What a Service Business Should Automate First

Most advice on this question hands you a list. Invoicing, lead follow-up, onboarding, reporting, scheduling — all true, all automatable, and none of it tells you which one to do on Monday.

The list is the wrong artefact. You already know what is painful. What you need is an ordering rule, and the rule is this: automate the step that is done most often, by the most expensive person, with the least judgement.

Three factors, and all three have to be present. Score every candidate on volume × salary × judgement and the ranking falls out of the arithmetic rather than out of whoever complained loudest this week.

Why all three, and not one

Each factor on its own points somewhere wrong.

Volume alone sends you to the highest-frequency task in the business, which is usually something trivial that already takes seconds. Automating a ten-second task done fifty times a week returns almost nothing and costs a build.

Salary alone sends you to whatever the most senior person does, which is usually the work you least want a machine touching. A partner’s judgement on a difficult file is the product, not the overhead.

Low judgement alone sends you to whatever is most mechanical, which is frequently something done once a quarter by a junior. Correct, mechanical, and worth almost nothing to fix.

The intersection is where the money is: a repetitive, rules-based step currently being performed by someone whose time is expensive. In a New Zealand service business that is very often the person who is also the owner.

Score your own candidates

Write down every step somebody complains about. Then score each one out of five on three axes and multiply.

A step scoring five, five, five is your first automation. A step scoring five, five, one — high volume, expensive person, high judgement — is not an automation candidate at all. It is the thing you hired that person for.

Two rules keep the exercise honest. Score judgement by asking whether two competent people would produce the same output from the same inputs; if they would not, it is judgement. And score salary by who actually does the work today, not by who is supposed to.

Check what your existing tools already do

Before you scope a build, find out whether you are about to pay for something you are already licensed for. This is the single most common way automation money is wasted.

If you run Xero, it already sends invoice reminders on a schedule you configure, and it already handles repeating invoice templates and invoice reminders without anything else in front of it. If your CRM is HubSpot, sequences and deal-stage automation are already in the product. Almost every scheduling tool sends its own reminders.

The build only makes sense for the gaps between tools, and for the steps no tool owns. That is a much shorter list than the one you started with, which is the point.

Then, and only then, choose the tool

Once you know the step, the platform question becomes easy — and the three candidates bill in genuinely different ways, which matters more than their feature lists.

Zapier meters tasks: roughly, each action a Zap performs. Make meters operations, counted per module run, so a workflow that loops over line items consumes them quickly. n8n can be self-hosted, which changes the billing question into an infrastructure question.

That difference decides more real cases than any feature comparison. A low-volume workflow across popular apps is cheapest on Zapier. A high-volume workflow that iterates over rows burns operations on Make and is better self-hosted. A workflow touching data you would rather not send offshore is an n8n question before it is anything else.

We wrote a decision procedure that resolves to one tool rather than a comparison that defers, and a longer comparison of n8n, Zapier and Make for New Zealand businesses if you want the detail behind it.

What to automate second — and why the answer changes

Here is the part almost every list-shaped article misses. Your first automation changes the ranking of everything else.

It changes it in three ways.

It moves work rather than removing it. Automating quote delivery does not delete the quoting process; it pushes the bottleneck to whatever happens next, usually follow-up. The second-ranked candidate on your original list is frequently no longer second by the time the first one ships.

It creates a new failure mode. An automated step fails silently in a way a person does not. Somebody now has to notice. n8n’s error handling exists for exactly this, and wiring an error workflow is usually the correct second piece of work rather than a second automation.

It creates data you did not have. Once a step runs through a workflow, it is timestamped. That is often the first honest measurement of how long anything actually takes — which is precisely the input the scoring table wanted and you had to estimate the first time.

So re-score after the first one lands. Do not queue up six builds against a ranking derived before you had any of this.

Four things not to automate first

Stop conditions are as useful as the rule, and they are missing from almost everything written on this question.

A process nobody has written down. Automating an undocumented process encodes one person’s version of it, including the parts they get wrong. Write it down first; sometimes that alone fixes it.

A process about to change. If you are mid-way through changing practice management systems, the integration you build this month is disposable.

Anything where being wrong is expensive and silent. Payment runs, client communications that commit you to something, anything touching a filing deadline. These can be automated — but not first, and not without an approval step. Which brings us to the last one.

Approval itself. If your process has a human sign-off, the automation’s job is to make that sign-off fast and well-informed, not to remove it. Under New Zealand’s GST rules the information you must hold and provide is taxable supply information, and getting that wrong at volume is worse than getting it wrong once. Likewise, if the workflow moves personal information, where and how it is stored is governed by the Privacy Act’s information privacy principles — a question worth settling before the build, not after.

A worked example

A New Zealand building-services firm with a handful of office staff.

Candidates: chasing signed variations, re-keying supplier invoices, weekly WIP reporting, and new-client onboarding.

• WIP reporting — monthly, mechanical, done by the office manager. Volume one, salary two, absence of judgement five. Score ten.
• Chasing variations — several times daily, but each chase is a judgement call about the relationship. Volume five, salary four, absence of judgement two. Score forty.
• Re-keying supplier invoices — many times daily, purely mechanical, currently done by the director because nobody else has Xero access. Volume five, salary five, absence of judgement five. Score one hundred and twenty-five.
• Onboarding — weekly, mixed judgement, admin staff. Volume three, salary two, absence of judgement three. Score eighteen.

Invoice re-keying wins, and it is not close. Notice that it won partly for a reason that is not really about the task: the director does it because of an access decision. Sometimes the first fix is not an automation at all — it is giving somebody a login. Check for that before you build.

If it survives that check, the build is the right call. And after it ships, chasing variations gets re-scored, because the director now has time and the bottleneck has moved.

Where to go from there

The order matters more than the tooling. Get the first one right and the second becomes obvious; get it wrong and you will spend the budget proving that automation does not work for your business, when what it proved is that the ranking was wrong.

If you want the wider context, our guide to business process automation in New Zealand covers what these builds involve. If the step you have picked turns out to need judgement rather than rules, AI agents for New Zealand businesses covers where that line sits. And if you have decided to build, who to talk to about building n8n workflows covers the three routes and what each one costs you later.

One more thing worth reading before you scope anything: business.govt.nz on capacity and demand. Automation is a capacity decision wearing a software costume, and the framing there is a better starting point than most automation content.

Common questions