5 repetitive tasks to automate first in a small business

Every owner knows the sentence: "I'll do it by hand, it only takes five minutes." Then it happens again on Thursday, and again next week, and the five minutes become a standing part of the job that nobody chose.

This post names five repetitive tasks that are worth automating first in a small business, says what each automation does, what it can get wrong, and what the tools cost. It also says where we would not automate. Gonzo Digital builds agents for some of these tasks, but a cheap flow tool is the right answer for several of them, and we say which.

Count before you buy

Articles on this subject usually open with a figure for how many hours the average employee loses to repetitive work. We looked for one we could trace to a source and did not find one we would stand behind, so we dropped the claim from the earlier version of this post.

Use your own numbers. For one week, write down each task you repeat, how many times it happened and how long each took. Multiply by your hourly cost. The arithmetic on our cost calculator shows the sums, and it is explicit that it is a calculation, not a forecast. A task that costs you 20 minutes a month is not worth a build. A task that costs 20 minutes a day probably is.

A good first candidate has four traits. It happens often. The steps are the same each time. A mistake is cheap to fix. And it has a clear trigger and a clear end, such as "a form is submitted" and "the record exists in the CRM". Tasks that fail the last two, or that involve a complaint or a large sum, should keep a person in them.

Task 1: Data entry between systems

The pain is copying a lead from a form into a spreadsheet, then into the CRM, then fixing the formatting. Each manual step invites a typo, and a typo in a client record can follow you for months.

What automation does. When someone submits a form, a flow creates or updates the CRM record and adds a row to a sheet. A CRM integration of this kind is the simplest automation there is: one trigger, two or three steps, and no customer-facing words.

What can go wrong. Duplicate records when the same person submits twice, and fields that do not match between systems. Decide in advance how to match an existing contact, usually by email address, and send rows that fail validation to a person instead of dropping them.

This is the clearest case for a flow tool. We do not sell a system for it.

Task 2: Invoicing and payment reminders

Chasing payments is the part of running a business that most people put off. Raising an invoice, sending it and remembering to ask again all depend on memory.

What automation does. When a project is marked complete, an invoice is created and sent, and a reminder is scheduled for a few days after the due date. When the payment is recorded, the reminders stop. That last rule matters more than the rest.

What can go wrong. A client who paid by bank transfer that nobody recorded still gets a "your invoice is overdue" email. A reminder to a good client at the wrong time does damage. Send reminders only from an accurate payment record, review the wording once, and keep disputes and money questions with a person. Our Quote-to-Invoice Automation is built around this rule: reminders stop the moment a payment lands.

Task 3: Following up new leads

A lead arrives, gets buried, and by the time someone answers, they have booked with another firm. The speed advice you will see says to reply within five minutes. Before you build around it, know where that number comes from: a 2007 study by a sales-software vendor and a 2011 article that shared its lineage, which record what firms did and are not controlled experiments. We went through what each measured in how fast to reply to a new inquiry. The sound lesson is that a real answer from a person soon beats a vague auto-reply.

What automation does. The moment a lead arrives, an acknowledgment goes out, a task is created for the right person, and the team is notified in Slack or email. If a second message is sent by text or WhatsApp, that is a separate decision.

What can go wrong. Messaging someone without their consent. The basics are in our consent notes, and the WhatsApp rules on the 24-hour window, templates and opt-in are in a separate post. A system that speaks for you in your name should also say it is automated.

Where the first reply must understand a question, not just acknowledge it, the Lead Response Engine is built for the job. It says it is an AI, stops on request and hands off to a person. It has run only in a sandbox.

Task 4: Scheduling content

Keeping a consistent presence means planning, scheduling and posting by hand across several apps, and missing days.

What automation does. Keep captions, images and dates in one sheet. A flow reads each row on its date and sends it to your scheduling tool or the platform. Change a caption once and it changes everywhere.

What can go wrong. The wrong post goes out on time. Add an approval column that must say "approved" before a row can publish. And a note on scope: Gonzo does not build social media tools, so this one is for a flow tool or the platforms' own schedulers.

Task 5: Onboarding paperwork

A new client waits for a contract, a confidentiality agreement or a welcome pack, and the delay looks disorganized.

What automation does. When a deal moves to "won", a contract goes out for e-signature, a welcome email follows, and documents are shared. When the contract is signed, the next steps start by themselves. We describe the three steps, and where they fail, in client onboarding automation. The Client Onboarding Pipeline is our version of it.

What the tools cost, and what the price leaves out

A common line is that a decent workflow costs a few dollars a month. It can. Here is what three vendors' pages showed on October 5, 2026. Zapier lists a free plan with 100 tasks a month, with paid plans starting at $19.99 a month for Professional.^[1] Make lists a free plan with up to 1,000 credits a month and a $9-a-month plan with 5,000 credits.^[2] n8n lists cloud plans billed annually, starting at EUR 20 a month for 2,500 workflow executions.^[3] The units differ: Zapier counts tasks, Make counts credits, and n8n counts whole-workflow executions, so they cannot be compared directly. We read Make's page through a page-reading tool, because a direct request was refused.

The subscription is the smaller cost. Three others are easy to miss:

• Usage limits. On Zapier, if you reach your plan's task limit you are switched to pay-per-task billing unless you turn that off, and if it is off your Zaps pause until the next period.^[1] A flow you rely on can stop at the wrong moment.
• Failure handling. Flows fail. Zapier lets you replay a run that errored, by hand or automatically, from its history.^[4] Someone has to look.
• Build and upkeep time. The hours to build, test and maintain it are yours, or someone's you pay.

Compare the total with the hours you counted earlier. If the sums favor automation, start with the cheapest tool that does the job.

Four myths, checked

"Automation is expensive." For simple flows it is not, as the prices above show. It is expensive when you pay for a build that saves 20 minutes a month.

"I need a developer." Not for a short flow between two apps: these tools use visual builders. The point where code starts to matter is when the automation speaks to your customers or touches money. It then has duties, such as disclosure, consent and a log, and each needs tested code. Our page on no-code or Python sets out where that line sits.

"Automation takes jobs." It takes defined steps, not a role. We cannot promise what it does to anyone's job, and we removed an earlier claim of a 40% productivity jump in one company because there was nothing to support it.

"It's only for large companies." Small teams can start on free plans, as above. Equal footing with large competitors is not something any tool can promise.

Where AI fits, and where it does not

A language model can draft a reply, sort an inquiry or pull fields out of a message. (If the word "agent" is new, what an intelligent agent is covers the five classic types.) Those jobs have fuzzy edges, which is where a fixed flow struggles. Anthropic's engineering team suggests starting with the simplest solution that works, and notes that autonomy brings higher cost and the potential for compounding errors, so test in a sandbox and add guardrails.^[5] In practice: use plain rules where the steps are known, use a model only for the wording or sorting, and put a person at any step involving money, a complaint or a legal question. The earlier version of this post promised "self-tuning growth engines" and lead scoring before you open an email. We cannot support that, so it is gone.

How to start

1. Count your own repeated tasks for one week.
2. Pick one that meets the four traits above.
3. Build it with the cheapest tool that fits, or ask us to scope it.
4. Add an alert for failed runs and a person for exceptions.
5. Check it after a week, then after a month, before adding the next.

If you have a task in mind, the free audit is how to find out whether it suits one of our nine systems. And if a flow tool is enough, our comparison says so plainly.