B2B SaaS support tickets with resolution threads and linked bugs

Summary

Nine years of tickets from the support desk of one mid-market SaaS vendor, each with the full customer and agent thread, internal notes, tags, priority, reply times and the closing resolution. About 14% link to an engineering issue, so a model can follow a problem from first report to shipped fix. Customer names, emails, phone numbers, account ids and pasted credentials are replaced with consistent placeholders.

Collected to order. We collect it from US companies for your order. Scope, volume and timing are confirmed in writing before you sign.

Pricing on request

Request scope and pricing

At a glance

Planned scope. Volume, sources and timing are confirmed in writing before you sign.

Records: 512Ksupport tickets
Size: 4.3 GB
Tokens, estimated: 620M
Years: 2017 to 2026
Firms: 1

Owner profile: US B2B SaaS vendor of workflow software for mid-market operations teams, 150 to 300 staff
Industry: Software
Data type: Support and service
Headcount: 150 to 300 staff
Region: United States
Languages: English
Formats: JSONL
Updated: 2026-10-14

Sources

System	Kind	Records
Zendesk	Support tickets	512K tickets, 2.9M comments
Jira	Projects and wiki	71K linked bug and escalation issues
Confluence	Projects and wiki	3,400 support runbooks

Good for

training support agents to draft first replies and pick the right macro
evaluating ticket triage, priority and routing against the agent's final call
learning when a support issue is really a product bug and how it gets escalated
scoring generated resolution notes against the note the agent wrote

Fields

Field	Type	Description
ticket_id	string	Stable de-identified ticket id
created_at	timestamp	When the customer opened the ticket
channel	enum	email, web form, chat or phone note
priority	enum	low, normal, high or urgent
tags	array<string>	Agent-applied tags for product area and issue type
comments	array<object>	Ordered thread with author role (customer, agent, agent_internal) and body
linked_issue	string	De-identified engineering issue key, if escalated
status	enum	solved, closed or pending at export
resolution_summary	string	The agent's closing note
first_reply_minutes	int	Minutes from creation to first public reply

Example record

{
  "ticket_id": "T-0418822",
  "created_at": "2024-03-11T14:07:52Z",
  "channel": "email",
  "priority": "high",
  "tags": [
    "sso",
    "login_failure"
  ],
  "comments": [
    {
      "role": "customer",
      "body": "Since this morning none of our dispatch leads can sign in with SSO. They get 'SAML response not valid' right after the identity provider redirect. Roughly 40 users are locked out at [SITE_3]."
    },
    {
      "role": "agent_internal",
      "body": "Their IdP rotated its signing cert last night and we still have the old one on file. Escalating so we stop pinning the cert. Admin contact is [PERSON_31] at [EMAIL_31]."
    }
  ],
  "linked_issue": "ENG-48213",
  "status": "solved",
  "resolution_summary": "Customer admin uploaded the new IdP signing certificate under Security settings and logins were restored. Bug filed to refresh from the metadata URL instead of a pinned cert.",
  "first_reply_minutes": 22
}

One record, de-identified. Placeholders such as [PERSON_12] and [COMPANY_4] stay the same across the whole dataset, so threads still read correctly.

Licenses available

Exclusive
One buyer. Nobody else can license it during the exclusive term, usually 2 years.
Non-exclusive
Licensed to more than one buyer.
First look
One buyer gets 2 to 4 weeks to make an offer before anyone else sees the dataset.

What ships with it

Datasheet: what it is, what it is good for, verified size, sources, years, format and limits.
Provenance file: the owner shown by code, staff notice and consent status, exclusions, mapped to the EU AI Act Article 53 training-data summary and California AB 2013.
Manifest with a hash for every file.
QC report and de-identification report.
Your own copy, with unique canary records.

How it is de-identified

Default exclusions removed, files flattened to text, identifiers found in three layers and replaced with consistent placeholders, and the result measured against a 99% bar for each identifier type in each source.

The full method