What Should You Automate First? A Scoring Framework for SMEs

Why does the first automation choice matter so much?

Because it sets the internal politics of everything that follows. A first project that returns visible hours in eight weeks makes the second project easy to approve. A first project that runs long, produces a debatable result and irritates the team makes automation a dirty word for two years.

The instinct is to automate the thing that causes the most pain. That is usually the wrong choice, because painful processes tend to be painful precisely because they involve judgement, exceptions and negotiation — the hardest things to automate reliably.

How do you score a process for automation?

Score each candidate out of 5 on these five factors. Anything scoring 20 or above is a strong first candidate. Below 15, park it for later.

Do this with the people who actually do the work, not just the leadership team. The scores that come back from the front line are consistently more accurate — particularly on rule clarity, where managers routinely overestimate how standardised a process really is.

Which processes usually win?

Across NZ and Australian SMEs, the same handful of processes score highly again and again, because they share the same profile: high volume, structured input, and an output someone checks anyway.

1. Inbound enquiry triage. Classify, summarise and route incoming emails or form submissions to the right person, with a draft response attached. High frequency, low risk, immediately visible to management.
2. Document data extraction. Pull structured fields out of invoices, purchase orders, timesheets or supplier statements and push them into your accounting system. The input is repetitive and the output is checkable at a glance.
3. Quote and proposal preparation. Assemble first-draft quotes from a request plus your existing pricing rules. A human still approves and sends — which is exactly the right division of labour.
4. Appointment and job scheduling. Match availability, send confirmations, handle reschedules. Particularly strong for trades, clinics and professional services.
5. Reporting and data consolidation. Pull weekly numbers from several systems into one summary. Low risk, unambiguously measurable, and it removes work nobody enjoys.

What should you not automate first?

Three categories cause most of the failed first projects:

• Judgement-heavy work. Pricing negotiations, performance conversations, complaint resolution, anything where the right answer depends on context a system can't see. These can be assisted by AI later, but they're a poor first project.
• Anything touching sensitive personal information before your guardrails exist. Health records, employment files, financial hardship cases. Under the Privacy Act 2020 these carry real obligations, and the Office of the Privacy Commissioner expects a privacy impact assessment and a human review step. Do this work — just not as your first, unproven project.
• Undocumented processes. If three people do the same task three different ways and all of them are right, you don't have a process yet. Document and standardise first; you'll often find that alone recovers most of the time you were trying to save.

How do you know your data is ready?

Data readiness is the most common reason a well-chosen project stalls. Before committing, check four things:

• Is the information in a system with an API, or is it in inboxes, spreadsheets and paper?
• Is it consistent enough that a rule would work — or does every record have its own quirks?
• Who owns access, and can they grant it within a week rather than a quarter?
• Does it contain personal information, and if so, have you agreed what may be sent to an AI tool?

Two 'no's here don't mean stop — they mean the data preparation is part of the project, and the timeline and budget should say so honestly upfront.

We map and score these processes as the first phase of every engagement — see how that works in our AI implementation roadmap for NZ SMEs.

What does a good first project look like in practice?

A well-chosen first automation has a shape you can recognise before it starts:

• It runs at least several times a day.
• One or two systems, both with APIs.
• A human approves anything that reaches a customer.
• There is a number from before the project you can compare against.
• It ships in weeks, not quarters.
• If it failed entirely tomorrow, the business would revert to the manual process and carry on.

That last point is the real safety test. If a workflow failing would stop the business, it is too important to be the one you learn on.