Phase 0

Decide & Plan

Decisions that lock scope before any work begins.

Define which Zendesk products are in scope. Document Support, Guide (help center), Talk, Chat/Messaging, Explore, and Sunshine custom objects per product. Each has its own data model and migration path.
Risk if skipped: Scope ambiguity around Talk recordings or Sunshine custom objects surfaces mid-migration and stalls cutover.
Confirm Zendesk plan tier and product entitlements. Sandbox availability starts at Professional and above; Custom Objects are restricted to Support Enterprise and Suite Team and above; Light Agent seats require Suite Professional or higher.
Risk if skipped: Plan-tier limits discovered after contract signing force last-minute upgrades or cut features from scope.
Decide the historical-data window per object. State explicitly: how many years of closed tickets, KB article revisions, satisfaction ratings, and Talk recordings move. Closed tickets older than the window stay in archived source exports.
Risk if skipped: Pulling all-time history balloons storage usage and migration runtime; pulling too little erodes agent context.
Choose data center region for the new Zendesk pod. Zendesk hosts Service Data in AWS regions including US, EU, APJ, and others gated by the Data Center Location add-on. Confirm the region matches data-residency obligations before provisioning.
Risk if skipped: Provisioning in the wrong region requires Zendesk Customer Support to spin up a fresh account and reload everything.
Capture GDPR, HIPAA, SOC2, and HDS compliance requirements. If PHI is in scope, sign Zendesk's Business Associate Agreement before import. Capture data-retention policy changes that take effect at cutover.
Risk if skipped: PHI loaded into an account without an active BAA creates a notifiable breach.
Assign a single accountable migration owner (DRI). One named person owns scope, schedule, and sign-off. Multiple stakeholders may contribute, but only one person decides.
Document explicit out-of-scope items. List what is NOT migrating: e.g. spam tickets, archived users, deleted KB articles, Talk voicemails older than X months. Get sponsor sign-off on the exclusion list.
Risk if skipped: Unwritten exclusions become user complaints after cutover.
Set the cutover window and source-system freeze date. Pick a low-volume weekend. Freeze must include source-system writes, automations, and integration triggers.
Define written success criteria. Concrete pass/fail thresholds: ticket count delta < 0.1%, KB article count exact match, attachment availability > 99.5%, zero failed user imports, custom field coverage 100%.
Risk if skipped: Without numeric criteria the migration is never formally finished and the team keeps firefighting.
Identify legal and procurement signoff steps. Surface contract review, DPA execution, BAA execution, and any required security questionnaires before the cutover date.

Phase 1

Pre-Migration Prep

Everything that must be true before the first byte leaves the source.

1a. Source-system audit and cleansing

Export current source record counts as a baseline. Capture counts for tickets (open, pending, solved, closed, archived), users (end users, agents, admins), organizations, groups, brands, KB categories/sections/articles, and macros/triggers/automations.
Risk if skipped: Without a baseline you cannot prove the migration was complete during Phase 6 reconciliation.
Run dedup analysis on source users keyed on lowercase email. Resolve merges in the source before export. Zendesk upsert on email collapses duplicates non-deterministically if both sides have the same address.
Risk if skipped: Duplicate users in source cause one agent's tickets to land under the wrong end-user profile.
Standardize email casing and trim whitespace on all user records. Lowercase the email column. Trim leading and trailing whitespace. Zendesk treats email matching case-insensitively but external IDs are case-sensitive.
Normalize phone numbers to E.164 format. Convert all phone fields to +CountryCodeNumber with no spaces or punctuation. Zendesk Talk requires E.164 for number assignment and outbound routing.
Risk if skipped: Mixed phone formats break Talk number lookup and prevent the customer-card phone match from firing.
Identify and resolve records missing required fields. Tickets without a requester, users without an email or an external_id, organizations without a name will all fail import.
Risk if skipped: Required-field failures abort batches of up to 100 records and force re-batching mid-cutover.
Audit tag usage and consolidate. Zendesk does not allow spaces in tags and converts spaces to dashes on import. Inventory tags with spaces or special characters and decide the canonical form before mapping.
Risk if skipped: Tag fragmentation post-migration breaks the views, triggers, and reports built on those tags.
Flag and remove spam, test, and bot-generated tickets. Decide a cut-off (e.g. tickets from blocked domains, tickets with a single body of 'test', tickets created by deactivated test accounts). Export only what should be preserved.

1b. Destination Zendesk setup

Provision the production Zendesk subdomain. Pick the permanent subdomain. Renaming a subdomain post-launch breaks bookmarks, integrations, and SAML configs.
Risk if skipped: A throwaway subdomain locked in by procurement haunts the URL forever.
Provision the sandbox environment. Create the standard sandbox from Admin Center on Suite Professional or above. A standard sandbox is a partial clone with no production tickets; verify what does and does not copy over from production config.
Risk if skipped: Skipping sandbox work means every problem in Phase 4 is a production-data problem.
Contact Zendesk Customer Support to enable data imports. Data imports are disabled by default. The account owner or an admin must open a ticket to enable bulk imports and exports.
Risk if skipped: Discovering imports are disabled on cutover day adds a multi-hour support delay.
Configure org-level locale, timezone, and business hours. Admin Center > Account > Localization. Timezone affects how created_at and updated_at render and how SLA clocks tick.
Risk if skipped: A wrong timezone means every migrated timestamp is off by hours in agent views.
Create all Brands before importing tickets. If the source is multi-brand, recreate each Brand in Zendesk first. Tickets carry a brand_id and the brand must exist on import.
Create all Groups before importing agents. Groups control ticket routing and visibility. Importing agents into a non-existent group silently drops the group assignment.
Create all custom ticket fields with the correct type. Use type checkbox, date, decimal, dropdown, integer, lookup, multiselect, regexp, text, or textarea. Once created the type cannot be changed.
Risk if skipped: Wrong field type forces deletion and re-creation, which orphans every reference in existing tickets, triggers, and macros.
Create all custom user fields and organization fields. User and organization fields appear on the side panel of every ticket. Build them with the same key names used in source so mapping is mechanical.
Bulk-import dropdown and multiselect field options. Per Zendesk, field options can only be CSV-imported during field creation, not after. Build the option list before creating the field.
Risk if skipped: Hundreds of options added one-by-one through the UI burns a week and produces typos.
Provision Custom Objects and Custom Object fields. Custom Objects are available on Support Enterprise and Suite Team and above. Per-account caps: Suite Team 3 objects, Suite Growth 5, Suite Professional 30, Suite Enterprise 50. Each object holds 100 fields, with 5–10 lookup fields per object depending on tier.
Risk if skipped: Hitting the Custom Objects ceiling mid-load forces a schema redesign.
Recreate ticket forms and conditional fields. Multi-ticket-form support requires Suite Growth or higher. Build forms to match the source's ticket-type semantics.
Disable all conditional 'required' rules on ticket forms during import. Per Zendesk's import guidance, every required condition must be set to Never during the import window. Re-enable required after the load completes.
Risk if skipped: Required-field conditions block ticket creation during the import with cryptic validation errors.
Create all custom roles with the correct permission matrix. Suite Enterprise supports up to 197 custom agent roles across eight permission categories. Build them before importing agents so role assignment is part of the user load.
Decide the Light Agent strategy. Light Agents are free read-only seats on Suite Professional or higher who can be CC'd on internal notes. Map source 'collaborator' or 'view-only' users to Light Agents during import.
Risk if skipped: Loading collaborators as full agents inflates the seat count and the bill.
Configure SLA policies and Business Hours schedules. SLA Policies require Suite Professional or higher. Create policies plus their underlying Schedules before importing tickets so SLA clocks attach correctly on load.
Risk if skipped: Tickets imported before SLA policies exist arrive with no SLA target attached, breaking downstream reports.
Set the file-type allowlist on the destination account. Admin Center > Objects and rules > Tickets > Settings > Downloads and attachments. Match the allowlist to the file types present in source export to avoid silent rejection.
Risk if skipped: Default allowlist may reject .key, .pages, or other source file types and skip those attachments.
Pause all destination-side triggers, automations, and outbound notification rules. Triggers fire on every ticket update; mass import will spam customers with notifications, fire Slack hooks, and re-route tickets. Set every trigger to inactive until cutover validation completes.
Risk if skipped: A single 'thank you for contacting us' trigger sends 100k+ emails to customers during import.

1c. People and access prep

Announce the source-system freeze window to all users. Communicate twice: 1 week prior and 24 hours prior. State the exact freeze start time and the destination go-live time.
Schedule destination-platform training sessions per role. Separate sessions for agents, light agents, admins, and end-user help-center users. Schedule them in the 1–2 weeks before cutover.
Designate first-responder agents for the first 48 hours post-cutover. Two to four agents on standby to triage user reports, with a direct Slack channel to the migration DRI.

Phase 2

Source Export

Pull every byte you might need; you cannot re-export after cutover.

Export the full user list with roles, organization, and custom fields. Include user_id, external_id, email, name, role, organization_id, tags, locale, time_zone, and every custom user field.
Export organizations with custom fields and domain mappings. Include organization_id, external_id, name, domain_names, tags, and all custom organization fields. Organizations must load before users that reference them.
Risk if skipped: Loading users before their organizations strands the org_id and forces a second pass to backfill.
Export groups with member rosters. Capture group_id, name, and the agent-to-group mapping for routing rebuild.
Export brands with logos and host mappings. Capture brand_id, name, subdomain or host_mapping, signature, and active state. Reference logos must be re-uploaded; URLs do not transfer.
Export all tickets including closed and archived. Use the Incremental Exports path with a start_time cursor for completeness. Search-based exports cap results and miss closed tickets older than 120 days.
Risk if skipped: Search-based exports silently drop archived tickets and produce a count mismatch in Phase 6.
Export ticket comments (public_comments and internal notes). Capture every comment per ticket along with the public boolean to preserve internal-vs-public on import.
Risk if skipped: Collapsing public and internal comments into a single field on import leaks private notes to end users.
Export ticket audit events for SLA and timestamp reconstruction. Ticket Audits return create, change, comment, SLA target change, satisfaction rating, and status events. Capture the via.channel object to preserve channel of origin.
Risk if skipped: Without the audit log, SLA breach evidence and original channel attribution are lost.
Export ticket attachments to local storage as binary files. For each attachment record, follow content_url and download the binary. Preserve the original filename and the parent ticket and comment IDs in a manifest.
Risk if skipped: Attachment URLs expire and re-fetch becomes impossible after source cancellation.
Export Side Conversations transcripts and attachments. Side Conversations are outbound emails or Slack threads attached to a ticket. Pull every Side Conversation and its events; most destinations do not have a native equivalent.
Risk if skipped: Side Conversation context evaporates because the destination has no parallel data structure.
Export Sunshine Custom Object types, schemas, and records. Native CSV exports flatten or omit Custom Object data. Use the dedicated Sunshine extract to dump object types, fields, records, and relationship records as JSON.
Risk if skipped: Custom Objects vanish from the native export and resurface only after months of post-cutover work.
Export Help Center categories, sections, articles, and translations. Extract Help Center categories, sections, articles, and per-article translations. Capture article_attachments separately.
Export all macros, triggers, automations, and views. These will be rebuilt, not migrated. Dump them as JSON for reference. Capture trigger conditions, actions, and ordering.
Export SLA policies and Business Hours schedules. Capture each policy's filters, target thresholds per priority, and the linked Schedule's time blocks per weekday.
Export satisfaction ratings and CSAT comments. Pull every satisfaction rating. CSAT history rarely has a native destination field; plan to land it in a custom field or archive separately.
Risk if skipped: CSAT trend data evaporates and the team loses the baseline that informs hiring and quality calibration.
Export Zendesk Talk call records, recordings, and voicemails. Capture Voice Comments and the recording URLs. Recordings expire; download binaries during the freeze.
Capture screenshots and JSON exports of dashboards, reports, and Explore queries. Reports do not migrate. Capture the configuration so they can be rebuilt in the destination.
Stamp every export file with its exact export timestamp. Record start_time and end_time of each Incremental Exports call. This timestamp is the delta cursor for the final pre-cutover top-up.
Risk if skipped: Without the cursor, the delta export at cutover either over-pulls and creates duplicates or under-pulls and loses recent tickets.

Phase 3

Transform & Map

The data-engineering core: shape source exports into Zendesk-loadable files.

3a. Mapping spreadsheet

Build the master field-mapping spreadsheet. For each source field record: source object, source field, source type, destination object, destination field key, destination type, transformation rule, nullability, and required-on-import flag.
Risk if skipped: Without a single source of truth for the mapping, every transformation tweak risks a regression nobody catches.
Choose upsert key per object. Use external_id for tickets, users, organizations, and custom object records. For users, email is a secondary unique key; one or the other identifies the record on the Create or Update Many bulk operation.
Risk if skipped: Loading without an external_id key makes re-runs duplicate every record.
Resolve picklist value mismatches. For every source dropdown, write the source-to-destination value map. Zendesk dropdown values are tags under the hood and tags cannot have spaces; spaces become dashes.
Risk if skipped: Unmapped dropdown values arrive as null and reports break silently.
Decide handling for source fields with no destination equivalent. Three options per field: drop, concatenate into the first private note, or create a new custom field. Record the decision in the mapping sheet.
Reserve tag namespace and prevent collisions. Tags cannot be reused across custom ticket fields. If a tag is used in a checkbox field, it cannot be reused as a dropdown option.
Risk if skipped: Tag collisions block future edits to the field configuration and surface as 'tag already in use' errors.
Verify ticket-field count stays under the 400-per-account or 400-per-form limit. Zendesk recommends a 400-ticket-field ceiling per account or per form to preserve performance.
Risk if skipped: Crossing the limit silently degrades agent UI load times and ticket-form submission reliability.

3b. Data transformation

Convert every datetime field to ISO-8601 with explicit timezone offset. Zendesk expects ISO-8601 UTC (e.g. 2026-05-27T14:00:00Z). Native exports may emit account-local timestamps without offset, causing silent timezone drift on import.
Risk if skipped: Timestamps imported without offset land in the wrong day in agent views.
Re-encode every CSV file as UTF-8. Per Zendesk import errors, files in ISO-8859 or Windows-1252 fail with 'invalid byte sequence in UTF-8'. Pass through Google Sheets or use the CSV UTF-8 export in Excel.
Risk if skipped: Encoding errors abort whole files and surface only after the upload completes.
Convert date fields in Sell-style imports to match account timezone format. The importer accepts DD/MM/YYYY and MM/DD/YYYY but only if the format matches the account's regional setting. Mismatched format silently flips day and month.
Risk if skipped: January 4 imports as April 1 with no error.
Truncate string fields exceeding destination character limits. Subject lines, name fields, and most custom text fields have a 255-character cap. Truncate with an explicit ellipsis and log the truncation.
Normalize regex-validated field values. Zendesk regex fields use Ruby regex and reject values that fail the pattern. Test every value against the regex before import. Regex patterns are capped at 255 characters.
Risk if skipped: A single failing value rejects the entire row.
Validate that tags contain no spaces and stay under the 50-tag-per-ticket soft cap. Replace spaces with dashes. Above 50 tags per ticket, performance degrades in views and reports.
Escape commas, quotes, and newlines inside CSV cells. Comment bodies frequently contain commas and CRLF sequences that break naive CSV parsers. Use a CSV writer that double-quotes and doubles internal quotes.
Risk if skipped: A single ragged comment shifts every column to the right for the rest of the file.
Reduce attachment file sizes to fit Zendesk inbound limits. Inbound email and web-form attachments cap at 50 MB per file; outbound email caps at 7 MB per file. Compress, split, or migrate to external storage links anything over the cap.
Risk if skipped: Oversize attachments are silently dropped and the linked binary becomes inaccessible.

3c. Relationship and audit-trail strategy

Use the Ticket Import path to preserve original created_at and updated_at. Standard ticket creation stamps created_at to now. The Ticket Import path accepts created_at, updated_at, and solved_at as overrides. Note: timestamps before 1970 are rounded up to 1970.
Risk if skipped: Using the wrong load path re-stamps every ticket to migration day, destroying SLA history and tenure context.
Set archive_immediately on closed source tickets. When the source ticket is closed, set archive_immediately=true on the Ticket Import call so Zendesk archives the ticket on creation. This avoids quota-burning open-ticket bloat.
Risk if skipped: Loading old closed tickets as open inflates the active-ticket dashboard for months.
Decide ticket-comment author strategy for missing source users. Comments require an author_id. If the source author was deleted, route the comment to a single 'Data Migration' system user and record the original author name in the comment body.
Decide record-ownership strategy for orphaned assignees. If the source assignee no longer exists, assign to a placeholder agent or null assignee. Record the original assignee name in a custom field 'Original Assignee'.
Plan attachment re-upload via the dedicated Uploads path. Attachments must be re-uploaded to obtain an upload token, then referenced in the ticket comment payload. External-storage links are an alternative if file sizes exceed limits.
Mirror lookup-relationship fields by recreating the references. Lookup fields take type=lookup with relationship_target_type values like zen:user, zen:organization, zen:ticket, or zen:custom_object:{key}. Recreate the same target types in destination and load lookup values in dependency order.
Risk if skipped: Lookup fields loaded before the referenced object exists arrive as null and require a second pass.
Migrate legacy Sunshine custom objects to the new Custom Objects framework. Per Zendesk, legacy custom objects retire in June 2026 and new creation stops January 15, 2026. Recreate object types in the new Custom Objects framework, recreate relationships as lookup fields on the 'many' side, then load records.
Risk if skipped: Building on legacy objects in 2026 means a second migration within months.

3d. Helpdesk-specific transforms

Reassemble ticket threads with public and internal comments in order. Order comments by created_at ascending. Preserve the public boolean to keep public_comments visible to the requester and internal notes restricted to agents and light agents.
Risk if skipped: Out-of-order or mis-flagged comments leak internal commentary to end users on first view.
Map Help Center categories, sections, and articles with translations preserved. Help Center hierarchy is Category > Section > Article. Create categories first, then sections inside each category, then articles in each section. Add translations on each article for every locale.
Risk if skipped: Loading translations after switching the default locale strands them under the wrong language.
Rebuild macros from scratch using the captured JSON. Macros are manual ticket actions. They do not migrate; they must be recreated in Admin Center > Objects and rules > Macros referencing the new field keys.
Rebuild triggers with destination field keys, not source field IDs. Triggers are event-driven business rules (fire on ticket create or update). Recreate them with the new ticket-field IDs; copying source trigger JSON verbatim breaks every field reference.
Rebuild automations with destination Business Hours. Automations are time-based rules (e.g. close pending tickets after 4 days of inactivity). They depend on a Schedule; create Schedules first, then automations.
Risk if skipped: Automations referencing the wrong schedule fire at the wrong time and close tickets prematurely.
Decide the SLA-clock strategy for in-flight tickets. Imported open tickets get SLA targets applied based on the active policies and the ticket's current state. Decide whether to override created_at to preserve original SLA clocks or to reset the clock at migration moment.
Risk if skipped: Naive load resets every SLA breach counter to zero and erases compliance evidence.
Land CSAT history in a custom ticket field if no native target exists. Zendesk's native satisfaction_rating accepts good/bad plus a comment. Source platforms with 5-star or NPS-style ratings should map the original score into a 'Legacy CSAT' integer custom field and the raw response into a text field.
Plan email channel cutover including MX record changes. Support email addresses become Zendesk support_address records. To route inbound email through Zendesk, either forward from source mailboxes or repoint MX records. Plan the DNS cutover for the cutover window.
Risk if skipped: MX change before Zendesk addresses are verified bounces every customer email.
Configure Web Widget and Mobile channels. Generate the Web Widget snippet from Admin Center > Channels > Classic > Web Widget or Messaging. Embed on website before cutover but keep messaging conversation creation disabled until go-live.
Port Zendesk Talk phone numbers or order new ones. Talk number porting takes 2–6 weeks with the carrier. Submit Letters of Authorization weeks in advance. For temporary numbers, order Talk numbers and forward source numbers during the freeze.
Risk if skipped: Porting timeline ignored means cutover day has no working phone channel.
Plan Chat-to-Messaging consolidation. Zendesk Chat (legacy) and Messaging are different products with different data models. Decide whether to consolidate onto Messaging and rebuild flows in the Flow Builder, or maintain Chat parity.
Collapse Side Conversations into a private-note transcript. If the destination does not preserve Side Conversations natively, write a single internal note per parent ticket with participants, original timestamps, message bodies, and file references inline.
Risk if skipped: Without the collapse, Side Conversation context is dropped silently and agents lose follow-up trails.
Map source agent seats to Zendesk seat types including Light Agents. Map full-time support reps to Suite agent seats and read-only contributors to Light Agent seats. Light Agents add CCs and internal notes but cannot own a ticket.
Risk if skipped: Loading every contributor as a full agent triples the seat invoice.

Phase 4

Sandbox Test Migration

Catch every problem in sandbox before it costs in production.

Load a representative 1–5% sample of each object into sandbox. Use the Ticket Import path in sandbox. Include open tickets, closed tickets, and tickets with side conversations and attachments to exercise every code path.
Chunk batches to the 100-record per-request cap. Bulk operations like Create or Update Many cap at 100 items per request. Larger payloads return a 400 Bad Request.
Risk if skipped: Exceeding the cap aborts whole batches and forces re-chunking under pressure.
Stay within the 30 concurrent jobs ceiling. Background jobs cap at 30 queued or running. Throttle requests against the remaining job count.
Risk if skipped: Crossing the limit returns TooManyJobs and silently leaves work unfinished.
Validate field-by-field on 10 random sandbox records per object. Open the ticket in agent view. Compare every visible field, including custom fields, side panel user fields, and side panel organization fields, against source.
Test edge cases: long strings, special characters, NULL handling. Push a comment with 32 KB of text, a ticket subject with emoji, a custom field with embedded quotes, and a user with a null organization to surface parser quirks.
Risk if skipped: Edge cases that pass the sample test still fail at production scale; document and fix each one.
Test timezone handling on datetime fields. Pick a ticket with a created_at near midnight UTC and verify it renders on the correct calendar day in agent view.
Test attachment re-upload and inline-image rendering. Upload the binary, attach the token to a comment, and confirm the attachment renders in agent view and downloads cleanly. Verify inline images in comment HTML still resolve.
Risk if skipped: Inline-image src URLs from source remain pointing at the dead source domain and silently break.
Test custom-object lookup relationships. Confirm that a ticket lookup field referencing a Custom Object Record resolves to the correct target record in agent view.
Run a full-volume rehearsal on the sandbox if quota allows. Measure throughput (tickets per minute) and total runtime. Extrapolate to production volume. Confirm the cutover window fits.
Risk if skipped: Production runtime two times the sandbox estimate blows the cutover window and forces a partial rollback.
Verify trigger and automation behavior is fully paused. Create a test ticket via the import path and confirm no triggers fire and no emails go out.
Risk if skipped: An active trigger in sandbox is a live trigger in production until the same admin pauses both.
Get a domain expert from the support team to spot-check sandbox tickets. A senior agent reviews 20 sandbox tickets end to end: thread completeness, custom field correctness, attachment access, side-conversation reconstruction.
Sign off the sandbox test results in writing. Migration DRI and support lead sign a one-page sandbox sign-off document before proceeding to Phase 5.
Risk if skipped: Verbal sign-off evaporates and Phase 5 fails because a known issue was 'going to be fixed later'.

Phase 5

Production Cutover

A tightly sequenced execution window. Idempotency everywhere.

Confirm source-system freeze and verify zero writes in the last hour. Spot-check source audit logs or last-modified timestamps. If writes are still arriving, the freeze is not effective.
Risk if skipped: Partial freeze means the delta export misses tickets created during the freeze ramp.
Take the final delta export from source using the Phase 2 timestamp cursor. Run the Incremental Exports call with start_time set to the cursor saved at Phase 2 end. This captures every change since the baseline export.
Re-confirm all destination triggers, automations, and outbound notification rules are paused. Spot-check Admin Center > Objects and rules and confirm no active rules. A trigger that was paused yesterday may have been re-activated by a colleague.
Risk if skipped: An active 'send response' trigger spams thousands of customers within minutes of the first ticket load.
Load Brands, Groups, custom fields, and Custom Object types first. These are the schema scaffolding. Every later load references them. Confirm they all exist before the user load.
Load Organizations via Create or Update Many Organizations. Key on external_id. Cap at 100 records per request. Capture every job_id for status polling.
Load Users via Create or Update Many Users. Key on external_id with email as secondary. Use the data importer for bulk updates. The legacy Bulk actions page caps at 2,000 rows per file.
Risk if skipped: Loading users without external_id makes re-runs duplicate every user.
Load tickets via the Ticket Import path with archive_immediately on closed tickets. Submit batches up to 100. Set archive_immediately=true for tickets with status=closed. Capture job_ids.
Risk if skipped: Using the standard ticket creation path re-stamps created_at to today.
Poll job status via Show Many Job Statuses. Poll many jobs in one call. Watch for status=completed with per-record error arrays.
Load Custom Object records via Create or Update Many Custom Object Records. Key on external_id. Custom Object records cap at 32 KB each and at 50 million per account.
Load Help Center categories, sections, articles, then translations in that order. Hierarchy must exist before children. Translations must reference an existing article_id.
Risk if skipped: Loading translations before articles errors out the entire translation batch.
Capture every job_id, row count, and error log per load step. Persist to disk. These are the only audit trail of what loaded and what failed.
Risk if skipped: Without per-step logs, Phase 6 reconciliation is guesswork.
If a load step fails partway, stop and diagnose before continuing. Do not push through with partial data. Fix the transform, re-run idempotently from the last known-good external_id.
Reactivate triggers, automations, and outbound notification rules in a staged sequence. Re-enable internal-routing triggers first, then external-notification triggers last. Smoke-test by creating a single test ticket and confirming the expected automations fire and no others.
Risk if skipped: Re-enabling every trigger at once detonates queued background jobs and floods customers.

Phase 6

Validate

Prove the migration was correct. No user access until this phase passes.

Compare total record counts per object: source vs. destination. Counts: tickets (by status), users (by role), organizations, groups, brands, KB categories, KB sections, KB articles. Investigate any delta > 0.1%.
Risk if skipped: Count mismatches discovered after user access exposes the gap to customers first.
Cross-check 25–50 randomly selected tickets field-by-field. Open each in agent view. Compare subject, requester, assignee, status, priority, every custom field, every comment, every attachment against source.
Verify open-ticket count matches source. Open + Pending + On-hold ticket counts. Open backlogs are the most operationally sensitive figure.
Verify ticket-thread integrity on 20 random tickets. Confirm every comment is present, in order, with the correct public_comments vs internal_note classification.
Risk if skipped: A single mis-flagged internal note leaks private commentary to a customer.
Verify attachment accessibility on 30 random tickets. Click each attachment in agent view. Confirm it downloads and renders. Verify inline images in comments display.
Verify ticket-to-requester and ticket-to-organization links. Sample 25 tickets. Confirm requester resolves to the correct user record and the user's organization is correct.
Verify Custom Object Record links from tickets. For every lookup field used, sample tickets and confirm the linked Custom Object Record opens.
Verify created_at and solved_at on a sample of imported tickets. Pick 20 closed tickets. Confirm their created_at and solved_at match source. This is the audit-trail-preservation test.
Risk if skipped: Re-stamped timestamps invalidate every tenure-of-issue and SLA-breach metric.
Verify Side Conversation transcripts landed in private notes. Sample 10 tickets that had Side Conversations in source. Confirm the transcript appears as an internal note with participants and timestamps.
Verify Help Center article rendering, including images and links. Sample 25 articles across categories. Confirm inline images render, internal links resolve to the new Help Center URLs, and translations show under the language switcher.
Risk if skipped: Broken inline-image src URLs persist post-launch and erode help-center trust.
Verify SLA policies attach to imported open tickets. Open 10 imported active tickets and confirm an SLA target shows in the side panel with the correct priority threshold.
Verify Business Hours schedule is applied to automations. Trigger a test automation (e.g. close-after-X-days) on a test ticket and confirm it respects the Schedule.
Verify CSAT history landed in the Legacy CSAT custom field. Sample 20 tickets with source CSAT and confirm the legacy score is visible to agents.
Verify file-attachment file types match the allowlist. Spot-check that .pdf, .png, and any other in-scope types upload and download. Confirm no attachments were rejected silently.
Run compliance verification per Phase 0 obligations. Confirm GDPR data-residency region matches expected. Confirm BAA is signed if HIPAA. Capture SOC2 audit evidence (storage region, encryption-at-rest confirmation).
Have 3–5 power users from the support team validate day-to-day workflows. Each runs a real ticket end to end: open, comment, side-conversation, status change, close. Document any data or workflow gaps.
Document UAT findings and remediate before sign-off. Bug list with priority. Critical and high issues must be fixed before go-live; medium and low can be scheduled for the first 30 days.
Get written sign-off from the migration sponsor. A signed go-live decision document. Names the date, the open-issue list, and the rollback decision criteria.
Risk if skipped: Without written sign-off, accountability for go-live decisions becomes finger-pointing later.

Phase 7

Post-Migration Cleanup

Close out so the team can move on.

Announce go-live to all end users. Email or in-app announcement with login URL, where-to-get-help contact, and a 'what's new' brief. Include the Help Center URL.
Run agent training sessions segmented by role. Separate sessions for full agents, Light Agents, admins, and supervisors. Cover triggers, macros, side conversations, custom-object workflows.
Set the source system to read-only and announce the read-only date. Disable writes via source-system admin settings or revoke write permissions on agents.
Risk if skipped: Leaving source writable means agents accidentally update tickets in source post-cutover and lose those updates.
Rebuild dashboards and Explore reports. Reports do not migrate. Recreate from the screenshots and JSON captured in Phase 2 using Zendesk Explore for analytics.
Re-enable and validate macros in the destination. Have agents run each macro against a test ticket and confirm the actions fire correctly with destination field keys.
Re-establish integrations with calendar, email-sync, BI, and CRM systems. Reconnect Slack, Jira, Salesforce, and any other integrations from the Zendesk Marketplace. Each integration needs new credentials.
Risk if skipped: Forgotten integrations leak ticket activity to the wrong system or to nowhere.
Activate ticket and user deletion schedules per data-retention policy. Admin Center > Account > Data deletion schedules. Configure end-user deletion schedules and ticket deletion schedules to match the retention policy decided in Phase 0.
Set Zendesk storage notifications and review storage usage. Admin Center > Account > Usage > Storage. Storage notifications fire above 80% usage; review the 7/30/90-day trend chart weekly for the first month.
Risk if skipped: Hitting storage limits silently rejects new attachments.
Archive source-system data exports to long-term storage. Move every export file plus the binary attachments to durable archival storage (object storage with versioning). Retention per the policy from Phase 0.
Risk if skipped: Without an archive, any audit, dispute, or 'what did the customer originally say' inquiry post-cutover has no source-of-truth.
Schedule the 30-day post-migration retrospective. Calendar invite issued on cutover day. Attendees: DRI, support lead, agent first-responders, technical lead. Agenda: what went well, what to do differently, residual data issues.
Document and schedule the source-system decommission date. Typically 90 days after cutover. Decommission stops paying for source licenses and removes the temptation to dual-source.
Close out the migration project formally. Final status report. Lessons learned. Project closure document signed by the sponsor.

Watch list

Risks to track throughout

These risks live across multiple phases — keep an eye on them from kickoff through cutover.

Throughput and concurrency ceilings under heavy load

Bulk operations cap at 100 records per request and 30 concurrent background jobs per account. Test runs, mapping fixes, and re-loads compound; a load that fits the cap on paper hits TooManyJobs in practice. Monitor the remaining job count from Phase 4 onward and throttle accordingly.
Audit-trail loss from using the wrong import path

Standard ticket creation re-stamps created_at and updated_at to the request time. Only the Ticket Import path preserves source timestamps. Pre-1970 dates round up to 1970. A single load step on the wrong path destroys SLA history, tenure context, and breach-of-policy evidence for every ticket in the batch.
Sunshine custom-object data loss in native exports

Zendesk's native CSV export flattens or omits Sunshine Custom Object records. Legacy custom objects retire entirely in June 2026 and new creation stops January 15, 2026. Source extraction must explicitly pull Sunshine object types, schemas, records, and relationship records, then load into the new Custom Objects framework. Skipping this step silently abandons the dataset.
Triggers and automations firing during import

Triggers fire on every ticket update including imports done via the standard path. A single un-paused 'thank you for contacting us' trigger can send 100k+ emails to customers within minutes. Confirm in Phase 1 that every trigger and automation is inactive; re-confirm at the start of Phase 5; re-enable in a staged sequence in Phase 5's final step, smoke-testing each batch.