Workflow automation: workflows that run themselves.
Workflow automation is the practice of taking a multi-step business process — the hand-offs, checks, and notifications a person currently does by hand — and running it as a machine-executed sequence with defined triggers, states, and exceptions. Aldenebai builds these workflows with n8n as the orchestrator, GitHub Actions as the workers, and Microsoft Teams for the notifications, so the process runs on its own and a human only steps in at the approval gates.
Scale model of the system
What does babysitting a process actually cost a team?
Every operations team has a process that works only because somebody watches it. A ticket lands and a person checks whether it is clear enough to act on. A job finishes and a person notices, or does not, and tells the next person. Something fails overnight and nobody knows until the morning stand-up. None of this is hard. All of it is constant, and it lands on the people you hired for their judgment.
The cost is not the ten minutes per step. It is the context switch, the item that sat in a queue for a day because its owner was in meetings, and the retry that never happened. Business process automation with AI takes the polling, the checking, the routing, and the chasing off people and leaves them the decisions.
At the marina-software client the ticket-to-PR pipeline described below removed hours of setup per ticket and made ticket-to-PR turnaround about 2× faster. Rounded from production data at a US marina-software company, 2025–2026; definitions and measurement notes are on each case study.
Repetitive workflows can be automated with human approval gates where they belong
From audit to a workflow running on its own
Map the process as it runs today
On the Automation Audit we walk the workflow live: every trigger, hand-off, check, and the person each step waits on. You get the written Automation Map within 48 hours, ranked by what each item takes and returns.
Place the approval gates
Together we decide which steps a human must approve and which can run unattended. Each gate becomes an explicit state in the workflow, so the machine cannot skip it or guess its way past it.
Build the orchestration
n8n takes the intake, triage, queue, and dispatcher. The heavy work runs as GitHub Actions jobs or scripts. Zapier goes where a light connection between two tools is all the step needs.
Wire notifications and the watchdog
Teams pings go to the person who owns the decision, carrying the state and the reason. A scheduled watchdog sweeps the queue and flags anything that has stalled past its limit.
Test it before it touches anything
The workflow runs against real inputs with live writes held back until it behaves. Gates, caps, and retry loops are tested the way software is tested. Built by a QA-automation engineer, so this step is not optional.
Run it on your infrastructure
Fixed scope and fixed price, agreed before any build. The workflow runs under your access controls with an audit trail, and your team keeps the exceptions.
Which workflow patterns show up in almost every company?
Ticket-to-PR orchestration
A Linear ticket labelled ai-task is picked up, triaged, queued, coded by an AI agent in a GitHub Action, opened as a pull request for a human to approve, compiled, and moved to QA once merged. The full build is documented on the AI dev agent page.
Queue and dispatcher with caps
One item at a time, mapped from project to the right repositories and branches, with a per-repo cap on open pull requests. Labels are the state, so anyone can read the queue from the tool they already use.
Retry loop with a ceiling
A red desktop build triggers an automatic retry in which the AI fixes the failure, for at most two passes. If it is still red, the item is labelled for a human. The same loop is shared with the autonomous bug fixer.
Watchdog and notifications
Every 30 minutes a watchdog flags anything stuck for more than 60 minutes. When a pull request is ready, the tester gets a Teams ping and the ticket moves to the right state without anyone chasing it.
Legacy terminal automation
Accuterm terminal processes that used to need a person typing through screens, automated end to end. The old system stays; the keystrokes go to the machine.
Light glue where it fits
Not every step needs an orchestrator. Where two SaaS tools need a simple hand-off, a Zapier zap does the job and stays readable by the team that owns it. The audit says which is which.
Why n8n holds the state and GitHub Actions does the work
Any n8n automation expert will tell you the orchestrator should never do the heavy lifting. n8n holds the queue, the state, the notifications, and the watchdog. Workers are stateless jobs that clone, run, and report back. When a worker dies, the orchestrator knows. When the orchestrator restarts, the state is in the labels, not in memory.
Who should hire a workflow automation consultant?
Ops leads, CTOs, and founders whose team runs the same sequence every day or every week, where the rules are clear but a person still has to babysit it. Engineering teams with a ticket queue that waits on setup. Back-office teams chasing sign-offs in chat. Anyone keeping a legacy terminal alive by hand.
It is a poor fit when every case is a fresh judgment call with no repeatable pattern. Those stay human, and the audit will say so rather than sell you a workflow that needs constant rescue.
The person who maps the workflow is the person who builds it, tests it, and runs it. No hand-offs, no agency chain. If the workflow should also keep other systems in sync, see integrations; if it should keep a CRM clean, see HubSpot and CRM automation.
Questions ops leads ask about workflow automation
What does a workflow automation consultant actually deliver?
A running workflow, not a slide deck. The Automation Audit is a free 30-minute call, and you leave with a written Automation Map within 48 hours: ranked opportunities, what each takes, what each returns. If it is a fit, you get a fixed-scope, fixed-price proposal for the first workflow, then the built and tested system running on your infrastructure with your team on the exceptions.
Why n8n instead of Zapier, or the other way round?
n8n is the orchestrator when a workflow needs a queue, state, retries, notifications and a watchdog, which is most workflows that matter. Zapier is used where a light connection between two SaaS tools is all the step needs. Many companies end up with both, and the audit says which goes where.
How do human approval gates work in an automated workflow?
The workflow stops at a defined state and notifies the person who owns that decision in Teams. Nothing merges, writes or sends until they act. In the ticket-to-PR pipeline a human approves every pull request before anything merges, and anything unclear is labelled ai-needs-human or ai-needs-detail instead of being guessed at.
Can business process automation with AI handle steps that need judgment?
It handles the bounded judgment: is this item fresh, is it clear enough to act on, is it a duplicate, what is the first draft. Open-ended judgment stays with the human at the gate. Where the machine tries to fix its own failure, the retry loop has a ceiling of two passes, and then a person takes over.
What happens when a step gets stuck or fails?
A watchdog runs every 30 minutes and flags anything stuck for more than 60 minutes. Known failure modes trigger a capped automatic retry; anything else gets a Teams ping to the owner with the current state and the reason. Nothing fails silently, and the state lives in labels and statuses your team can already see.
Related pages
Find the workflow you are still babysitting.
The Automation Audit is free: 30 minutes, and you leave with a written map of which processes to automate first, where the approval gates go, what each takes and what each returns. Yours to keep either way.