Getting uncoordinated changes in research IT clusters under control
A large research institution runs lab, data and office IT with high-performance clusters. Around 1,000 changes a month create audit risks.
- Science
- IT-Service-Management
Key metric
Time spent coordinating changes
1,600960h/month
Monthly coordination effort for changes in research IT operations
At a glance
Starting point
The research institution runs several high-performance clusters, lab equipment, data platforms and office IT for more than 5,000 employees. Around 1,000 changes are planned every month: updates to cluster nodes, patches for lab software, configuration changes in storage and network, and adjustments to research applications. Requests come in through the ticket system, email and spreadsheets, often with incomplete details on affected systems, time windows and risks. Change managers and operations teams together spend more than 1,600 hours a month on clarifications, manual dependency checks and maintaining records in various documents.
- 1,000
- Changes per month
- 1,600 h
- Change effort per month
The challenge
Manual coordination of changes is reaching its limits. For every request, operations teams have to check which clusters, lab devices and data platforms are affected, whether maintenance windows collide and which research projects would be interrupted. Information is scattered across tickets, architecture overviews, runbooks and project plans. Critical dependencies often only become visible shortly before the maintenance window, leading to last-minute postponements, rollbacks and extended outages. At the same time, internal and external audits require traceable documentation of risks, approvals and the measures carried out. Producing these records takes extra time and remains patchy under pressure.
Typical symptoms in the company
Why this remained unsolved
The problem had been known for a long time but was accepted as the unavoidable complexity of running research operations. A classic IT project with a new tool seemed too costly, because many teams, systems and existing processes would be affected. Spreadsheets and ticket templates were revised several times, but could not capture the multitude of dependencies and special cases. Rule-based workflows failed because risks and impacts depend heavily on the context of each research project and change constantly. Coordination therefore remained a labor-intensive, error-prone process: too big for simple adjustments, too complex for rigid automation.
Our solution
The institution introduces an agentic workflow that centrally bundles change requests from the ticket system, email and spreadsheets, structures them and accompanies them through their entire lifecycle. An intelligent assistant reads new requests, maps them to the affected systems and services and fills in missing details by asking the requesters. For each change, the workflow creates a standardized risk and impact report that takes clusters, lab equipment and ongoing research projects into account. On this basis, maintenance windows are proposed, conflicts with other changes detected and approvals prepared. Once the change has been carried out, the workflow automatically generates a record of the measures, results and any rollbacks, which is filed directly in audit folders and knowledge bases.
Where generative AI comes in
Generative AI understands the free text in change requests, maps it to affected systems and services and writes structured risk and impact reports. It drafts proposals for maintenance windows, standardized approval documents and closing records, handles differing descriptions, project contexts and special cases, and turns them into consistent, auditable documents.
- Central intake component that collects change requests from the ticket system, email and spreadsheets, deduplicates them and converts them into a uniform format.
- Context assistant for risk analysis that evaluates architecture overviews and creates impact reports for clusters and lab equipment.
- Planning module that proposes maintenance windows, detects conflicts with other changes and generates decision documents for change boards.
- Documentation agent that creates standardized closing records and audit evidence from the measures taken, logs and feedback.
- Dashboard for change managers and operations teams that makes status, risks, approvals and open questions transparent across all changes.
How the agentic change workflow runs
Bundle intake
New change requests are read in from the ticket system, email and spreadsheets, classified automatically and mapped to the affected systems and services.
Analyze risks
The workflow fills in missing details, evaluates architecture and project contexts and creates a structured risk and impact report for each change.
Plan maintenance windows
Based on the impact reports, the system proposes suitable maintenance windows, detects conflicts with other changes and prepares decision documents.
Manage approvals
Change managers review the proposals, adjust them where needed and document approvals, conditions and escalations directly in the workflow.
Generate records
After the change has been carried out, the documentation agent creates a closing record with measures, results and rollbacks and files it in an auditable way.
Measured: after 6 months in production
Period: after 6 months in production.
| Metric | Value | Basis |
|---|---|---|
| Time spent coordinating changesKey metric | Reduced by around 40 percent | Time tracking before/aftermeasured |
| Share of unplanned rollbacksQuality measure | Down from about 8 to 4 percent | System measurementmeasured |
| Quality of change documentationQuality measure | Audit teams rate 90 percent of records as complete | User surveyestimated |
Cost-effectiveness
The solution depends heavily on the quality of the existing architecture and project documentation. Where documents are outdated or missing, manual checks remain necessary, and the agentic workflow needs clear responsibilities.
Integration and operations
Transferability
The agentic change workflow fits wherever many changes with complex dependencies have to be coordinated and classic tools reach their limits. It is especially valuable where free-text requests, scattered documentation and different teams come together. Generative preparation of context and records makes it possible to assess risks in a more structured way and serve audits better. Prerequisites are clear roles and a central ticket system or comparable sources.
A similar case in your organization? Let’s talk.
Briefly describe your process. We’ll get back to you with an initial assessment of whether the pattern can be transferred, and what it realistically takes.