Home · Business · IT, Data & Security · Automation & integrations
As of 13 August 2026, AI can only partly route documents for approval.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
1 hourto something you’d act on.
Cost, all in£0
Skill neededpower-user
Who has to check ita colleague
What the alternative costsA developer-built approval workflow is the alternative; no price is given in the supplied data.
If this goes wrong: a document is sent to the wrong person or left unapproved, causing delay, disclosure or an incorrect business decision.
What to actually do
Hand it to a person
The route this page recommends
A person who owns the outcome does this end to end, worth it when the failure is dear.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, power-user skill, and roughly 1 hour until you can act on the result.
How to actually do it
- Open the document repository and the automation or workflow platform that will receive documents and send approval requests.
- Gather the document types, available metadata fields, approval roles, value thresholds, sequence rules, fallback roles and documents that must not be routed automatically.
- Paste those details into the prompt, using roles or groups instead of names and excluding document contents unless they are needed for the design.
- Ask the model to produce the routing table, exception paths, access checklist and test plan, then resolve every item it marks as missing or ambiguous with the process owner.
- Build the workflow in the automation platform using the approved routing table, with least-privilege access, audit logging and a manual hold for excluded or incomplete documents.
- Run each test case with harmless sample documents, compare the destination, approver, notification and audit record with the expected result, and obtain a colleague's approval before activating the workflow.
Prompt
Act as a workflow architect for a UK organisation. Design a document approval-routing workflow from the information below, but do not claim to have connected to any system or sent any document. Organisation context: [brief description] Document source and destination systems: [for example, shared drive, document management system or email] Document types: [list] Fields available on each document: [list] Routing rules: [state which approver or approval group applies to each document type, department, value band or other condition] Approval sequence: [one approver, several in sequence, or several in parallel] Required response time: [state the internal target, if any] People or groups who may approve: [list roles or groups, not personal data] Fallback approver: [role or group] Documents that must never be routed automatically: [list] Sensitive information and access restrictions: [describe] Produce: 1. A plain-English routing table covering every rule. 2. The exact trigger, conditions, actions and rejection paths for the workflow. 3. A list of missing inputs and ambiguous rules. Do not invent them. 4. An exception process for missing fields, unavailable approvers, duplicate documents, rejected documents and failed notifications. 5. A least-privilege access and audit-log checklist. 6. A test plan with sample cases for every route, including a case that should not be routed automatically. 7. A deployment checklist that requires a human to approve permissions and test results before activation. Keep the design platform-neutral unless a platform is named. Mark every assumption clearly.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot know whether the routing rules reflect your current organisation chart, delegation limits or internal policy.
- AI cannot safely infer an approver when a document is urgent, disputed, incomplete or outside the stated rules.
- AI cannot grant itself legitimate access to your repositories, identities or confidential documents.
- AI cannot prove that every permission, exception path and audit record works correctly in your live systems.
- AI does not take responsibility for a document sent to the wrong person or left without approval.
What caps this at PARTLY: private data access, judgement under ambiguity and verification cost.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 1 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 1 |
| Total | 6 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can AI route documents for approval?
- Partly. AI can design and help implement rules that route documents by type, department or value, but you still need to provide the rules, connect the systems and test the live workflow.
- Can AI decide who should approve a document?
- It can apply explicit routing rules to an identified approver or group. It should not guess when the rules are incomplete, the approver is unavailable or the document falls outside normal cases.
- Can AI set up an approval workflow?
- It can draft the workflow and generate implementation instructions, and some agent-building tools can help build the surrounding application. A person still has to configure permissions, connect the document systems, test failures and approve activation.
- Is it safe to let AI route confidential documents?
- Only with controlled access, clear exclusions, audit logging and tests using representative sample documents. Your organisation remains responsible for disclosure if the workflow sends a confidential document to the wrong recipient.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.