Published on  September 30, 2026 / 5 min read

How to Build an Approval Workflow

How to Build an Approval Workflow

Plenty of internal processes need someone to say yes before anything happens. Expense requests, content publishing, leave requests, purchase orders. Most teams run these on email and memory, which works until the volume grows or the person who remembered everything goes on holiday.

An approval workflow is that process written down and enforced by software.

What an approval workflow is made of

A request moves through one or more approval steps before it is finalized. Four pieces do the work:

A request submitted by a user. A status field, usually Pending, Approved and Rejected. A list of approvers or managers. Actions that change the status based on their decision.

That structure is what lets a team track decisions and see the state of a process without asking anyone. Everything else is variation on it.

Where it gets used

Expense approvals, purchase requests, leave requests and document review cover most of what teams build first. They share a shape: a form, a queue, a decision, a notification.

The interesting variations start when the routing gets conditional. Requests above a threshold need a second approver. Requests from a particular department go to a particular manager. Anything that sat untouched for three days escalates. None of that is hard to build, but it is worth deciding before you start rather than bolting on later.

What you need before you start

A connected data source, and a clear answer to three questions.

Who can submit? Who approves, and is it one person, a role, or a chain? What happens when nobody responds?

The third question is the one people skip, and it is the one that determines whether the workflow survives contact with reality. A request with no timeout sits in Pending forever, and the requester goes back to sending emails, which is what you were trying to stop.

Building it in Jet Admin

Connect your database first. Jet Admin supports over 200 integrations and reads your tables and fields automatically, so the interface has something to build on.

Then create the submission form. Keep it short. Every field you add is a field someone has to fill in before they can ask for something, and long forms push people back to messaging you directly.

Next, configure what happens on approval and rejection. In Jet Admin these are actions that update the request record based on the approver's decision. This is also where notifications belong: tell the approver when a request arrives, and tell the requester when it resolves.

Then set permissions, so approvers see the queue and requesters see their own requests.

Finally, test with real requests before anyone depends on it. Submit one, approve it, reject another, and try leaving one untouched to see what happens. The unattended case is important to test because the process needs an explicit response when a decision is delayed.

Two things that go wrong

The workflow gets more approval steps than the decision deserves. Three approvers on a $50 expense is a process designed by someone who has never submitted one. Each step adds delay and diffuses responsibility, and past two steps most of what you are buying is the appearance of control.

The status field stops matching reality. This happens when people can approve things outside the system, by replying to a message or telling someone in person. If a decision can be made off the record, the record becomes decorative. Either the workflow is the only route, or it is documentation of a process that happens somewhere else.

Model decisions separately from their consequences

An approval is a recorded decision about a particular version of a request. Sending a payment, publishing content, or provisioning access is a subsequent operation with its own result. Keeping those states separate prevents an “Approved” label from being mistaken for proof that the downstream work succeeded.

For a purchase-request example, retain a stable request identifier, requester, submitted details, required approver, current revision, and decision history. A decision record should identify who decided, when, and which version they approved. If the amount, supplier, or scope changes materially, apply the organization’s rule for whether the request needs another decision.

The exact approval threshold and escalation period belong to the organization’s policy, not to a universal template. Write them down before configuring routing. A useful initial workflow can be small while still making exception handling explicit.

State or eventMeaningNext step
SubmittedA complete request is ready for a decision.Assign the authorized approver.
In reviewA particular approver owns the decision.Approve, reject, or request clarification.
ApprovedThe recorded request revision was approved.Execute the authorized downstream operation.
Execution failedThe decision exists but the operation did not complete.Investigate or retry under the defined policy.
CompletedThe intended result has been verified.Record the outcome and notify the relevant person.

Build the requester and approver journeys

Use a short submission form with the fields needed to make the decision. Show the requester their own requests and the current reason for delay or rejection. Give the approver the original details, supporting evidence, and the permitted actions in one place rather than a button without context.

Use the configured identity and permission model to enforce who may decide. Hiding an approval button does not establish authorization. Test an attempted decision through an alternate view or request path and verify that a requester cannot approve their own request unless the documented policy allows it.

Jet Admin’s automation overview describes workflow building and triggers. Validate the actual connector, update operation, and notification setup in the deployed app. A proposed approval process still needs its own routing, data model, and permission tests.

Handle duplicate decisions, timeouts, and delegation

Two approvers may open the same request at once. Before applying a decision, confirm that the request is still at the expected revision and state. A stale screen should refresh or report the conflict rather than overwrite a later decision without explanation.

A timeout during downstream execution should not automatically create a second payment, publication, or account. Use a supported idempotency mechanism or reconcile the destination before retrying an uncertain operation. Keep the authorization record distinct from the operation’s attempt history so support can explain what happened.

Specify how an approver can delegate, what happens during an absence, and when a pending request escalates. Escalation should route to a known responsibility, not silently treat lack of response as consent. Test both the notification and the actual reassignment.

Prove the workflow before broad rollout

Test a valid submission, missing evidence, rejection, a changed request, an unauthorized decision, two simultaneous decisions, and a failed external operation. Retain the expected outcome for each case and inspect the resulting records, not only the notification text.

Agree on the measures that matter for the process: pending age, decision turnaround, reopened requests, execution failures, and requester clarification. A faster approval is useful only when the required evidence and authority are preserved. Use the pilot to remove unnecessary steps while retaining the controls the decision actually needs.

Assign ownership for policy updates, permissions, failed runs, and connector credentials. When the policy changes, version the configuration and decide how existing pending requests are handled. A workflow needs an operating owner as well as an initial builder.

What you get

Instead of tracking requests through inboxes, the whole process sits in one place: who asked, what they asked for, who decided, when, and what the decision was. That history is usually the reason the workflow gets built in the first place, and it is worth more than the time saved on any individual request.

What is Jet Admin

Jet Admin is the AI app builder for turning your existing data into real business software — no code required. Describe what you need, and Jet's AI Builder instantly generates the app, connected to your live database or API, with role-based access and audit logs already built in.

Teams use it to build everything from admin panels and internal tools to CRMs, customer portals, and inventory systems — on the data they already have, with no per-seat fees and no migration required.

Get started free→