Data Migration Validation for Help Desk, ITSM & CRM
Data migration validation for ticket and CRM data: reconcile counts, diff fields, check links, dates and files before you sign off. Read the checklist.
Published Updated 12 min read

On this page
- What does data migration validation mean for ticket data?
- What should you validate, object by object?
- Method 1: record count reconciliation
- Method 2: field-level diff on a sample
- Method 3: relationship and timestamp checks
- Method 4: attachments and inline images
- Which platform limits affect validation?
- Data integrity checklist: before, during and after cutover
- For MSPs and SIs: acceptance criteria for the SOW
- For software vendors: validation as an onboarding step
- How MigrateX does this
- Data migration validation FAQ
- Changelog
Data migration validation is the work of proving that the target system holds the same records as the source, with the same field values, relationships, authors, timestamps and attachments, before you sign off. For help desk, ITSM and CRM data, it comes down to four methods: reconcile record counts, diff fields on a sample, check links and dates, and check files. Each produces evidence someone can sign.
This guide is for teams about to accept a migration of tickets, incidents, cases or CRM records, and for MSPs and systems integrators who owe a client a QA deliverable.
What does data migration validation mean for ticket data?
It means showing, record by record, that nothing was changed, lost or invented in transit. NIST defines data integrity as "the property that data has not been changed, destroyed, or lost in an unauthorized or accidental manner" (NIST CSRC glossary, citing CNSSI 4009-2022, as of 4 Nov 2026). Validation is how you prove that property after a move.
Three words get used loosely here. NIST defines only the first; the other two are our working definitions.
| Term | Question it answers | How you measure it | Ticket example |
|---|---|---|---|
| Data integrity | Was anything changed, destroyed or lost? | Counts and checksums match, with every difference explained | Closed tickets in the source equal closed tickets in the target, plus an exceptions list |
| Data accuracy | Is each value right? | Field-level diff against the mapping spec | Priority "Urgent" in the source is "Urgent" in the target, not "High" |
| Data fidelity | Does the record still mean the same thing? | Reviewer opens records and checks context | Internal notes still internal, replies in original order, original author and date shown |
Data fidelity vs data integrity
Integrity is the strict test: same records, same bytes. Fidelity is the human one: a ticket can keep every comment and still fail fidelity if each shows the integration user as author and the load date as its time. A full-fidelity migration passes both.
Validation is not the same as a test migration. A test migration is a rehearsal you run before signing a contract. Validation is the evidence you produce on every run, test and production alike, and the production run is the one that counts.
What should you validate, object by object?
Validate every object in scope. If scope is not settled yet, start with our guide on whether to migrate ticket history.
| Object | What to check | Defect it catches |
|---|---|---|
| Tickets, incidents, cases | Counts by type, status and created year | Missing years, status mapping that drops a value |
| Public replies vs internal notes | Visibility flag per comment | Internal notes exposed to requesters |
| Comment authors | Author per comment, including departed agents | Every comment shown as the integration account |
| Timestamps | Created, updated, solved or closed, and per comment | Dates reset to the load date, time zone shifts |
| Attachments and inline images | Count, size and checksum per record; images render | Missing files, broken images pointing at the old instance |
| CCs, followers and tags | Lists match per ticket | Silent drops when a user was not created first |
| Custom fields | Value per field per sampled record | Unmapped picklist values left blank |
| SLA and CSAT data | Historic outcomes in an agreed field | Timers recalculating on closed history |
| Knowledge articles | Count, body, images, translations | Missing translations, dead internal links |
| CIs and assets | Count by class; links to tickets | Orphaned CIs, links stored as text |
| Problem, change and parent links | Links resolve to migrated records, both directions | Links that point nowhere |
| Requester and organization links | Each record linked to the right person and company | Records linked to the wrong or default organization |
Method 1: record count reconciliation
Count reconciliation is the cheapest check: freeze source counts, run the load, compare the same buckets on both sides. Never compare one grand total, which can match while a year is missing and duplicates fill the gap.
Bucket by record type, status and created-date band (a year, or a quarter for recent data). Every bucket must satisfy one equation:
source count = target count + excluded + failed
"Excluded" means spam, test and other records you agreed to leave behind, listed by ID. "Failed" comes from the loader's error log. Anything else is a defect.
Typical count queries look like this:
# Zendesk: count tickets matching a query
GET /api/v2/search/count?query=type:ticket status:closed created>2024-12-31 created<2026-01-01
# ServiceNow: Aggregate API, grouped
GET /api/now/stats/incident?sysparm_count=true&sysparm_group_by=state,priority
# HubSpot: search with limit 1, then read "total" from the response
POST /crm/v3/objects/tickets/search
-- Salesforce: SOQL count by status and created year
SELECT Status, CALENDAR_YEAR(CreatedDate) yr, COUNT(Id)
FROM Case
GROUP BY Status, CALENDAR_YEAR(CreatedDate)
The ServiceNow endpoint and its sysparm_count, sysparm_group_by and sysparm_query parameters come from the ServiceNow Aggregate API docs (as of 4 Nov 2026).
Store frozen counts with a timestamp, and reconcile again after the final delta sync.
Method 2: field-level diff on a sample
Counts prove records arrived; a field-level diff proves their values are right. It is the core of data integrity testing: pull each sampled record from both systems, normalize, and compare field by field against the mapping spec.
Pick a sample that can fail
Use two samples, not one:
- A targeted sample of hard records, chosen by ID before the run. Our test migration checklist lists the shapes to include. Easy records prove nothing.
- A random sample, to estimate the defect rate across everything else.
How big should the random sample be?
No public standard sets a sample size for ticket migrations. Probability gives a floor. If a fraction p of records is broken, a random sample of n records contains at least one broken record with probability 1 − (1 − p)ⁿ. To have a 95% chance of catching a defect:
| Defect rate you want to catch | Random sample needed (95% chance of at least one hit) |
|---|---|
| 5% (1 in 20) | 59 records |
| 2% (1 in 50) | 149 records |
| 1% (1 in 100) | 299 records |
| 0.5% (1 in 200) | 598 records |
Sample each object separately. Zero defects in 299 random records lets you say, at 95% confidence, that the true rate is roughly 1% or lower, not that it is zero.
Normalize before you compare
Most false alarms come from formatting, not data. Compare stripped text rather than HTML, convert dates to UTC, trim whitespace, and translate source picklist values through the mapping table first, so "on-hold" vs "Pending" is expected, not a defect.
Log each mismatch with record ID, field, source value, target value and a verdict: defect, expected, or accepted.
Method 3: relationship and timestamp checks
A record can have every field right and still be wrong if it is linked to the wrong parent or dated to the load day.
Relationships
Check each link from both ends: a sampled incident's problem, change and CI links must resolve to migrated records, and the problem must list the incident. For CRM data, confirm each case sits on the right contact and company. Then run orphan counts in the target: records with no requester, organization or parent where the source had one. Any non-zero result needs an explanation.
Timestamps and authors
Compare created, updated and solved dates on sampled records, plus each comment's time and author.
Zendesk's Ticket Import API sets ticket and comment dates. Salesforce keeps source audit values only once the "Set Audit Fields upon Record Creation" permission is on. Jira Service Management's request comment API takes no author or date at all, so each comment shows the calling account. The constraints table below has the detail.
Where the target cannot set an original date or author, store it in a custom field or a header line in the comment, agreed before the run, and validate that instead.
Method 4: attachments and inline images
Attachments fail quietly: a ticket can arrive with fields and comments intact while its files are missing, and the count still passes. For each sampled record:
- Count files on both sides. Count inline images separately. Zendesk leaves inline attachments out of a ticket's attachment list; list them through List Comments with the
include_inline_imagesparameter (Zendesk ticket attachments docs, as of 4 Nov 2026). - Compare size and checksum. A matching file name with a different size is a truncated or re-encoded file.
- Open a handful. Check that PDFs, images and logs open from the target.
- Check where inline images load from. An image still served from the old instance looks fine until the source is switched off. Search target comment HTML for the old domain.
Then test each target's size limit with one oversized file, because limits differ by platform and API path:
| Target | Documented limit (as of 4 Nov 2026) |
|---|---|
| Freshservice | Total size of all of a ticket's attachments "cannot exceed 40 MB" (Freshservice API docs) |
| ServiceNow | com.glide.attachment.max_size defaults to 1024 MB (ServiceNow docs) |
| Salesforce REST API | 2 GB per file for ContentVersion, 500 MB for other eligible objects, 500 MB per sObject Collections request (Salesforce REST API Developer Guide) |
| Jira Service Management, CSV import | Files referenced by URL in an Attachments column, HTTP or HTTPS (Atlassian support) |
Which platform limits affect validation?
The import path decides what you can validate: if the target cannot store an original date, you can only check the field you put it in.
| Platform and path | Original timestamps | Comment author and visibility | Validation gotcha |
|---|---|---|---|
| Zendesk, Ticket Import API | Ticket and comment dates settable; pre-1970 rounded up | author_id shown in examples, rules not documented; public per comment | "Metrics and SLAs are not supported for imported tickets"; up to 100 tickets per bulk request; archive_immediately suggested for 750,000+ tickets |
| Freshservice, public API | created_at described only as "Ticket creation timestamp"; Create a Note lists no date attribute | Note takes user_id; private defaults to true | List All Tickets returns only the past 30 days by default; use updated_since |
| Jira Service Management, CSV import | Date Created column, in the format set in the wizard | All imported comments become public in JSM spaces | Atlassian recommends about 1,500 work items per file |
| Jira Service Management, REST API | Not settable on request comments (JSM REST API) | Comment API takes body and public flag only | Every comment shows the calling account as author |
| ServiceNow, Table API | sys_ fields "are typically system parameters that are automatically generated and cannot be updated" | Not covered on the Table API page | Plan where sys_created_on originals go before mapping |
| HubSpot, CRM API | Notes: hs_timestamp sets the note's time on the timeline | hubspot_owner_id sets the note creator shown | Imported activities without an Activity date get the import time; search stops at 10,000 results |
| Salesforce, API | Audit fields settable on create with the "Set Audit Fields upon Record Creation" permission, on Case, CaseComment, ContentVersion, Task and other listed objects; never SystemModstamp (Salesforce System Fields) | CreatedById settable with the same permission | Audit dates must be on or after 1 Jan 1970 |
Sources: Zendesk, Freshservice, Atlassian, ServiceNow, HubSpot (notes guide, import file guide) and Salesforce documentation, checked 4 Nov 2026.
Moving from Zendesk? Our guide on how to export tickets from Zendesk covers getting a complete source extract to validate against. For a CRM target, our Zendesk to Salesforce Service Cloud migration guide covers the mapping side.
Data integrity checklist: before, during and after cutover
This data migration testing checklist follows the order of the work.
Before the load
- Agree scope by object and date band, in writing.
- Write the mapping spec, including which field holds original dates and authors where the target cannot set them.
- Freeze and store source counts by type, status and year.
- Pick the targeted sample by record ID; generate the random sample.
- Create users first, departed agents included, so authors and CCs can resolve.
- Mute target notifications, triggers and workflows for the load window.
During the load
- Capture the failed-record log with a reason per failure, re-run failures, and log the second result.
- Reconcile counts after the bulk load, before the delta.
After cutover
- Run the final delta sync and reconcile counts again.
- Run the field diff, relationship, orphan, timestamp and attachment checks on both samples.
- Log in as a requester and confirm no internal note is visible.
- Sign the reconciliation report and exceptions log, then set the source read-only.
For MSPs and SIs: acceptance criteria for the SOW
Write validation into the statement of work so "done" has a definition. These criteria are our recommendations, not an industry standard (more on how we work with MSPs and systems integrators):
| Criterion | Pass |
|---|---|
| Count reconciliation | Every bucket balances: target + excluded + failed = source |
| Failed records | Each failure listed by ID with a reason and a fix or written acceptance |
| Field diff, targeted sample | Zero unexplained defects |
| Field diff, random sample | Sample size and detectable defect rate stated; zero unexplained defects |
| Relationships | Sampled links resolve both ways; orphan counts explained |
| Dates, authors, attachments | Originals in the agreed field; file counts and checksums match; no image served from the old instance |
| Deliverable | Reconciliation report, exceptions log, sample diff, signed by both sides |
For software vendors: validation as an onboarding step
If you sell the target platform, a validation report shortens the road from signature to go-live: a customer who sees their own counts reconciled and hardest tickets diffed trusts the data from day one. Build the four methods into onboarding and hand the customer the report. See how we work with software vendors.
How MigrateX does this
MigrateX runs the four checks above and hands over a reconciliation report with the exceptions log and sample diff, reconciled after the bulk load and again after the final delta.
The same checks hold at scale: for a UAE technology company we moved 5 million records with multiple delta syncs.
MigrateX is SOC 2 Type II, GDPR compliant and HIPAA compliant, with a DPA and BAA available (see security). To see the checks on your own data, ask for a free 100-record sample with a validation report, and get a fixed quote within 24 hours.
Facts checked November 4, 2026 from NIST CSRC glossary, Zendesk developer docs, Freshservice API docs, Atlassian support, ServiceNow docs and Salesforce Object Reference. No vendor paid for this article. How we research
Data migration validation FAQ
What is data migration validation?
It is proving that migrated data matches the source before sign-off: reconcile counts by type, status and date band, diff fields on a sample, check relationships, dates and authors, and check attachments. The output is a reconciliation report and exceptions log both sides sign.
What is the difference between data integrity and data fidelity?
Data integrity means data has not been "changed, destroyed, or lost in an unauthorized or accidental manner" (NIST). Data fidelity is looser: whether the record still means the same thing. A ticket can keep every comment yet show the wrong author and date, a fidelity failure. Validate both.
What is high fidelity data in a migration?
Migrated data that keeps its context as well as its values: original authors and dates, comment order, note visibility, parent links and working attachments. An agent can read a migrated ticket exactly as in the old system.
How long does data migration validation take?
There is no neutral public benchmark. It depends on objects in scope, sample size, defects found and re-runs needed. Count reconciliation is mostly scripted; the field diff and review of hard records take longest, and you repeat validation after the final delta sync.
What sample size should I use to validate a migration?
No public standard sets one for ticket data. Use a targeted sample of hard records chosen by ID, plus a random sample sized by probability: 299 random records give a 95% chance of catching a defect that affects 1% of records, and 59 records catch a 5% defect rate. Sample each object separately.
Changelog
- 31 Mar 2026: First published as "Data Fidelity in Migration: What It Means".
- 4 Nov 2026: Rewritten and moved from /blog/data-fidelity-in-migration. Now a validation guide for help desk, ITSM and CRM data: four methods, sample-size table, platform constraints table and a cutover checklist, with every fact sourced.
Written by
Ayush BothraFounding Member, MigrateX
Ayush writes about ITSM platforms, pricing and what it takes to move a service desk to a new platform without losing its history.


