Phase 0

Decide & Plan

Lock scope, owners, and the cutover window before touching data.

Document the business reason for moving to Intercom. Capture whether the driver is AI deflection via Fin, channel consolidation, or messaging-first UX. The answer shapes which data must come with you versus what can be archived.
Define migration scope by object: Contacts (Users and Leads), Companies, Conversations, Tickets, Help Center articles, Custom Data Attributes, Custom Objects. List each object in or out of scope. Tickets are a separate object from Conversations and were added in 2023, so decide explicitly whether prior support threads land as Conversations or Tickets in Intercom.
Risk if skipped: Treating Conversations and Tickets as interchangeable causes mid-migration rework when Intercom rejects the wrong payload shape.
Decide historical-data depth for conversations and tickets. Pick a date floor (e.g. last 24 months of closed conversations plus all open ones) and document it. Wider scope multiplies effort and outcome-billing exposure once Fin is live.
Risk if skipped: An undefined cutoff lets scope creep silently double the cutover window.
Pick the Intercom plan tier — Essential, Advanced, or Expert. Test Workspaces, SLA policies, multibrand routing, and advanced workflow features are tier-gated. Confirm the chosen plan supports every feature on the must-have list.
Risk if skipped: Discovering a feature gap after cutover forces an emergency plan upgrade or unplanned workaround.
Pick the regional hosting location: US, EU, or AU. Regional hosting cannot be changed after the workspace is provisioned. Match the choice to where the bulk of customer data must reside under GDPR or local privacy law.
Risk if skipped: Provisioning in the wrong region forces a full re-implementation if regulators require data residency in a specific jurisdiction.
Document compliance posture: GDPR, SOC 2 Type II, HIPAA attestation, ISO 27001. Pull the relevant artifacts from the Intercom Trust Center for the security-review file. HIPAA requires a signed BAA before any PHI moves.
Risk if skipped: Importing regulated data without the BAA or DPA in place creates a privacy incident on day one.
Assign a single accountable migration DRI plus named owners for data, channels, automation, and end-user enablement. Capture names, escalation order, and decision authority in one place.
Define written success criteria. Spell out the row counts, open-conversation counts, Help Center article counts, and KPI baselines that must reconcile before the migration is declared complete.
Set the cutover date plus the source-system freeze window. Schedule cutover for a low-volume support period. Communicate the freeze to the wider team at least two weeks in advance.
Document explicit out-of-scope items. Examples: do not migrate spam tickets, deleted contacts, archived KB drafts, or social-channel direct messages older than the date floor. Recording exclusions kills future scope creep.

Phase 1

Pre-Migration Prep

Get the source clean and the Intercom workspace ready before any data moves.

1a. Source-system audit and cleansing

Export current record counts from the source for Contacts, Companies, Conversations, Tickets, and Articles. These counts become the reconciliation baseline used in Phase 6.
Run email-based dedup on the source contact list. Normalize all emails to lowercase first, then dedup. Keep the most complete record when duplicates collapse. Fuzzy name+company matching produces 10–25% false positives in B2B lists, so route fuzzy results to manual review rather than auto-merge.
Risk if skipped: Duplicate contacts migrated into Intercom split conversation history across multiple profiles and break Fin context retrieval.
Standardize phone-number formatting in the source. Apply E.164 to every phone field. Intercom's import path can reject malformed numbers when validation is on, so the safer option is to ship clean numbers from the start.
Risk if skipped: Invalid phone numbers stall the import or get silently dropped, depending on validation setting.
Resolve all contacts missing required identifiers. Every Intercom Contact needs at minimum an email or a user_id. Rows missing both will fail.
Risk if skipped: Required-field failures abort batch jobs partway through and force a re-run.
Decide and document the upsert key per object. For Contacts the upsert key is typically external_id or email. For Companies use company_id. Casing matters — Intercom expects lowercase user_id, company_id, external_id keys.
Inventory source custom fields and their current population rate. Drop fields populated on under 1% of records unless legally required. Carrying dead fields into Intercom inflates the CDA count toward the 250 soft cap.
Re-encode source exports as UTF-8. Non-UTF-8 text silently mangles emoji, accented characters, and Asian-language content.
Risk if skipped: Garbled multibyte characters surface as broken glyphs in Inbox replies and bias Fin retrieval.

1b. Intercom workspace setup

Provision the production Intercom workspace in the chosen region. Use a corporate email for the owning admin. Document the workspace ID.
Enable a Test Workspace from Settings on the Advanced or Expert plan. Test Workspaces are free on Advanced and Expert and are the only safe place to dry-run the full migration. Use it as the destination for every Phase 4 sandbox load.
Risk if skipped: Sandbox-testing directly in production contaminates the live workspace with test rows that are painful to delete.
Configure workspace-level settings: timezone, business hours, working language, currency display. These affect SLA clocks and outbound message scheduling once data lands.
Provision Intercom teammates with the correct seat type and role. Seat types include full agents and Lite seats. Lite seats limit feature access. Match the source agent list one-for-one so historical assignments can map.
Risk if skipped: Missing teammates force conversations to load against a generic system user, breaking ownership reports.
Create all Custom Data Attributes (CDAs) on Contact, Company, and Conversation before any import. CDAs must exist before values can be written. Pick the matching CDA type — String (up to 255 chars), Number (integer, no leading zeros), Boolean, or Date. Attribute names ending in _at are auto-typed as dates.
Risk if skipped: Importing a value into a missing CDA is silently dropped on most paths.
Stay under the 250 active CDA soft limit. Archive unused legacy attributes from the source before mapping. Hitting the cap blocks future field creation and slows the settings UI.
Avoid disallowed characters in CDA keys. Periods, dollar signs, backticks, tildes, and similar characters get auto-converted to underscores. Standardize key naming up front so the live key matches the mapping spreadsheet.
Define Custom Objects for any source data that does not fit Contacts, Companies, Conversations, or Tickets. Examples: subscriptions, devices, orders. Create the schema and relationships in Settings > Data before import.
Configure Ticket Types with the fields each type needs. Tickets in Intercom are typed (e.g. back-office, customer-facing). Each type has its own attribute set. Match source ticket categories to Intercom Ticket Types up front.
Risk if skipped: Loading tickets without their type pre-configured produces orphan tickets that lose field structure.
Enable Messenger HMAC identity verification for the destination domains. HMAC prevents impersonation of logged-in users on the Messenger. Generate the secret in Settings > Security and pre-stage the integration code.
Risk if skipped: Going live without HMAC exposes every logged-in user to identity spoofing through the Messenger.
Pause every Workflow, Series, and Outbound message in the destination workspace. Any live automation will fire as imported contacts match its rules — sending real emails or push notifications to your customers. Confirm pause status before the first row lands.
Risk if skipped: An active welcome Series can blast thousands of greeting emails to customers who were already onboarded years ago.

1c. People and channel prep

Announce the source-system freeze window to all agents and stakeholders. Include cutover date, freeze start time, and a single point of contact for exceptions.
Schedule Intercom Inbox training for support agents at least one week before cutover. Cover Inbox basics, Macros, ticket routing, and the new Help Center editor. Record the session for late joiners.
Designate first-responder agents for the first 48 hours after cutover. These agents triage user issues, broken Macros, and routing errors before they escalate.
Plan the channel cutover sequence: email forwarding, Messenger install, WhatsApp, SMS, social. Each channel has its own connection mechanism. Sequence them in the cutover runbook so traffic flows to Intercom in a controlled order, not all at once.

Phase 2

Source Export

Pull a complete, timestamped snapshot of everything in scope.

Export source contacts with every standard and custom field. Include the field that will become external_id in Intercom — this is the upsert key.
Export source companies with their identifiers, plan, size, and any custom attributes. Capture the company_id value you will use for the Intercom Company. Every Company needs a company_id to be created.
Export the contact-to-company association table. Intercom Companies and Contacts are linked many-to-one through company_id, so the join file is required to rebuild the relationship.
Export closed conversations and tickets as a separate file from open ones. The recommended sequence migrates all closed conversations first (the bulk) and a small delta of open conversations during cutover. Splitting at export time enforces that pattern.
Risk if skipped: Bundling open and closed together forces a re-export at cutover and inflates the freeze window.
Export conversation parts in original order with author, timestamp, channel, and body. Each Intercom conversation is a series of parts (messages, notes, assignments). Preserve sequence so the reconstructed thread reads correctly.
Risk if skipped: Out-of-order replies become a confusing audit trail that erodes agent trust in migrated data.
Export internal notes separately or flagged. Intercom represents internal notes as conversation replies with message_type=note. Flagging at export keeps them from leaking to customers post-import.
Risk if skipped: Internal notes loaded as customer-visible replies expose private agent commentary to the end user.
Export attachments — both the binary files and the metadata linking each file to its conversation, ticket, or contact. Messenger uploads accept up to 100 MB per file. Files above that need a hosted-link strategy decided in Phase 3.
Export agent users with email, role, group, and reporting structure. This is the source-side equivalent of the Intercom teammate list and drives ownership mapping.
Export the Help Center: collections, sub-collections, articles, multilingual translations, and inline media references. Capture the article URL, slug, author, status (published/draft), tags, and any redirects you need to preserve for SEO.
Risk if skipped: Skipping redirects breaks every external link into the old knowledge base on cutover day.
Export Macros, Triggers, Automations, SLAs, CSAT surveys, and reports as configurations or screenshots. These do not transfer programmatically. Capturing the source config is the only way to rebuild them in Workflows, Series, and Outbound.
Record the exact UTC timestamp of each export. This becomes the delta boundary at cutover. Document the timestamp at the top of the runbook.
Hash and store every export file before transformation begins. A SHA-256 of each file lets you prove later that production data was not mutated by accident.

Phase 3

Transform & Map

Convert exports into Intercom-shaped payloads with every relationship and history preserved.

3a. Mapping spreadsheet

Build a master mapping spreadsheet keyed by object: Contact, Company, Conversation, Ticket, Article, Custom Object. For every source field record: destination attribute, attribute type, transformation rule, nullability, and required-on-create flag. This becomes the single reference all transform code reads from.
Risk if skipped: Without a single mapping document, undocumented field decisions surface as data loss at validation.
Map source identifier to Intercom external_id on Contact. external_id is the primary upsert key for Contacts. Choose the source field most likely to be globally unique and stable.
Map source company identifier to Intercom company_id on Company. company_id is the upsert key for Company. Casing must match exactly.
Decide handling for source fields with no Intercom equivalent. Options: drop, fold into a Conversation Data Attribute, concatenate into the contact bio, or create a new CDA.
Resolve picklist value mismatches between source and Intercom CDAs. Example: source ticket priority "P1/P2/P3" maps to Intercom CDA values "High/Medium/Low". Document the translation table.
Split or truncate any text value that exceeds the Intercom 255-character limit on String CDAs. String CDAs cap at 255 characters. Long source notes need to be moved to conversation notes, custom long-text targets, or split across multiple fields.
Risk if skipped: Silently truncated strings lose critical context (e.g. the end of a long license key).
Decide the Contact subtype for each source row: User or Lead. Intercom Contacts split into Users (logged-in customers) and Leads (anonymous prospects). Make the call per row before transform.

3b. Data transformation

Convert all date fields to Unix epoch seconds in UTC. Intercom date attributes expect Unix timestamps in seconds. Settings UI also accepts DD/MM/YYYY. Pick one format per pipeline and stick to it.
Risk if skipped: Mixing seconds vs milliseconds shifts every timestamp by a factor of 1,000, putting events in 1970 or the year 52000.
Name every date CDA with an _at suffix. Intercom auto-types attributes ending in _at as dates. Without the suffix the value is interpreted as a number.
Strip leading zeros from numeric values. Intercom Number CDAs cannot start with "0" — "01234" must be stored as 1234. ZIP codes and IDs with required leading zeros must move to String CDAs.
Risk if skipped: Stored numbers lose leading zeros, corrupting postal codes, account numbers, and ID strings.
Re-encode every text payload as UTF-8 before serialization. Apply a final pass even if the export claimed UTF-8 — mixed encodings still survive in legacy CRMs.
Escape newlines, embedded commas, and quote characters in CSV exports. Intercom's CSV importer mis-parses unescaped values and splits a single record across two rows.
Map source ownership to an Intercom teammate ID per conversation. Match by email. For source users without an Intercom counterpart, route to a dedicated 'Migration' teammate so the assignment field is never null.
Decide and implement the historical-timestamp strategy for conversations. Intercom records the import time as the conversation created_at. To preserve the original date, store it in a Conversation Data Attribute (e.g. source_created_at_at). Document the decision so reports treat it as the canonical date.
Risk if skipped: All migrated tickets appearing as 'created today' breaks SLA reporting and ages-out analytics.
Decide attachment strategy: re-upload binaries into Intercom or rewrite as external storage links. Messenger uploads cap at 100 MB per file. Anything larger must be hosted externally (e.g. S3, Drive) and referenced by URL in the conversation body.

3c. Relationship and ordering decisions

Document the strict load order: teammates → Companies → Contacts → CDAs/Custom Objects → Conversations/Tickets → Notes → Attachments. Per Intercom's historical-migration guidance, every conversation and ticket must reference an existing contact. Contacts must reference an existing Company. Loading out of order produces orphan records.
Risk if skipped: Out-of-order loads fail validation and leave thousands of partial records to clean up manually.
Pre-populate every Conversation and Ticket payload with from_type and from_id pointing at the imported contact. Intercom Conversations require from_type (User, Lead, or Contact) and from_id. Build the join during transform so the payload writes are simple.
Verify lowercase casing on every external_id, company_id, and user_id key. Intercom keys are case-sensitive. Mixed-case keys silently create new records instead of updating existing ones.
Risk if skipped: Casing drift creates duplicate Contacts and Companies on every retry.

3d. Helpdesk-specific transforms

Reassemble each conversation thread from source ticket comments by creating the conversation with the first message and then loading each subsequent reply in order. Conversations are created with one initial message, then every subsequent reply is appended with the original timestamp captured in a CDA. The ordering of the appends determines the rendered thread.
Risk if skipped: Skipping ordered replays scrambles the agent–customer dialogue and erodes case context.
Translate source private comments into Intercom internal notes. Internal notes are loaded the same way as a reply but with message_type set to note. Anything else becomes a customer-visible reply.
Risk if skipped: Mis-typing a note as a reply leaks internal agent commentary to the end customer.
Rebuild Help Center collections, sub-collections, and articles in Intercom's Articles area. Create collections first, sub-collections next, then articles. Preserve slug, locale, author, and publication status. For multilingual content create one article per locale under the same collection.
Risk if skipped: Loading articles before their parent collection exists drops articles into an unrouted default folder.
Rewrite every internal Help Center cross-link to point at the new Intercom article URL. Old article URLs will not resolve after cutover. Build a slug→new-URL map and rewrite inline links before publish.
Risk if skipped: Broken internal links degrade Fin's retrieval and frustrate self-service users.
Rebuild Macros from source canned responses inside Intercom's Macros editor. Macros do not transfer programmatically. Capture the source text and any triggered actions (tag, assign, close), then recreate one by one. Group by team owner.
Recreate source automation rules as Workflows, Series, or Outbound messages. Workflows handle conditional automations (assignment, tagging, bot logic). Series handle multi-step outbound journeys. Outbound handles single proactive messages. Pick the right tool per rule.
Rebuild SLA policies under the tier-restricted SLA feature. SLAs require the Advanced or Expert plan. If you are on Essential, the SLA clock cannot be reconstructed natively — capture target response times in a CDA and report on them externally.
Risk if skipped: Loading old tickets on Essential silently strips every SLA expectation from the historical record.
Re-create CSAT surveys via Intercom's post-conversation rating mechanism. Historical CSAT scores from the source platform should land in a Conversation CDA so reports can still slice by them. New CSAT goes through Intercom's native post-conversation rating.
Plan the multi-channel connection: Messenger (web/mobile with HMAC), inbound email forwarding, WhatsApp, SMS, Instagram, Facebook, and phone via the Twilio integration. Each channel has its own setup. Document the cutover order in the runbook so traffic does not hit Intercom before agents are trained.
Configure Fin AI Agent content sources before turning Fin on. Fin is billed at $0.99 per outcome. Pointing it at the wrong Help Center collections or unreviewed articles costs real money on every hallucinated answer.
Risk if skipped: Enabling Fin against unscoped content runs up an unbounded outcome-billing tab and damages customer trust.

Phase 4

Sandbox Test Migration

Catch every problem in the Test Workspace before it costs you in production.

Load a representative 1–5% sample of each object into the Test Workspace. Include at least one of every Contact subtype, Conversation channel, Ticket Type, Custom Object, and locale.
Validate field-by-field on 10 random sandbox Contacts. Open each record in the Intercom UI and confirm every CDA, ownership field, company association, and historical timestamp matches the source.
Confirm the conversation thread renders top-to-bottom in the correct order. Open at least 20 migrated conversations and verify reply order, note privacy, attachment links, and author attribution.
Risk if skipped: An incorrect reply order discovered in production is much harder to diagnose than catching it now.
Test long strings near the 255-character CDA limit. Confirm trimming behaves as expected and no silent truncation occurs.
Test special characters: emoji, accented Latin, CJK, RTL Arabic and Hebrew. These commonly break on legacy CSV pipelines. Catch the regression in sandbox.
Test the _at date convention by writing one timestamp far in the past (e.g. 2015) and one far in the future. Verify both render as dates, not numbers, in the UI.
Risk if skipped: A misnamed date attribute reports as a giant integer and silently breaks date-based segmentation.
Test attachment upload at boundary sizes: 1 KB, 5 MB, 99 MB, and 100 MB. Confirm the 100 MB Messenger upload ceiling and the external-link fallback for anything larger.
Test contact bulk operations at the per-call record ceiling. Intercom's contact bulk operations accept up to 10 records per call. Right-size batch jobs so they hit the ceiling without going over.
Test relationship preservation: contact-to-company, ticket-to-contact, article-to-collection. Pull 20 sandbox records and verify each foreign-key link in the UI.
Test Conversation Data Attributes that exceed the 10-metadata-key cap on Events. Event metadata caps at 10 keys per event. Excess keys are silently ignored.
Time the sandbox loads end-to-end and extrapolate to full production volume. Project the total wall-clock time and compare against the planned cutover window. Adjust batch size or window if the projection exceeds the freeze period.
Risk if skipped: An unrealistic cutover window leaves agents locked out of both systems when the freeze runs over.
Have a senior support agent spot-check 25 sandbox conversations for correctness. Domain expertise catches subtle data issues that automated checks miss.
Capture written sandbox sign-off from the migration sponsor and the support lead. Production cutover does not begin until this sign-off is on file.

Phase 5

Production Cutover

Execute the migration in a tight, idempotent, dependency-ordered sequence.

Confirm the source-system freeze is in effect and verify no new writes have happened in the last 60 minutes. Capture the freeze confirmation in the runbook with a timestamp.
Take a final delta export of every in-scope object since the Phase 2 baseline. Delta exports are smaller and faster than full re-exports and cover only changes since the prior cutoff.
Confirm every Workflow, Series, and Outbound message is still paused in the destination workspace. An accidentally enabled welcome message can blast tens of thousands of customers within minutes of import.
Risk if skipped: A live Workflow triggers customer-facing emails the moment imported records match its rules.
Load teammates first, then Companies, then Contacts. Foreign-key dependencies require parent records before child records. Skipping the order produces orphan rows.
Load Custom Data Attributes and Custom Objects before any value-carrying records reference them. Per the historical-migration sequence, CDAs must exist before conversations or tickets carry CDA values.
Risk if skipped: Conversation imports running against missing CDAs silently drop the attribute value.
Load closed conversations and tickets first, then open ones during the cutover window. This sequencing matches Intercom's recommended pattern — the bulk historical load can happen in the days before, and only the smaller open-ticket delta needs the strict freeze window.
Use idempotent upserts keyed off external_id, company_id, or user_id on every load. Idempotent writes mean a retry after a transient failure does not duplicate the record.
Risk if skipped: Non-idempotent retries silently double-create Contacts and Companies.
Capture every import job's response: status, retry count, throttling headers, and error payload. Persist the log to durable storage — it is the forensic record if a row count is off later.
Reconcile row counts at each load step against the source delta. Catch a missing batch within minutes, not days.
Stop and diagnose any load step that fails midway — do not push through with partial data. Fix the transform, then re-run idempotently from the last known good checkpoint.
Risk if skipped: Continuing past a partial failure layers errors that take days to unwind.
Re-enable Workflows, Series, and Outbound messages only after all loads complete and Phase 6 validation has begun. Bring automation back online in a controlled order, with a smoke-test message to a known internal address first.
Switch inbound email forwarding to the new Intercom inbox address at the official go-live timestamp. This is the technical moment of cutover for the email channel. Capture the timestamp.

Phase 6

Validate

Prove the migration before letting users into the destination.

Compare total Contact count: source vs destination. Investigate any delta. Common causes: dedup collapse, dropped Leads, casing-driven duplicate creation.
Compare total Company count: source vs destination. Companies often have looser identity rules than Contacts — a delta here often reveals casing or company_id drift.
Compare total Conversation count, broken down by status (open, closed, snoozed). Open-vs-closed splits matter because the SLA clock applies only to open conversations.
Compare total Ticket count, broken down by Ticket Type. If a Ticket Type shows zero on destination but had volume on source, the type was likely misconfigured in Phase 1.
Risk if skipped: A zero-count type means an entire ticket category was silently dropped.
Reconcile Help Center article counts per collection and per locale. Multilingual articles often miss locales when the loader treats each locale as a separate article rather than a translation.
Cross-check 25–50 randomly selected Contacts field-by-field between source and destination. Random sampling catches transformation bugs that don't show up in row counts.
Verify every CDA value type renders correctly in the UI. Strings under 255 chars, Numbers without leading zeros, Booleans as true/false, Dates as dates not integers.
Verify Contact-to-Company associations on a 25-record sample. Click into each contact and confirm the linked company matches the source join file.
Verify conversation thread order, author attribution, and note privacy on 30 migrated conversations. Spread the sample across channels (Messenger, email, WhatsApp, SMS) and across time.
Risk if skipped: Channel-specific defects only surface when sampling covers every channel.
Verify attachment accessibility on 20 conversations that have attachments. Click each attachment and confirm the file downloads. Externally hosted links should resolve and not 403.
Confirm the historical-timestamp strategy worked. Pick 10 migrated conversations and confirm the source_created_at_at CDA (or chosen equivalent) holds the original date, even though Intercom's native created_at shows the import time.
Run a compliance check appropriate to the chosen region. EU workspaces: confirm data resides in the EU region and the DPA is signed. HIPAA workspaces: confirm BAA is in place and PHI fields land in attributes covered by the BAA.
Risk if skipped: A missed compliance step on cutover day is much harder to remediate than to verify now.
Have 3–5 support agents validate that their daily workflows work with the migrated data. Cover routing, Macros, Workflows, Help Center search, and the Inbox filters they rely on.
Document any data issues surfaced during user-acceptance testing. Triage each into: blocker (fix before go-live), nice-to-have (fix in Phase 7), or accept-as-is.
Verify Teammate Activity Logs are recording cutover actions. Teammate Activity Logs retain one year of history and are available on all plans. Confirm log writes are flowing — this is the audit trail used for any later forensic review.
Capture written sign-off from the migration sponsor before announcing go-live. Sign-off closes the validation phase. Without it, scope creep from Phase 7 leaks back into Phase 6.

Phase 7

Post-Migration Cleanup

Decommission the source, lock the destination as the system of record, and capture lessons learned.

Announce go-live to all end users with login instructions and the support contact for issues. Pair the announcement with a single hand-raise channel (e.g. #intercom-help) for first-week issues.
Run role-segmented training sessions on the Intercom Inbox, Help Center editor, and Workflows. Tier the depth by role — full agents need more, Lite-seat reviewers need less.
Set the source helpdesk to read-only. Announce the read-only date in advance so anyone with pending work has time to migrate it.
Set up redirects from old Help Center URLs to the new Intercom article URLs. Use the slug→new-URL map built in Phase 3. SEO and existing inbound links depend on this.
Risk if skipped: Missing redirects evict the entire knowledge base from search rankings.
Rebuild reports and dashboards in Intercom Reports. Source reports do not transfer programmatically. Rebuild the top 10 by usage first.
Bring Workflows, Series, and Outbound back online in a measured sequence. Start with internal-only Workflows, then welcome messages to a 1% test segment, then full audience.
Re-establish integrations: CRM sync (Salesforce, HubSpot), calendar, BI exports, and any other custom integrations. Each integration needs its own auth and its own smoke test.
Turn on Fin AI Agent against a vetted subset of the Help Center. Fin is billed at $0.99 per outcome. Limit the audience and content scope at launch, monitor outcome volume daily, and scale up only after answer quality is validated.
Risk if skipped: Switching Fin on with no scoping pushes outcome billing to unbounded levels in the first week.
Schedule a 30-day post-migration retrospective. Capture what worked, what didn't, and any data issues that surfaced after the validation phase closed.
Document the source-system decommission date. Typically 90 days post-cutover, after the final reporting and retention window has elapsed.
Archive source-system exports to long-term storage with checksums. Retain for the regulatory window applicable to your industry. Verify SHA-256 hashes on a sample annually.
Close the migration project formally with a final status report and lessons-learned doc. Capture the runbook, the issue log, the timing breakdown, and the final reconciled counts.

Watch list

Risks to track throughout

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

CDA soft cap and 255-character string limit

Custom Data Attributes have a 250-active soft limit per workspace, and String CDAs cap at 255 characters. Source systems with hundreds of fields or long free-text values silently lose data when these limits are hit. Audit field population rates in Phase 1, archive rarely-used attributes, and move long text to conversation notes or external storage before transform.
Fin outcome billing drift

Fin AI Agent charges $0.99 per outcome with no seat fee, which makes total cost unbounded and hard to forecast. An over-eager Fin pointed at unreviewed Help Center content runs up real billing on every answered question. Scope Fin's content and audience tightly at launch, monitor outcome volume daily, and set internal alerts on weekly outcome counts.
Conversations vs Tickets confusion

Intercom Conversations and Tickets are distinct objects with different field shapes — Tickets was added in 2023 as a separate concept. Treating them as interchangeable causes mid-migration rework and orphan records. Lock the source-to-destination mapping in Phase 0 and document, for every source ticket category, whether it lands as a Conversation or a typed Ticket.