Automation Audit: How to Decide What to Automate and What to Keep Human

Before you automate anything, the real question is usually "where does my time actually go?" An automation audit is the short piece of work you do before you build anything: you list what is done by hand, decide what a machine can safely do, and decide what a person should keep.

This guide gives you an original 10-area audit that Ai Spark uses as a working method, a scorecard, a worked example, and a worksheet you can copy into a spreadsheet today. It also says plainly what we have tested ourselves and what we have not.

What an automation audit is (and is not)

An automation audit is a structured look at how work moves through your business: how an enquiry arrives, who touches it, which apps hold the information, and where things get stuck, repeated or forgotten. The output is a short ranked list of tasks, each with a decision.

It is not a software purchase, a promise of savings, or a plan to remove people. If the idea of automation is still new to you, read What is AI automation? first. If you already know you want a tool, the audit also makes the choice easier, because you will know what the tool must do. Our n8n vs Zapier vs Make comparison is the next step after this one.

How to read the evidence labels in this guide

Automation content online is full of confident numbers. We use four labels so you can see what each claim rests on.

The audit in one picture

The audit has six moves. The last one matters most: it sends you back to look again, because a workflow you built last month is a task you can audit this month.

Step 1: List what you do for one week

Keep a note open on your phone or desk. Each time you do something you have done before, write it down with a rough time in minutes. Ask your team to do the same. Typical entries look like this:

• Copied a new website enquiry into a spreadsheet (3 minutes)
• Sent the price list to a WhatsApp enquiry (2 minutes)
• Reminded a client about tomorrow's visit (2 minutes)
• Downloaded bank transactions for the accountant (20 minutes, weekly)

Step 2: Check ten areas

A list of chores shows you the obvious work. The ten areas below show you the hidden problems, like a lead that arrives somewhere nobody looks. This is Ai Spark's own method, built from our own work on lead capture and follow-up workflows. It is a working checklist, not an industry standard. For each area you get what to check, why it matters, what could be automated, whether a person should stay involved, and what evidence to collect.

1. Repetitive tasks

• What to check: anything done the same way more than once a week, with minutes per time.
• Why it matters: small chores are easy to forget when you estimate from memory.
• Automation opportunity: copying, pasting, sending the same message, filing.
• Keep human? The judgement parts, such as choosing what to say to a difficult client.
• Evidence needed: your own one-week list, not a guess.

2. Lead capture

• What to check: every way an enquiry can reach you (form, email, WhatsApp, calls, social messages) and where each one ends up.
• Why it matters: an enquiry that lands where nobody looks cannot be answered.
• Automation opportunity: send each channel into one list, and save the lead first before anything else happens.
• Keep human? Judging whether the lead is a good fit.
• Evidence needed: trace your last ten enquiries from arrival to first reply, with dates.

3. Data entry

• What to check: where the same detail is typed more than once.
• Why it matters: retyping is where typos and mismatched records come from.
• Automation opportunity: pass fields from the form to the sheet or CRM.
• Keep human? Checking unusual or incomplete entries.
• Evidence needed: a count of how many places each detail is typed.

4. Follow-up

• What to check: what happens after first contact, who decides when to chase, and what makes follow-up stop.
• Why it matters: follow-up that depends on memory can slip when you are busy.
• Automation opportunity: reminders to you, prepared drafts, scheduled nudges.
• Keep human? Wording that makes promises, and the decision to stop chasing.
• Evidence needed: a sample of follow-ups that were due versus sent, and your written stop rules.

5. CRM updates

• What to check: whether records and pipeline stages change when real things happen, and who moves them.
• Why it matters: a pipeline you cannot trust is not worth keeping. Our guide to building a sales pipeline explains stages, and What is a CRM? covers the basics.
• Automation opportunity: move a record when a clear event happens, such as a booked call.
• Keep human? Stages that need judgement, such as "proposal accepted".
• Evidence needed: compare ten records against what actually happened.

6. Notifications

• What to check: who needs to know what, how fast, and which alerts get ignored.
• Why it matters: too many alerts get muted, and then the important one is missed.
• Automation opportunity: one clear alert when a new lead arrives or a workflow fails.
• Keep human? Sensitive or urgent matters, which should reach a person directly.
• Evidence needed: a list of current alerts and which ones people actually act on.

7. Scheduling

• What to check: how many messages it takes to book one appointment, and how reminders and rescheduling are handled.
• Why it matters: back-and-forth booking can be a source of delay.
• Automation opportunity: a booking link and automatic reminders.
• Keep human? Exceptions, special requests and important clients.
• Evidence needed: a sample count of messages per booking.

8. Reporting

• What to check: which numbers you collect by hand, where each comes from, and who reads the report.
• Why it matters: a report nobody reads is wasted time, and a wrong number is worse than none.
• Automation opportunity: a weekly summary built from the sheet or CRM.
• Keep human? Deciding what the numbers mean.
• Evidence needed: spot-check a few figures against the raw records. In our own n8n test, the time written to the sheet did not match local time until we understood a time zone setting, so check dates first.

9. Human approval points

• What to check: where a person must see something before it goes out or changes money or records.
• Why it matters: if approval is hard, people skip it. If it is missing, a mistake goes straight to a customer.
• Automation opportunity: make approval easy by preparing drafts and review queues, not sending.
• Keep human? Yes, by definition.
• Evidence needed: a written list of approval points and who owns each one.

10. Error and risk points

• What to check: what happens when a field is empty, an email is wrong, a contact is duplicated, an app is down, or someone has asked not to be contacted.
• Why it matters: a workflow can fail without anyone noticing unless you plan for failure.
• Automation opportunity: failure alerts, save-first design, duplicate checks.
• Keep human? Decisions about customer data and privacy.
• Evidence needed: a test run using deliberately bad data, with the result written down.

Step 3: Score each task

Give every task from your list 1, 2 or 3 points on each of five questions. This is a simple prioritising tool of our own design. The cut-off numbers are a starting point you can adjust, not a scientific rule.

Step 4: Give each task one outcome

Add the five scores (maximum 15), then apply the human gate.

The human gate. A task is gated if it moves money, issues a refund, deletes data, gives legal, medical or safety advice, handles a complaint, or sends a message in your name that makes a promise. A gated task is never fully automatic, however high it scores.

A worked example

ILLUSTRATIVE EXAMPLE — NOT A CLIENT CASE STUDY. "Green Row Garden Care" is a made-up two-crew gardening company. Ai Spark has not audited or built anything for it. The scores are invented to show the method. No savings, revenue or conversion figures are claimed.

Simple future workflow for the first automation:

Trigger: a new form submission arrives
Step 1: save the enquiry to the Leads sheet first
Step 2: prepare a draft reply from a template
Step 3: alert the owner that a draft is waiting
Step 4: the owner reads, edits and sends the reply
Step 5: update the sheet status
If the email address is missing: alert the owner, do not continue
If the contact has asked to stop: no automated message

Step 5 of the audit is "build small and test with bad data". This example is not a tested system. We have tested a similar pattern in n8n, which is described next.

What we have actually tested at AI Spark

This section lists only what we have checked ourselves. It is short on purpose. We have no client audit case studies to report, and we do not present any.

We have not tested AI-written replies, WhatsApp automation or the full audit method on a paying client. The scorecard and the ten areas are our own working method, and we have not measured how well it predicts results.

Where a person has to stay involved

A few outside sources are worth knowing before you automate anything that talks to customers. We summarise them in our own words, and none of these replaces advice for your situation.

• AI text can be wrong. Anthropic's documentation says that AI models, including Claude, can produce text that is factually incorrect, and that its suggested techniques reduce this but do not remove it, so important information should be checked. OFFICIAL SOURCE ONLY. This is why our example prepares drafts for a person to approve.
• A draft is a built-in approval point. The n8n Gmail node supports creating drafts as well as other operations, so a workflow can prepare a message that waits for you. OFFICIAL SOURCE ONLY, though we did use the draft operation in our own test.
• Plan for failure. n8n lets you set an error workflow that runs when another workflow fails, starting with an Error Trigger node, and a Stop and Error node lets you fail a run on purpose under conditions you choose. OFFICIAL SOURCE ONLY. We have not set up an error workflow ourselves yet.

Marketing emails and opt-outs. If a workflow sends commercial email, the US Federal Trade Commission's CAN-SPAM guidance lists requirements such as honest subject lines, a postal address, a clear way to opt out and honouring opt-out requests within 10 business days. It also says that hiring someone else to send your email does not remove your responsibility. OFFICIAL SOURCE ONLY. This is general information, not legal advice, and rules differ by country. We have only reviewed the US guidance here, so check the rules that apply to you.

The Automation Audit Worksheet

Copy the layout below into a spreadsheet. One row per task. Everything you type in is a rough estimate, and so are the totals.

To paste a ready-made header row into a blank sheet, copy this line and split it into columns:

Task, Frequency per week, Minutes per occurrence, Current tool, Who performs it, Error risk, Automation potential, Human approval required, Priority, Notes

When you review a workflow after a few weeks, compare your first estimate with what you actually observe. Do not assume the original number was right.

What to do after the audit

1. Pick one task marked Automate that you can describe in one sentence, such as "when a form is submitted, save the lead to a sheet and prepare a draft reply".
2. Write the workflow on paper, one line per step, including what happens when something is missing.
3. Choose the tool after you know what it must do. Our comparison and the CRM automation workflows guide can help.
4. Build it small and test it with bad data, as in our n8n lead follow-up tutorial.
5. Check the results for the first weeks and compare them with your estimates before you automate the next task. n8n keeps a list of executions and separates manual test runs from production runs, which makes this easier to review. OFFICIAL SOURCE ONLY.

Want help running an audit?

If you would rather not do this alone, Ai Spark can walk through your lead capture, follow-up and CRM process with you, map it on paper first, and tell you what we think is worth automating and what is not. We make no promise of a particular saving, because that depends on your business. Get in touch and tell us how enquiries reach you today.

Frequently asked questions

Sources

• n8n Docs: Handle errors gracefully (checked 29 Oct 2026)
• n8n Docs: Gmail node, Draft operations (checked 29 Oct 2026)
• n8n Docs: Executions (checked 29 Oct 2026)
• Claude Platform Docs: Reduce hallucinations (checked 29 Oct 2026)
• US Federal Trade Commission: CAN-SPAM Act, A Compliance Guide for Business (checked 29 Oct 2026)

Last reviewed 29 October 2026. Our own tests are described in the section "What we have actually tested at AI Spark".