What Should a Small Business Automate First?

Detailed Explanation

The instinct is to automate whatever feels most impressive — a customer-facing chatbot, an AI agent, a fully automated pipeline. That's usually the wrong place to start. The best first automation project is the most boring one: a process your team already does the same way, over and over, that is currently costing real time or causing real errors.

Three questions narrow this down quickly:

1. How often does it happen? A process done fifty times a week is worth far more attention than one done twice a month, even if the second one feels more painful in the moment.
2. How rule-based is it? If you can write the steps down as a clear decision flow ("if the invoice total matches the PO, approve; if not, flag for review"), it's a strong automation candidate. If the steps depend on judgement, negotiation, or reading between the lines, it isn't ready yet.
3. How much does it currently cost you? Count the hours spent doing it, the cost of the errors it produces, and any downstream delays it causes (a slow onboarding process that pushes back a new hire's start date, for example).

A process that scores high on all three — frequent, rule-based, and currently expensive in time or errors — is the one to automate first. A process that only scores high on one, however dramatic it feels day to day, usually isn't.

Once that first process is running, the question changes from "what's the single best starting point" to "in what order do we tackle everything else" — see how do you prioritize which processes to automate next for sequencing a roadmap across multiple competing candidates.

A Simple Way to Find the Candidate

Walk through a normal week and list every task that is:

• Repetitive (done the same way more than a few times a week)
• Manual (someone retypes, re-enters, or copies data between systems)
• Time-boxed (it has a clear start and end — "process this invoice", "send this reminder" — rather than being an ongoing judgement call)

For most small businesses, the shortlist that comes out of this exercise looks similar regardless of industry:

• Invoice and billing admin — matching invoices to purchase orders, chasing overdue payments, re-entering supplier invoices into accounting software, and routing purchase and expense requests through an approval chain instead of an email back-and-forth.
• Scheduling and reminders — appointment confirmations, follow-up emails, renewal reminders.
• Data entry between systems — copying a new lead from a web form into the CRM, or a new order from email into a spreadsheet, which often means connecting systems that don't integrate natively.
• Document handling — sorting incoming files, extracting figures from PDFs, generating standard documents from a template (often built with a tool like Power Automate if you already run Microsoft 365).
• Routine email triage — sorting incoming email into categories, flagging urgent messages, sending acknowledgement replies.
• Sales follow-up — chasing new leads on a schedule so a prospect never goes quiet just because someone forgot to send the next email.
• New-hire admin — employee onboarding paperwork and IT provisioning, which tends to be rare enough per hire but painful enough when missed that it's worth automating early.

Pick whichever of these matches your actual week most closely, rather than the one that sounds most "AI". If your highest-friction process is specific to your industry rather than generic back-office admin — proof of delivery, client intake, RFI tracking, shop-floor reporting — it may outrank anything on this generic list; see the industry automation hub for the named process most businesses in your industry automate first.

Things to Consider

• Fix the process before you automate it. Automating a broken or inconsistent process just makes the mess happen faster. If nobody can agree on the "right" way the task is currently done, standardise it first.
• Use what you already own. Most businesses already run software — accounting platforms, a CRM, Microsoft 365 — with built-in automation features (rules, workflows, approvals) that cover a first project without buying anything new.
• Keep a human in the loop at first. For anything customer-facing or involving judgement calls, start with automation that drafts or flags rather than one that acts unsupervised, then reduce human review as it proves reliable.
• Measure before and after. Note the time the process currently takes and the error rate before you start. Without a baseline, it's hard to tell whether the automation actually helped — see how do you measure the ROI of automation for a practical framework for capturing that baseline and using it.
• Start narrow. Automating one supplier's invoice format, one email inbox, or one form is a smaller, more achievable project than "automate all our invoices" — and it gives you a working example to expand from. Treat this narrow first attempt as a deliberate pilot rather than a quiet soft-launch — see how do you run a pilot before rolling out an automation project for scoping it properly.
• Not every candidate needs AI to automate it. Rule-based tools handle most of the processes above just fine — see what is AI automation (and how is it different from regular automation) for how to tell which of your candidates actually need a model versus a simpler rule-based flow.
• Know roughly what a first project costs before picking one. A narrow first candidate is also easier to budget for — see how much does AI automation cost for a small business for typical subscription and setup cost ranges.
• A narrow candidate is also easier to get approved. A small, cheap, reversible first pick makes the internal pitch simpler — see how do you get leadership buy-in for an automation project for turning this candidate into an ask leadership will actually approve.
• Decide who owns this before it's live, not after. Picking the process is only the first half of the decision — see who should own automation projects in a small business for choosing between an existing employee, a dedicated hire, and a consultant.
• Write down exactly how the process runs before you automate it. Once you've picked a candidate, don't skip straight to building — see what is process mapping, and do you need it before automating for capturing the process's real steps and exceptions first.

Common Mistakes

• Starting with the most complex process because it's the most painful. Painful and complex processes are usually painful because they're inconsistent — they need cleaning up before automation will help.
• Buying a new platform before checking existing tools. Many businesses pay for a new automation tool before realising Power Automate, their CRM's workflow builder, or their accounting software already does what they need.
• Automating end-to-end on the first attempt. Trying to automate an entire process from start to finish in one project increases the chance of missing an edge case. Automate the highest-value step first, then extend it.
• Skipping a review step for customer-facing output. Sending AI-drafted or automated replies straight to customers without any check, before you've confirmed accuracy, risks visible mistakes reaching people outside the business.
• Choosing a process based on novelty rather than volume. A rarely-run process that would be fun to automate is a worse first project than a dull, frequent one that quietly eats an hour every day.

These mistakes are the recurring pattern behind most failed projects — see why do automation projects fail for the full list and how to avoid it.