A change request workflow: an AI risk check, a board decision, and a record of what was approved
SynaptaGrid8 min read
In short
A change request workflow is the path a proposed change takes from the person who wants it to a recorded decision: someone assesses the risk and the impact, a change board decides whether it goes ahead, and the request ends approved, sent back for changes or rejected. The slow part is usually the assessment. Someone has to read the request, work out who is affected and how it could go wrong, and write that up before the board can even start. A good workflow does that assessment first and puts it where the board will read it, keeps the decision with people, and writes the result onto a record. We built a small version in our own test workspace in which an AI tool, written for this job and used as an action in a normal step, rates the risk and summarizes the impact, and the workflow writes both onto the change record before the board decides. We ran it three times, once approved, once sent back and once rejected, and checked the records afterwards.
What a change request workflow is
A change is anything that alters how the business runs: moving a server, changing a supplier, switching a system on or off. Most changes go well. The ones that do not tend to share a cause, which is that nobody wrote down who would be affected or how to undo the change before it was approved. A workflow writes down who assesses the change, who decides, and what the possible answers are.
The workflow does not judge the change. The board does. It makes sure every request is assessed before it is decided, that the assessment is in front of the board when it decides, and that every change ends in a status anyone can look up.
The version we built
The steps to include, and who owns each
- Record the request
- When a change is proposed, a record is created with its title, a status of submitted and an approved flag that is off. Every later step reads from and writes to that record.
- Assess the risk and impact
- This is where an AI tool helps. It reads the request and returns a risk rating, a short impact summary and the questions the board should ask. It saves the board from doing the first read of every request, and it does not decide anything.
- Put the assessment where the board reads it
- The rating and the summary are written onto the change record, so the board sees them with the request instead of in a separate message.
- Board decision
- The change board, or the person with authority to approve changes, decides. This is a judgement about the business, so it stays with a person.
- Send back without rejecting
- A change with a fixable gap, such as a missing rollback plan, can be sent back for changes. That ends this round with a clear status, and a revised request is a new record.
- Reject
- The board can refuse a change. The record is marked rejected, so the decision can be found later.
- Carry the change out
- Scheduling and doing the change usually happens in other systems or by other teams. Decide who does it and add that as its own step.
Where approvals and checks belong
Separate the assessment from the decision. An assessment describes what could go wrong, and a decision accepts that risk. When the same person writes the risk and approves it, nobody has checked the decision. In this version the assessment comes from an AI tool and the decision from a person, and the person can read the assessment against the original request.
Name the board by role, such as the change board and the owner of the affected system, rather than by person. If high-risk changes need an extra approver, or if low-risk changes should skip the board, add that on your own policy; we did not build it.
The old way and the new way
The table compares change requests handled by email and a shared document with the same requests run as a workflow.
- Who writes the risk assessment
- The old way: The requester, or nobody
- The new way: An AI tool drafts it; the board reads it
- Where the assessment is
- The old way: In an email or a separate document
- The new way: On the change record, next to the request
- A request with a gap
- The old way: Rejected, or approved anyway
- The new way: Sent back with a clear status
- Finding a past decision
- The old way: Search old messages
- The new way: Open the record and read its status
- Time to the first read
- The old way: Whenever someone finds time
- The new way: Done as soon as the request arrives
- Judgement
- The old way: Still the board’s
- The new way: Still the board’s: whether to accept the risk
How we decided
We built the version in the diagram in our own test workspace in SynaptaGrid Automation, with a small record type in SynaptaGrid Data for the change request. The AI part is a custom tool we wrote for the job: a prompt that says how to rate risk, an input that takes the request text, and an output with a fixed shape of a risk rating, an impact summary and up to three questions. In the workflow it is an action in an ordinary step, set up the same way as any action that calls another system. The next step reads its output and writes the rating and the summary onto the record.
We ran it three times in November with the same request text: moving a shared file server used by all staff to cloud storage over one weekend, with no written rollback plan. In every run the tool rated the change high and wrote a three-sentence summary onto the record, and the board step then waited for a decision. In the first run the board approved, and the record ended as Approved with the flag on. In the second run the board sent it back, and the record ended as Sent back for changes. In the third run the board rejected it, and the record ended as Rejected. Both flags were off in the last two.
We checked the three records in Data afterwards. Each showed the high rating and the summary, plus the right status. All three runs worked first time.
The summary was not perfect. It said the finance team’s month-end reports would run on the new system, where the request only said they run from the server on Monday morning. The difference is small, but it shows why the board reads the summary against the request and does not treat it as fact. The choice of steps is ours and a judgement; it is not a standard.
Six limits, so this guide does not claim more than the runs showed. One test account held the board decision, so the runs show the sequence and the branching, not hand-offs between different people. The tool saw one request text, so we did not test how its rating changes for small, safe changes, and we did not compare its ratings with a person’s. The workflow does not act on the rating: a high-risk change went to the same board as any other would, and routing by risk is not built. We did not put the three questions the tool returns on the record. We did not build the work of carrying out the change, a rollback check or reminders when a decision waits too long. And the template carries one test account as the board and the actions tied to our test record type, so to reuse it you would change the step to a role and point the actions at your own record type, which we did not do.
Standards and accepted practice behind this
These are the practices the guide rests on. We align with them. This page does not claim a certification.
- Change enablement
- The ITIL framework treats a change as something to assess for risk and decide through a defined authority, with the amount of control matched to the risk. This guide covers one route through that: assess, decide, record.
- Documented approval
- Auditors and security frameworks expect to see who approved a change and when. A step that records the decision and a record that carries the status are that evidence.
- A rollback plan
- Most change policies expect a way to undo a change before it is approved. Our test request had none, which is what the tool flagged, and a send-back route is where a missing plan is handled.
- Human oversight of AI
- Where AI helps with a decision, accepted practice is to keep a person accountable for it and to let them see the AI’s inputs and outputs. Here the tool drafts and a person decides.
- Source: EU AI Act (Regulation 2024/1689), Article 14: human oversight (opens in a new tab)
- BPMN
- The standard notation for drawing a process. The diagram above is a plain flowchart, and a BPMN diagram of the same steps would show the same decisions.
- Source: OMG: BPMN 2.0 specification (opens in a new tab)
Where SynaptaGrid fits
SynaptaGrid Automation lets you write the steps as a workflow, use an AI tool you wrote as an action in a step, pass its result to the next step, create and update records in SynaptaGrid Data, and see where every change stands. You can start from the version above, change the steps and the tool’s prompt to match your own change policy, and keep the result as a template.
Check the limits above before you rely on it: whether the board is assigned by role rather than by person, whether the rating should change who decides, whether the work of making the change must be tracked, and who should be told when a request waits.
Frequently asked questions
- What steps should a change request workflow have?
- Record the request, assess its risk and impact, a board decision with a way to send it back, a way to reject, and a status written on the record.
- Can AI assess the risk of a change?
- It can draft the assessment: a rating, a short summary and questions for the board. In our test run it rated a risky change high and summarized it in three sentences, with one small slip. A person still reads it against the request and decides.
- Should AI decide which changes need the board?
- That is a policy choice, and we did not build it. In our version every change went to the board whatever its rating, which is the cautious way to start.
- Can a workflow write the assessment onto the change record?
- Yes. In our test run the next step took the rating and the summary from the AI tool’s result and wrote them onto the record, so the board saw them next to the request.
SynaptaGrid
Build your change request steps
Start with the steps above in a test workspace and run one change through each outcome before you use it for real.