path between Zammad and Crisp. Crisp's official import tool supports adapters for Zendesk, Gorgias, Help Scout, Tidio, GrooveHQ, and WHMCS — but not Zammad. You must extract data via the Zammad REST API, transform it to match Crisp's conversation-message schema, and load it via the Crisp REST API or a custom adapter.
What data is lost when migrating from Zammad to Crisp?
SLA data and escalation history, ticket priority (no native Crisp field), organization hierarchy (flattened to contact-level company fields), agent assignment history, ticket linking and merging relationships, Core Workflow trigger logs, and checklist/audit log details. Knowledge base articles must be migrated separately.
How long does a Zammad to Crisp migration take?
For a typical instance with 5,000–20,000 tickets, expect 7–13 days of engineering time covering extraction, transformation script development, loading, validation, and cutover. Larger instances with heavy attachments take proportionally longer due to per-file API calls and quota constraints.
What are the Crisp API rate limits for data imports?
Crisp plugin tokens use a daily quota system rather than per-minute rate limits, allowing burst requests until the daily allocation is exhausted. For large imports, request a quota increase through the Crisp Marketplace before starting. Estimate total API calls as roughly (n × 5) + (n × m), where n is conversations and m is average messages per conversation.
How do Zammad ticket states map to Crisp conversation states?
Crisp supports three states: unresolved, pending, and resolved. A common mapping is new/open → unresolved, pending reminder/pending close → pending, and closed → resolved. Merged tickets typically need a reference note since Crisp has no merge concept.