In short
An IT access request workflow is the path a request for access to a system takes from the employee to a recorded decision: the manager confirms the person needs the system for their work, security decides how much access is appropriate, and the request ends granted in full, granted with limits or denied. Most access problems come from a missing middle option and from access nobody can account for later. Security cannot say yes, but read only, so it chooses between granting more than the person needs and refusing someone who has a real need, and nobody can tell afterwards who approved what. A good workflow gives security three choices, keeps the business question and the risk question apart, and writes the decision 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, reads the request and suggests full or limited access with a reason, and the workflow writes the suggestion onto the access record before the manager decides. The suggestion is advice for security, not a decision. We ran it three times, once with full access, once with limited access and once denied by the manager, and checked the request records afterwards.
What an IT access request workflow is
An access request asks for permission to use a system: a finance tool, a customer database, a code repository. A workflow writes down who approves it, who decides the level of access, and what the possible answers are. Without one, requests travel by chat, access is granted by whoever answers first, and nobody can list who has what.
The workflow does not grant access in the system itself. It makes sure the right people decide, that the decision includes how much access, not only whether, and that every request ends in a record that can be found when someone asks why a person has access.
The version we built
The steps to include, and who owns each
- Record the request
- When a request arrives, a record is created with the system requested, a status of requested and an access level of pending. Every later step reads from and writes to that record.
- Suggest a level
- This is where an AI tool helps. It reads the request, suggests limited or full access and gives a short reason, such as that the need is temporary. The next step writes the suggestion onto the access record so security sees it with the request. It is advice only: security still decides the level, and the manager still decides whether the person needs the system.
- Manager approval
- The manager confirms that the person needs the system for their work. The manager knows the job, which security cannot see.
- Security review of the level
- Security decides how much access is appropriate. This is a different question from whether the person needs the system, so it is a different step with a different owner.
- A middle answer
- Besides granting and denying, security can grant limited access, for example read only or one part of the system. Without this option security either grants more than the person needs or refuses a real need.
- Two places to stop
- A request can be stopped by the manager, who does not support it, or by security, who deny it. Both lead to the same ending: the record is marked denied with no access.
- Mark the result on the record
- Each ending sets the status and the access level: full, limited or none. The workflow does this itself, so the record shows what was decided.
- Grant the access in the system
- Creating the account or the permission happens in the system itself, or in an identity tool. Decide who does it and add that as its own step.
Where approvals and checks belong
Put each decision with the person who has the information. The manager knows whether the work needs the system, and security knows what the access allows and what it exposes. Splitting them stops a manager from granting access they cannot assess and security from refusing without knowing the job.
Name each approver by role, such as the requester’s manager and the security team, rather than by person. If a particular system needs the system owner’s approval too, add that on your own policy; we did not build it.
The old way and the new way
The table compares access requests handled by chat and tickets with the same requests run as a workflow.
- First read of a request
- The old way: Security reads it cold
- The new way: An AI tool suggests a level and a reason first
- Where the request is
- The old way: In a chat or a ticket queue
- The new way: At a named step, with an owner
- Answers available
- The old way: Yes or no
- The new way: Full access, limited access or denial
- Who approved the level
- The old way: Often unclear
- The new way: A separate security step that is recorded
- What the person was given
- The old way: Check the system and guess
- The new way: The record states the access level
- Finding a request later
- The old way: Search chat history
- The new way: Open the record and read its status
- Judgement
- The old way: Still the manager’s and security’s
- The new way: Still theirs: whether it is needed, how much is safe
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 request, and ran it three times in November. The AI part is a custom tool we wrote for the job: a prompt that says how to review a access request, an input that takes the request text, and an output with a suggested level of limited or full access and a short reason. In the workflow it is an action in an ordinary step, set up like any action that calls another system, and the next step writes its result onto the record. In the first run the manager approved and security granted full access. The workflow created the request record with the status Requested and the level Pending, then updated the same record to Access granted with the level Full, and the run finished with the granted result. In the second run security granted limited access, so the record ended as Access granted with the level Limited. In the third run the manager rejected the request: the run skipped security, marked the record Access denied with the level None and ended with a denied result.
We checked the three records in Data afterwards. They showed Access granted with Full, Access granted with Limited, and Access denied with None. All three carried the AI suggestion: limited access, with the reason that the need was temporary and tied to one task. In the first run security granted full access even though the tool had suggested limited, which is the design working as intended: the tool advises and security decides. The note was nearly identical in all three runs, and it was written before the manager decided, so even the rejected request carries one. The AI tool did not work first time: our earlier attempts to call it failed until we named the model in full in the tool’s settings, and our first version did not use its output in a later step. One run also failed at the AI call with an error we did not investigate; a fresh run with the same input passed. The version described here is the one whose three outcomes we checked.
One path was not run on its own. A denial by security leads to the same denied step the manager’s rejection uses, and we ran that step through the manager’s rejection only. We chose a middle answer for security and a separate manager approval because those are the places access handling usually goes wrong. 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. The tool saw one request, so we did not test how its suggestion changes for a different system or a longer need, and we did not run a manager approval followed by a security denial. One test account held every human step, so the runs show the sequence and the branching, not hand-offs between different people. Nothing was granted anywhere: the workflow does not create accounts or permissions, and we did not connect it to an identity or ticketing system, so the access level on the record is a statement of the decision, not a change in a real system. We did not build an approval by the owner of the system, an expiry date for access, or a periodic review. We did not test reminders when a step waits too long or notifications to the requester. And the template carries one test account as every approver and the actions tied to our test record type, so to reuse it you would change each step to a role and point the actions and the tool at your own, 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.
- Least privilege
- A person should hold only the access the work needs. The limited-access answer is how a request ends with less than was asked for.
- Source: NIST glossary: least privilege (opens in a new tab)
- 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 input and output. Here the tool only suggests a level, and security makes the decision.
- Source: EU AI Act (Regulation 2024/1689), Article 14: human oversight (opens in a new tab)
- Segregation of duties
- The person who asks, the person who approves the need and the person who sets the level should not all be the same.
- Source: NIST glossary: separation of duty (opens in a new tab)
- Documented approval
- Security frameworks and audits expect to see who approved access and when. A recorded step and a record that carries the level are that evidence.
- Periodic access review
- Many frameworks expect access to be reviewed from time to time. This guide covers the request, not the review, so plan that separately.
- 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, assign each decision to a person, 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 as part of the run, and see where every request stands. You can start from the version above, change the steps to match your own access policy, and keep the result as a template.
Check the limits above before you rely on it: whether each step is assigned to a role rather than a person, whether granting must reach an identity or ticketing system, whether some systems need their owner’s approval, and who should be told when a request waits.
Frequently asked questions
- What steps should an access request workflow have?
- Record the request, a manager approval, a security review of the access level with a limited option, a way to deny from either step, and a status and level written on the record.
- Why add limited access as an answer?
- Because the choice is often not full access or nothing. A limited answer lets security give what the work needs and no more, which is the least privilege principle.
- Who should approve access to a system?
- The requester’s manager for the business need and security for the level, decided by role rather than by name, plus the system’s owner if your policy requires it.
- Can AI decide who gets access?
- It should not. In our test run an AI tool suggested limited access with a reason and the workflow wrote it on the record, and security still chose the level, in one run granting full access against the suggestion. The tool advises and a person decides.
- Can a workflow update the access record for me?
- Yes. In our test run the workflow created the request record when the request arrived and set its status and access level at each ending. It does not change access in the system itself unless you connect it to one.
SynaptaGrid
Build your access request steps
Start with the steps above in a test workspace and run one request through each outcome before you use it for real.