What to automate first: five workflows for a Vancouver small business.

Most automation lists offer a long menu and no way to choose. We start with five workflows, in order, and a clear boundary for where a person stays involved. Four do not need AI. They are simple rules that can save time and create cleaner records for the parts of the work where AI may help later.

Do not begin with a product demo. Write down what comes in, who owns the next action, where the work waits, and which record the team trusts. A new tool cannot repair an unclear handoff; it usually gives the same uncertainty another place to hide. Once the workflow is visible, decide whether the right fix is a clearer process, a configuration change, one connection between tools, or a new system.

The rule for choosing the order

Ask two questions of each candidate: how often does it happen, and what does a mistake cost? Begin with frequent work where the error is cheap and easy to correct. That is where a rule is safest and the saved hours add up. Zapier’s survey of small business workers found 94 percent doing repetitive tasks, especially data entry, document creation, and lead follow-up. Intuit’s 2024 survey of firms with 10 to 99 employees put manual data entry and reconciliation at 25 hours a week. They are vendor surveys of American businesses, so the exact figures are only indicative. The underlying problem will be familiar to a Burnaby plumbing company.

1. Missed calls and lead response

A missed call can be costly when the owner is on a roof or halfway through a job. Our recommendation is simple: reply while the request is still fresh, then put the response and its next step in one place. Harvard Business Review’s report on lead-response timing is useful further reading, but it concerns web leads and should not be treated as a trades-specific missed-call rate. You probably already know when your phone tends to go to voicemail.

Start with a rule. A missed call triggers a text within a minute: "Sorry we missed you, what do you need and when?" The reply lands in one place with the caller’s number and a timestamp, and the owner gets a notification. That can recover leads that would otherwise call the next company. An AI agent that answers and qualifies calls is a later step, best kept to bounded after-hours work. That is the use case we built ReplyFirst for in home-service trades. A person still confirms any price or booking.

2. Appointment reminders

Appointment reminders are often a sensible early automation because a no-show has a visible cost. The strongest evidence here comes from healthcare: a 2018 review of 105 studies put the average no-show rate around 23 percent, with long lead times and a previous no-show as the strongest predictors. That does not establish the same rate or pattern for trades. It is a prompt to check your own missed-appointment data.

Set a confirmation when the booking is made, then a reminder 24 hours before with a one-tap confirm or reschedule. If no one confirms, flag it for the front desk that morning. It is straightforward, rule-based work. There is no reason to wait for an AI project to do it.

3. Invoicing and payment reminders

Xero’s March 2026 data from roughly 12,000 Canadian customers shows invoices paid 11.6 days late on average, up from 10.5 the quarter before, with total time to be paid near 30 days. The report does not establish why each invoice was late. For a small business, that makes a clean invoice workflow and a visible follow-up list a reasonable place to inspect first.

Generate the invoice from the completed job record, not from memory. Send a polite reminder three days before it is due and a firmer one seven days after. The owner should have one list of outstanding invoices. AI can help draft an awkward third reminder once the first two rules have run. A person still decides when the relationship matters more than the seven-day rule.

4. Review requests and responses

Reviews are easy to leave to chance. BrightLocal’s 2026 consumer survey found 97 percent of respondents read local reviews, 68 percent expected four stars or better before considering a business, and 80 percent were more likely to use a business that responds to all reviews. Nineteen percent expected a same-day response. The sample is American, so treat it as a signal to inspect your own review behaviour rather than a Vancouver measurement.

Send a review request with a direct link once the job is complete and paid, never before. AI can draft a response in your voice; a person should read it before posting. That review is worth the minute it takes. A generic reply to a negative review can be worse than silence.

5. Quotes and follow-up, with a person on the button

Quoting happens often and matters a great deal, which makes end-to-end automation tempting. A wrong price is still the costliest error on this list. Use rules around the quote: capture the right questions, put site photos in the job record, send a follow-up after three days, and flag a quote that sits for two weeks. AI may draft from your price book and past jobs. A person reads every number before it goes out. That is the practical line between assistance and an unreviewed commitment.

Tier 1

Automate now

Frequent, well defined, cheap to get wrong.

Missed-call text-back, booking confirmations and reminders, invoice generation and the first two reminders, post-payment review requests, internal notifications when a form arrives. Each is a rule you can write in one sentence, and each runs hundreds of times a year.

What qualifiesYou can write the rule in one sentence and nobody would be upset to learn it was automated.

How to build itA form or booking tool that feeds one record, plus an automation platform for the notifications. Days of work, no model required.

Tier 2

Automate with a check

Frequent, needs judgment, a person confirms.

Review responses, the third payment reminder, quote drafts, summaries of a long customer thread before a handoff, first-draft replies to inbound questions. The variation is too high for rules and the cost of a bad output is real, so AI drafts and a person approves.

What qualifiesThe step needs reading or writing on every pass, and a wrong answer would embarrass you or cost money.

How to build itAI draft into a queue, one-tap approve or edit, nothing leaves without the tap. Measure how often the draft is edited; if it is rarely, the check is still worth keeping.

Tier 3

Not yet

Undefined, rare, or a customer would object.

Anything you cannot describe as a process, anything that happens twice a year, hiring decisions, complaint resolution, pricing changes, and any conversation a customer would be angry to discover was a machine. Automating a workflow you have not mapped copies the confusion into software at a monthly fee.

What qualifiesNobody can draw it on a whiteboard the same way twice, or the customer relationship is the product.

What to do insteadMap it first. Our readiness checklist scores whether a workflow is ready; if workflow friction scores low, that is the work, not a tool.

What not to automate, and the Vancouver-specific part

Three Canadian checks belong in the design review. If a workflow affects people in Quebec, check how Law 25 applies before using automated decision-making. Before automated texts or emails, check the CRTC guidance on consent and message requirements; voice calling has separate rules. For review requests, do not filter recipients by expected rating; the Competition Bureau addresses deceptive marketing. Build one honest request into the completion step and make it available to every eligible customer.

Start with the calls you miss. A quote may have a bigger dollar value, but it also has more ways to go wrong.

What to do Monday

Count last week’s missed calls, no-shows, late invoices, and unanswered reviews. Start with the number that catches you off guard. Write the rule in one sentence, choose a tool that can follow it, then check the same number again in 30 days. The AI Readiness Check is a useful first pass if you need a view of lead capture, workflow friction, and tool sprawl. For the reasoning behind the tiers, read automation vs AI.

Rules first.The phone before the pitch deck.

Common questions

What are some good automations for small businesses?

Five places worth comparing are a text-back for missed calls, appointment confirmations and reminders, invoice generation with automatic payment reminders, review requests sent after a job is paid, and follow-up on unanswered quotes. All five are rules-based, need no AI, and produce the clean records that make an AI step worthwhile later.

Which business process is best suited for automation?

A process that happens often, follows the same steps every time, and is cheap to get wrong. Missed-call text-back and appointment reminders are the clearest examples: hundreds of repetitions a year, one sentence of logic, and no customer is upset to learn a reminder was automated. Processes that need judgment on every pass, like quoting, should be assisted rather than automated, with a person confirming the output.

Can you give me an example of workflow automation?

A customer calls while the owner is on site and the call goes unanswered. Within a minute the system texts the caller asking what they need and when, logs the reply against the caller’s number in one record, and notifies the owner. If the caller asks for a quote, the request is captured with the right questions and a follow-up is scheduled for three days later. Every step is a rule; the only judgment, the price, stays with a person.

Can ChatGPT automate tasks?

ChatGPT can draft, summarize, and classify text when a person prompts it, which makes it useful for the judgment steps in a workflow, like writing a review response or a third payment reminder. It does not, by itself, watch for a missed call, send a reminder on a schedule, or move a record between tools. Those are rules-based automations, built with a form, a record system, and an automation platform, with AI added only at the steps that need reading or writing.

What is the best workflow software for small businesses?

The one that fits the rule you wrote, not the one with the longest feature list. For most small businesses that means a booking or intake tool that feeds one record system, an automation platform such as Make or Zapier to move data and send notifications, and an AI step only where drafting is needed. Choose the record system first, because every other tool has to read from it.

See what a workflow automation project includes, from mapping the work to testing the handover.