A document approval workflow: review it, approve it for publication, and know which version is live
SynaptaGrid8 min read
In short
A document approval workflow is the path a policy, procedure or other controlled document takes from a draft to a published version: a reviewer checks it for accuracy and clarity, the owner of the subject decides whether it is correct to publish, and the document ends published, returned to its author or rejected. Most document problems come from confusing review with approval and from not knowing which version is current. A reviewer who also approves checks their own judgement, and a draft that circulates by email can be mistaken for the final one. A good workflow separates the two roles, lets the reviewer send a draft back without ending it, and marks the status and a published flag on a record so everyone can see what is live. 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 old and new wording of a changed section, classifies the change and writes a short summary for the approver, and the workflow writes both onto the document record before the reviewer looks at it. We ran it three times, once published, once returned to the author and once rejected by the owner, and checked the document records afterwards.
What a document approval workflow is
Some documents have to be right before people rely on them: a safety procedure, an expenses policy, a customer-facing terms page. A workflow writes down who reads the draft, who has the authority to say it can go out, and what the possible answers are. Without one, drafts move by email, comments are lost, and a published copy can sit beside three older ones.
The workflow does not edit the document or judge it. People do. It makes sure the review and the approval both happen, in order, and that every document ends in a status anyone can look up, including whether it is published.
The version we built
The steps to include, and who owns each
- Record the document
- When a draft is submitted, a record is created with the title, a status of draft in review and a published flag that is off. Every later step reads from and writes to that record.
- Summarize the change
- This is where an AI tool helps. It reads the old and the new wording of the changed section, classifies the change as a change of wording, process, safety or something else, and writes a short summary of what changed. The next step writes both onto the document record, so the reviewer and the owner see what changed before they read the text. The tool does not approve or reject anything.
- Review
- The reviewer reads the draft for accuracy, clarity and fit with other documents. A reviewer gives an opinion on the text; they do not decide that the company stands behind it.
- Send back without rejecting
- The reviewer can return the draft to its author with changes. That ends this round with a clear status. The author submits a revised draft, which is a new request, so each round has its own record.
- Owner approval
- The owner of the subject, such as the head of the function the policy covers, decides that the document is correct to publish. This is an authority question, separate from the reviewer’s reading, so it is a separate step.
- Reject
- The owner can refuse a document the reviewer passed, for example because the policy has changed. The record is marked rejected.
- Mark the result on the record
- Each ending sets the status and the published flag. Only an owner-approved document is marked published, so the flag answers the question which version is live.
- Put the document where people read it
- Publishing to an intranet, a help site or a document store usually happens in another system. Decide where an approved document goes and add that as its own step.
Where approvals and checks belong
Separate review from approval. A review improves the text, and an approval accepts responsibility for it. When one person does both, nobody has checked the decision, and the document carries one person’s view as the company’s.
Name each role, such as a reviewer from the team and the function owner, rather than a person. If some documents need a second approver, for example legal for anything customer-facing, add that on your own policy; we did not build it.
The old way and the new way
The table compares documents handled by email and shared folders with the same documents run as a workflow.
- What changed in the draft
- The old way: Compare two versions by eye
- The new way: An AI tool classifies the change and summarizes it first
- Where the draft is
- The old way: In an inbox or a shared folder
- The new way: At a named step, with an owner
- Who may say it is final
- The old way: Whoever sends the last email
- The new way: The owner, at a recorded approval step
- Which version is live
- The old way: Ask the author
- The new way: The record’s published flag says so
- A draft that needs changes
- The old way: Comments in several threads
- The new way: Sent back with a clear status; the next round is a new request
- Finding a past decision
- The old way: Search old messages
- The new way: Open the record and read its status
- Judgement
- The old way: Still the reviewer’s and the owner’s
- The new way: Still theirs: whether the text is right, whether to publish
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 document, 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 read a change, an input that takes the old and new wording as text, and an output with a change type from a fixed list and a short summary. 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 reviewer passed the draft and the owner approved it. The workflow created the document record with the status Draft in review and the published flag off, then wrote the AI result onto it, then updated the same record to Published with the flag on, and the run finished with the published result. In the second run the reviewer sent the draft back: the run skipped the owner, marked the record Returned to the author and ended. In the third run the reviewer passed the draft and the owner rejected it, so the record ended as Rejected.
We checked the three records in Data afterwards. The published one showed Published with the flag on. The returned one showed Returned to the author, and the rejected one showed Rejected, both with the flag off. All three carried the AI result: the change type process, and a one sentence summary saying the spare parts storage moved from the main warehouse to a cage by the production line and that counting went from yearly to quarterly under the supervisor. The three summaries were word for word the same, because the same text went in each time and the tool is set to give the same answer for the same input. The summary says what changed and does not say who else the change affects, which is what a reviewer would still want to ask. 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. Earlier runs of the workflow also failed at a record write, so we rebuilt the record type and pointed the actions at it, and we capped the tool’s summary below the length a text field on the record accepts. The version described here is the one whose three outcomes we checked.
We chose a separate reviewer and owner, a send-back route that does not reject, and a published flag on the record because those are the places document control 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 changed section in plain text, so we did not test a long section, a change to a table or a change that looks like wording and alters the meaning, and we did not compare its change type with a person’s. One test account held every human step, so the runs show the sequence and the branching, not hand-offs between different people. We did not attach a document, compare versions or publish anywhere: the workflow works on a record about the document, and the published flag is a stand-in for putting it live. We did not build a second approver, a link from a returned draft to its revision, or a review date after which the document must be checked again. We did not test reminders when a step waits too long or notifications to the author. 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.
- Document control
- Quality management standards, such as the ISO standard for quality management systems, expect documents to be approved before use and the current version to be identifiable. The published flag and the recorded approval support that.
- 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 summarizes the change and the reviewer and the owner decide.
- Source: EU AI Act (Regulation 2024/1689), Article 14: human oversight (opens in a new tab)
- Segregation of duties
- The person who reviews a document and the person who approves it should differ, so approval is a second look and not the same opinion repeated.
- Source: NIST glossary: separation of duty (opens in a new tab)
- Documented approval
- Auditors may ask who approved a policy and when. A step that records the decision and a record that carries the status are that evidence.
- Periodic review
- Controlled documents are usually reviewed again after a set period. This guide covers approval, not the review date, 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 document stands. You can start from the version above, change the steps to match your own document 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 the document itself must be attached and versioned, whether approved documents must reach a publishing system, and who should be told when a draft waits.
Frequently asked questions
- What steps should a document approval workflow have?
- Record the document, a review with a way to send it back, an owner’s approval for publication, a way to reject, and a status and published flag written on the record.
- What is the difference between reviewing and approving a document?
- A review checks the text for accuracy and clarity. An approval accepts responsibility for the document and agrees that it can be published. They are best done by different people.
- How do I know which version of a document is current?
- Keep a record for each submitted version and mark only the approved one as published, so the flag answers the question.
- Can AI summarize what changed in a document?
- It can read the old and new wording of a section, classify the change and write a short summary. In our test run it called a change to how spare parts are stored and counted a process change and summarized it in one sentence. It does not judge whether the change is right: the reviewer and the owner still decide.
- Can a workflow update the document record for me?
- Yes. In our test run the workflow created the document record when the draft was submitted and set its status and published flag at each ending. It does not publish the document itself unless you connect it to where documents are published.
SynaptaGrid
Build your document approval steps
Start with the steps above in a test workspace and run one document through each outcome before you use it for real.