# Processes

> The work that has to move between people - requests, approvals, hand-offs - described once and then run by the platform, with the automation that removes the steps nobody should be doing by hand

Source: https://aipril.dev/docs/processes

## Every company already has these

Somebody requests a day off. A discount needs a manager's sign-off. A new supplier has to be
checked before the first invoice. A customer complaint has to reach the person who can answer it,
and somebody has to notice if it does not.

None of this is exotic, and none of it is optional. What varies is only whether it is written down
anywhere.

## Where it usually lives

In email, in chat, and in the heads of the three people who know how it really works.

The request arrives in someone's inbox. They forward it to whoever they think approves that kind of
thing. That person is away, so it sits. Two weeks later somebody asks what happened to it and there
is no answer, because nothing recorded that it was ever raised - only that it was mentioned.

The alternative most companies try is a classical BPM system, and it fails in the opposite
direction: a separate product, drawn by an analyst in a notation nobody else reads, integrated with
the systems where the data actually lives through a project that takes a quarter. It becomes the
thing you avoid changing, so it stops matching how the work is really done - and people go back to
email.

## What is different here

**You describe it, and it gets built.** What a request contains, who fills it in, who approves it,
what may follow what - said in words. The assistant produces the working process, including the
rejection path, the timeout and the status nobody remembers to add until March. The specialist who
used to be required for this is the single biggest reason companies still run approvals out of a
mailbox.

**It runs where the data is.** A route resolves who approves from the real
[org structure](https://aipril.dev/docs/ai-organisation/people.md), a form's fields hold real references and not typed text, the tasks
are readable by [analytics](https://aipril.dev/docs/analytics.md) without an export, and a row in a
[spreadsheet](https://aipril.dev/docs/spreadsheets.md) can start one.

**Substitution is not a special case.** Somebody being away is a normal condition of an
organisation, not an exception the process discovers at the worst moment.

**People answer from where they already are.** A reply in Telegram or by email lands on the task as
a comment - the participants who never open the system still take part in it.

**And most of it should not involve a person at all.** Everything that is a hand-off rather than a
decision - the reminder, the copy, the notification, the nightly check, the update to the other
system - runs by itself.

## Where to go next

- [What you can do](https://aipril.dev/docs/processes/what-you-can-do.md) - the process from the inside: requests and
  their fields, routes and approvals, who owns a task and what happens when they are away.
- [Forms and records](https://aipril.dev/docs/processes/forms.md) - what a request is made of: field types, the checks
  that hold before it is saved, the three ways to work the same records, and where a form can open.
- [Automation and orchestration](https://aipril.dev/docs/processes/automation.md) - what runs without anybody: triggers,
  the nodes a scenario is built from, installing your own, and long-running scenarios that survive
  a restart.
- [AI in processes](https://aipril.dev/docs/processes/ai.md) - describing a process in words and getting one, and the
  agent as a step inside a running scenario.
- [Under the hood](https://aipril.dev/docs/processes/under-the-hood.md) - how the two engines work, what is guaranteed
  and what is bounded on purpose.