---
id: KB-CL-010
url: https://app.codecontract.io/help/collaboration/where-to-start-automating
idioma: en
categoria: colaboracion
subcategoria: automatico
audiencia: usuario
nivel: basico
actualizado: 2026-08-13
tambienEn: [es]
relacionados: [KB-CL-002, KB-CL-005]
citadoPor: [KB-CL-013, KB-CL-019]
---

# Where to start automating

_The first thing worth leaving to run itself, and why that one._

**Responde a:** what should i automate first · recommended first automation · where to start with automation · automating without complications

The short answer is chasing: reminders at days 3, 7 and 12. It gives back the most time, can break the least, and needs no explaining.

## Why that one and not another

| Criterion | Automatic chasing |
| --- | --- |
| How much time does it return? | A lot: it is the most repetitive work there is |
| What if it fails? | Little: one reminder too many or too few |
| Does anything need deciding? | No: it is a clear rule with no exceptions |
| Is the effect visible? | Within two weeks, in delivery |

> [!IMPORTANT]
> The four answers together are what make it first. An automation that returns a lot but could break something serious is not first; one that breaks nothing but returns nothing is not either.

## The order for the rest

1. **Automatic chasing** — Spaced reminders. Week one.
2. **Expiry warnings** — With margins based on who renews each item. Week two.
3. **One of the dull rules** — Open a task when something has been stuck too long.
4. **And only then, the rest** — Once the first three have run a month with nobody touching them.

> [!WARNING]
> Do not start with what decides. Approving, rejecting or blocking someone are your decisions, and automating them before trusting the rest is the fastest way to have to backtrack.

## How to know it worked

If two weeks later you have forgotten it exists, it is working. An automation you have to watch daily is solving nothing: it is moving the work from one place to another.

> [!NOTE]
> Every automatic action consumes just as if a person did it. A rule firing a thousand times spends a thousand: check the breakdown in the first month, not the sixth.

**What if my case is different?**

The criterion still applies: high return, little harm if it fails, no decisions.

**Can I automate several things at once?**

You can, and then you will not know which caused what. One at a time.

**Can it be switched off?**

Yes, at any moment.

## Ejemplos

**A company wants to automate document approval to move faster.**

- Starts with automatic chasing
- Leaves approval with a person

→ Recovers most of the time without delegating any decision it answers for.

**Automation starts with the company's most complex process.**

- Starts with what repeats weekly and needs no judgement
- Measures how much time it consumes today before touching it
- Checks at four weeks whether the saving is real

→ The first automation pays for itself in a month and gives a basis for choosing the second.

**Something that happens three times a year is automated.**

- Counts the repetition before investing

→ Effort goes where it pays.

**A process that still changes weekly is automated.**

- Stabilises the process before automating it

→ The automation is not rebuilt monthly.

**It is automated by copying the paper circuit, dead steps included.**

- Removes steps that existed only because of paper

→ The new process is shorter than the one it replaces.

**What decides is automated instead of what chases.**

- Automates requesting, chasing and warning

→ The repetitive disappears and the judgement stays.

**Nobody measures whether the automation helped.**

- Compares against the time it used to take

→ The next decision is taken with data.

**Five things are automated at once and none works well.**

- Builds one, verifies it and moves to the next

→ Each is understood and can be fixed.