End-to-end approval workflow: 7 checks before launch

An end-to-end approval workflow starts when someone outside your company submits a form and ends when the decision is written back where your team keeps its records. Between those two points sit a Power Pages site, a Power Automate flow, a Dataverse table and a group of approvers, and each handoff can fail in its own way. This checklist walks through seven places to look before anyone presses the button for real.

What makes the design hold up?

Five questions cover most of it. Who can start the request? Who can see which records? Who decides, and in what order? What happens if nobody answers? Where is the result stored? Each one maps to a setting you will meet below: web roles, table permissions, the approval type, run limits and a Dataverse row. Dynamics 365 Sales and Customer Service keep their data in Dataverse, the same layer the site and the flow work on, so the answer to the last question is often a table you already have.

If you can answer all five on one page, the build is mostly assembly. If you cannot, no amount of clicking in the designer will close the gap.

Seven checks for an end-to-end approval workflow

1. Secure the flow with a web role

Attaching a flow to the site generates a unique URL, and the web roles you assign decide who may call it. Pick the narrowest role that works. A signed-in supplier role is a safer default than leaving anonymous visitors in, because the flow can write to your records. Review the role list whenever the flow changes what it does.

2. Match parameter names and types

The names in the request body have to match the trigger’s inputs exactly, and the supported types are Text, Boolean, File and Number. Flows configured with secure inputs cannot receive parameters from the site. A trigger with no parameters still expects an empty payload. Test once with a deliberately wrong name, so you recognize the failure when it shows up later.

3. Choose the approval type on purpose

Everyone must approve, first to respond, custom responses and sequential approval each stop the flow at a different moment. Write down your company’s real sign-off rule first, then pick the type that copies it. Sequential approval, for instance, requests decisions one at a time and in order, and each approver must answer before the next one receives the request.

4. Scope table permissions by account

Account access gives a role the records linked to the signed-in user’s account, so colleagues at one supplier share a view while another supplier sees nothing of theirs. A permission with no web role does nothing, which is worth testing with two real accounts. Contact access, by comparison, ties records to the individual user, which fits requests that belong to one person.

5. Plan for the 30 day limit

Pending approvals count toward a 30 day run limit, and then the pending step times out. Add a reminder and an escalation path early, and write a final status to the record so the requester never stares at an open request with no future. The run duration is counted from the run’s start time, so the clock starts the moment the visitor submits, not when the approver first opens the email. Decide on reminders before launch, because after the first stuck request is too late.

6. Group multi-row updates

When one decision touches several Dataverse rows, use the connector’s changeset request. It treats the operations as one unit: if any step fails, the steps that already worked are undone, so you never end up with a half-applied decision. The changeset request sits in the Dataverse connector next to actions such as Add a new row and Update a row.

7. Register the flow in every environment

Moving site components to a new environment does not carry the flow registration along. Unregistered flows return a forbidden error when the site calls them, so put the register step on the release checklist for test and production. The error appears only when the site calls the flow, so a build that looks complete can still fail on its first real request.

Before you launch the workflow

Run one request through the whole chain with real accounts, from the form to the final record, and watch each handoff. Include one rejection and one request nobody answers, beyond the happy path, and compare the final record with what the requester saw. A dry run exposes missing roles and mismatched names within minutes, while production exposes them in front of customers. Keep this checklist with the solution too. When someone changes the flow later, the same seven checks tell them what to retest.

If the checks raise questions about portal design itself, this article on B2B and B2C portals is a useful companion to read next.