An internal service desk built and owned by your company beats a Jira Service Management renewal mainly for large IT and employee-services operations (well past a hundred agents) whose approval chains and asset records are central to how the business runs, and only if engineers will maintain it. Below that size, Atlassian's per-agent pricing is hard to undercut, so renew. The test: agents × per-agent list price × 36 months, set against engineers × build months × $7,000 plus annual running costs.
Atlassian charges per agent each month. Its Jira Service Management pricing page shows Premium at $55.28 per agent per month with monthly billing (list price checked November 21, 2026; check your contract). Read that figure with two caveats. It is the monthly-billing price at the page's default team size, a size the page doesn't state, and per-agent pricing is tiered by team size, so a desk of 150 or 400 agents may see a different rate. Paying annually is advertised as "save up to 17%".
List prices checked November 21, 2026. Check your contract; negotiated prices differ. Per-agent figures are monthly-billing prices at the page's default team size and are tiered by team size.
Two usage meters run alongside headcount. Beyond its 1,000 included conversations, the virtual service agent is listed from $0.30 per assisted conversation, and Assets objects past the free allowance start at $0.02 per object per month. Both scale with the whole company, not with IT: every employee can raise requests, and every laptop, SaaS license and server you register is an object. Renewal quotes typically rise when HR, facilities or finance desks join the instance, when a single must-have feature forces Standard up to Premium, when the asset register grows, and when the desk relies on more Marketplace apps.
For most IT teams under about a hundred agents, staying put is correct. Jira Service Management sits next to the engineering work your developers already track in Jira, and its prices at that scale leave little room for owned software to recover a build.
That split fits IT especially well. The agent count is often inflated not by engineers but by HR, facilities and IT operations staff handling routine access, equipment, onboarding and offboarding tickets. Those flows already hinge on your identity provider and HR system, which is where custom code earns its keep.
Staying costs agents × per-agent price × 36. Leaving costs the build (engineers × months × $7,000), Atlassian fees through the build plus a single month of parallel running, and a running cost from that parallel month until month 36. Break-even is the earliest month from which leaving's cumulative spend stays at or under staying's. The rows use Premium at $55.28 per agent per month (list price, check your contract), the monthly-billing price at the page's default team size. Since pricing is tiered by team size, swap in your contracted per-agent rate first. Running cost is our assumption: a fraction of one engineer at $7,000 a month plus hosting and monitoring.
Premium list price checked November 21, 2026 (list price, check your contract), on monthly billing at the page's default team size and tiered by team size. Excludes virtual service agent overage, Assets objects and Marketplace apps. Build size and running cost are assumptions.
Look first at 50 agents. Atlassian costs about $2,764 a month there, while a quarter of an engineer plus hosting runs $2,250, so leaving never catches up and you should renew. At 150 agents the crossing comes in month 29, close enough that an annual-billing rate could wipe it out. At 400 agents it arrives in month 20. Under the split, Jira licenses that engineering keeps belong on both sides. Enter your tiered rate, Assets count and app spend in the 3-year build vs buy TCO calculator. The table locates a crossing on stated assumptions; it does not forecast savings.
List every service project with its request types, queues, SLA goals, automation rules and approval steps, and write down the actual approver for each. Add monthly volume per request type and the systems it touches: identity provider, HR system, device management, purchasing. Approvals hide the real logic; "manager approval" might mean an HR reporting line, a cost-center owner or one named security reviewer. You're done when request types are ranked by volume and every approval rule is written in plain English and confirmed by whoever owns it.
Atlassian's backup manager, under Settings > System > Backup manager, creates a zip holding activeobjects.xml, entities.xml and a data folder of attachments. Backups with attachments need 48 hours between runs, and Atlassian suggests total data under 25 GB with attachments around 50–100 GB at most. Opsgenie-powered alerts and on-call, and Assets for JSM, are not in that backup. A CSV export with all fields gives each comment its own column, but Jira Cloud has no native bulk download of attachment files. Teams on Atlassian Backup and Restore should know its backups stay 30 days in Atlassian storage, can't be downloaded to yours, and leave out Marketplace app data. You're done when you have checked by hand whether request types, SLA history and customer accounts survived, and Assets plus each app's data has its own tested extraction route.
Tables for requests, request types, comments, attachments, approval steps (approver, decision, timestamp), SLA timers and an asset register with typed attributes and relationships: laptop to employee, application to owner, server to service. Join requests to assets by foreign key, so "every ticket about this laptop" is a query instead of a search. Store each Jira issue key on the migrated request so old links resolve. You're done when the backup loads cleanly and an auditor's usual question, who approved this access and when, comes straight out of the new tables.
Take the two or three request types with the most traffic, usually access, equipment and onboarding. Approval rules become code with unit tests, reading reporting lines from the HR system rather than a stale copy. Release the portal form, agent queue and approval step behind a feature flag to one department. You're done when that department raises, approves and closes real requests without touching Jira Service Management.
Put SSO in front of the portal, then connect the identity provider for access grants, the HR system for joiners and leavers, and device management for hardware. Under the split, the owned desk creates and updates Jira issues through the Jira Service Management REST API: base path /rest/servicedeskapi, basic auth for scripts or OAuth 2.0 (3LO) for integrations, and paging via start, limit, size and isLastPage. Respect the rate limits: apps have had a points-based hourly quota since March 2, 2026; bursts allow 100 requests per second for GET and POST and 50 for PUT and DELETE; a single issue takes 20 writes per 2 seconds and 100 per 30 seconds. API token traffic sits outside the new quotas, though burst limits still apply. Status changes flow back through webhooks that accept JQL filters; dynamic app webhooks lapse 30 days after creation or refresh, and failed deliveries get up to five retries, so schedule refreshes and a nightly reconciliation job. You're done after a week of mirrored traffic with zero throttling errors and no lost status updates.
Send one department's new requests to the owned desk while everyone else stays on Jira Service Management, then add departments in turn. Measure first-response time, approval cycle time, SLA breaches and reopen rate against that department's own history. You're done when each department has spent a full month on the new desk with numbers its lead accepts.
Redirect old portal links, take a last backup and CSV export, and set service projects to read-only. Atlassian's cloud licensing page says your site remains accessible until 15 days after the current subscription period ends, and paid-site data is deleted 60 days after cancellation. Finish archiving well inside that. You're done when the final export is in your storage with recorded checksums and old request keys open archived records.
On these projects our AI pod teams hand coding assistants the dull work, such as turning backup XML into table mappings, generating approval-rule tests from past requests, and drafting export and reconciliation scripts. People still review every change, and it ships only once tests pass.
A tech lead owns the schema and the cutover. One to three senior full-stack engineers build the portal, queues and approval engine. A platform engineer handles identity and device-management integrations (hire DevOps engineers); a data engineer helps when years of attachments and a big asset register are involved; and part-time QA covers approval tests and the department pilots, which our QA and testing services also do. Many teams begin with a pod of 3 to 5 senior engineers (a tech lead plus engineers), and a full pod is about seven. If you're only moving the employee portal and approvals under the split, a tech lead with one or two engineers is a reasonable start.
Ryz engineers on your team start from $7,000 per month (our published range is $7,000–$15,000 per engineer per month, mostly mid-level to senior), quoted per team. If your own numbers favor leaving, scope your team with us.
At low agent counts, rarely. Multiply agents by your per-agent rate by 36, and compare that with engineers times build months times $7,000, plus Atlassian fees during the build and one parallel month, plus running cost. With Premium at list price and our assumptions, 50 agents never break even, 150 cross in month 29 and 400 in month 20. Tiered pricing at your team size can move all three.
The backup manager produces a zip of XML files and attachments, with 48 hours required between backups that include attachments, and a full-field CSV export covers flat issue data. Assets for JSM and the alerting and on-call features sit outside the Jira backup, Marketplace app data is absent from Atlassian's backups, and CSV export cannot bulk-download attachments, so each needs its own route.
The count of request types in active use, how tangled the approval rules are, how large the asset register is, how many Marketplace apps you depend on, and whether engineering stays on Jira. Moving department by department keeps each stage contained, and the week-one inventory is what produces a believable plan.
Engineers multiplied by months multiplied by the monthly rate, which starts at $7,000 per engineer, for the build. Then add a running cost you control, normally part of a maintainer's time plus hosting and monitoring, and the Atlassian fees that continue until the final department moves.
Yes, and many IT teams should consider it. Jira remains home for engineering projects and incident escalation, while the employee request portal, approval engine and asset register move to owned software that opens Jira issues over the API when a developer is needed. Developers keep their tool, and agent licenses for the business desks are no longer needed.
Questions we didn't answer? Email info@ryzlabs.com.
Thanks — your message has been sent. We’ll get back to you soon.
Something went wrong while sending your message. Please try again or email info@ryzlabs.com.