What should a small business automate first?

The short answer

A small business should automate a repeated, predictable task with a clear input, a clear result and an easy way for someone to check it. Enquiry routing, internal reminders and copying information between existing tools can be good starting points. Keep risky decisions with a person.

You do not need an AI agent to do every boring job. If a rule can describe the work reliably, ordinary automation may be enough. Consider AI when the task involves variable language, classification or drafting, and make room for mistakes and review.

The first project should solve one recognisable problem. "Put new enquiries in the right inbox" is a workable brief. "Automate the business" is not.

How do you choose the first task?

List a few jobs that repeat during a normal working week. Look for one that causes avoidable copying, waiting or missed handovers, then answer these questions before choosing software:

• How often does it happen? Count actual cases, including quiet periods.
• What does it cost in effort? Include checking and correcting the work, not just the initial typing.
• Are the rules clear? Can two people follow the same instructions and reach the same result?
• Can the system access the right information? Check permissions, data quality and whether the existing tools support the connection.
• What happens when it goes wrong? A misfiled internal note and an incorrect payment have very different consequences.
• Who owns the exceptions? Name a person and keep a manual route available.

Start with a task that repeats enough to matter, has clear rules and has manageable consequences if it fails. If nobody can explain the current process, write it down first. Automating an unclear handover can make the confusion happen faster.

Do you need rules, AI or a person?

Ordinary automation follows defined rules. AI assistance can interpret messy text or prepare a draft, but its output still needs checking. Neither removes the need to decide who is allowed to act.

Choosing the simplest useful approach

Approach	Example first task	What to check
Ordinary automation	Route a form using the enquiry type selected by the visitor.	Required fields, destination, duplicate handling and delivery failures.
AI assistance	Prepare a short summary of a free-text enquiry for someone to review.	Missing context, invented details, sensitive information and review effort.
Human decision	Agree a price, approve a refund or decide whether a project is a good fit.	Authority, business context and the consequences of the decision.

These are possible tasks, not claims that every example is running in our products. A single workflow can combine all three approaches.

If you are choosing a provider rather than choosing a first task, our guide to AI automation services for small businesses explains the broader commissioning questions.

A real example: organising incoming enquiries

Our own Transmission enquiry system starts with a small, defined job: collect enough information for a useful first reply and send it to the relevant inbox.

The project brief asks what someone wants to build, their current tools, rough budget and timeline. Server-side validation checks the submitted fields. Project briefs and Test Pilot requests have separate routing settings, and the visitor gets a confirmation or a fallback message if delivery is unavailable.

That is the implemented intake flow. It does not need an AI agent to decide which of those two forms someone submitted. The decision about project fit and the substantive reply remain with a person.

For a business receiving long enquiries, an AI-written summary might be a useful later addition. That is a hypothetical next step here, not a feature of this intake flow. It would need checking against the original message, appropriate data handling and evidence that review takes less effort than reading the enquiry directly.

The lesson is to make the basic handover work before adding interpretation. We are not claiming a measured time saving or conversion increase from this example.

What should stay manual at the start?

Keep approval with a person when a mistake could affect money, access, customer trust or information that is difficult to recover. Refunds, contract changes, deleting records and sensitive customer replies are poor places to give an untested system unrestricted control.

Also decide what information the workflow genuinely needs. Do not connect an entire inbox or customer database simply because a tool makes it easy. Limit access, decide what gets retained and make it possible to correct mistakes.

A useful fallback is specific: who receives the problem, what information they get and how they finish the job manually. "The AI will work it out" is not a fallback.

How do you know the automation is worth keeping?

Record how the job works before changing it, then compare similar work after the trial. Track:

1. The number of cases handled.
2. Time spent doing, checking and correcting the task.
3. Errors, missed steps and cases needing manual help.
4. Running costs and maintenance effort.

For example, suppose you observe 40 cases taking six minutes each. That is four hours of handling time. If a trial handles 40 comparable cases with two minutes of review each, plus 30 minutes of troubleshooting, it takes one hour and 50 minutes. The difference is two hours and ten minutes, before accounting for setup effort and ongoing costs.

Those numbers are illustrative, not a Wacky Works customer result or a savings promise. Use your own observations. Faster handling is not an improvement if important details are lost or mistakes create more work elsewhere.

Copy this first-workflow brief

Fill in these prompts before buying a tool or commissioning a build. Redact personal information from any examples and do not include passwords or access keys.

Bring us one workflow

Before replacing the tools you already use, work through buy, build or integrate. Once the task is clear, use the automation services buying guide and maintenance checklist to agree delivery and ongoing ownership.

You do not need a technical specification to start a conversation. Send us the job you want to improve, the tools involved and what a useful result would look like. Leave sensitive records out of the form.

If you want dedicated time to work through the options first, a paid AI consultation is another route. It is not a requirement for sending a project enquiry.

Related capability

Automations

Workflow automations that save time across handoffs, repeated admin, and reporting.