This month, we set up an automation that reads messages received through a contact form, sends them to a language model, and has the model classify each one and draft a reply. The setup took about an hour in Zapier, without any code. A year or two ago, the same result would have required a small software project.
Zapier has existed since 2011, and Make (known as Integromat until its renaming a couple of years ago) has also been available for a long time. For most of that period, these tools moved data from one system to another: a form on one side, a spreadsheet on the other. It is now possible to insert, in the middle of a workflow, a step that reads and interprets content.
This makes automation considerably more attractive for a small business. It also makes it considerably easier to build something fragile.
What is worth automating in a small business?
The most reliable rule we apply is simple: automate the tasks you perform the same way every time.
A task that follows the same steps with the same inputs, and that you perform every week, is a good candidate. When each case differs and requires judgment, the task is better left with a person, at least for the time being.
The following five automations suit most small businesses, from dive shops to gîtes to contractors. They are ordinary, and that is precisely why they work.
1. Booking or contact form to calendar and confirmation
A visitor completes the form on your website. The automation creates the entry in your calendar or booking sheet, sends a confirmation email and notifies you. There is no copying between tabs, and no need to apologize to a client for a forgotten follow-up.
2. Invoice reminders
When an invoice passes its due date in your accounting tool, a courteous reminder is sent automatically, followed by a second one a week later. You intervene only if these reminders do not produce a result. Few people enjoy pursuing payments, and this approach handles the task without any awkwardness.
3. Follow-up after a visit or a job
A day or two after the stay, the course or the repair, the client receives a thank-you message with a link to leave a review. This automation is more important than it appears. Reviews come from asking at the right moment, and business owners often forget to ask. We have observed this situation many times.
4. New newsletter subscriber, with a welcome message
A person signs up on the website or ticks the box when booking. They are added to the correct list in your email tool and receive a short welcome message. This is a basic step, yet many small businesses still copy these addresses by hand once a month, or not at all.
5. Sorting incoming email with AI
This is the new capability. An automation reads new messages in your general inbox, labels them (booking request, supplier, complaint, spam, question) and, for simple questions, prepares a draft reply for you to review.
The automation drafts; it does not send. This distinction is essential, and we return to it below.
Should a small business choose Zapier or Make?
This question comes up frequently. When you are starting out, the choice matters less than one might expect.
Zapier is easier to learn. Its logic is very linear: when one event occurs, a given action is performed. It connects to a very large number of applications. Make is more visual, handles workflows that branch in several directions more effectively, and can be less expensive once you run a high volume of operations. Both allow you to add an AI step.
We recommend choosing the tool that connects to the applications you already use and building a single workflow. If the tool proves difficult to work with, try the other one. The Zapier learning centre is a useful free resource for understanding the basic concepts, even if you eventually choose Make.
How should you get started with automation?
Start with one automation, then add the next. The most common mistake is to build ten automations over a weekend. The result feels productive, until something changes (a form field is renamed, a password expires, an application is updated) and three of them stop working at the same time, without any warning. The issue is often discovered two weeks later, through a dissatisfied client.
We recommend a gradual approach:
- Choose the task that causes you the most frustration.
- Automate it and monitor it for a month.
- Record, in a simple place, what it does and which accounts it uses.
- Then add the next one.
This record is more important than it appears. Six months later, it is unlikely that anyone will remember why an email is sent every Tuesday.
What happens when an automation fails?
Every automation fails at some point. The key question is whether the business stops when it does. The answer is to keep a manual fallback, and a person in charge of the AI step.
For each workflow, you should know how to perform it by hand. If the booking automation stops working on a Saturday in July, can someone still take a booking and confirm it? If not, the business now depends on a single point of failure.
For AI steps, one additional rule applies: a person reviews anything sent to a client. Language models perform well at sorting and drafting. They can also state an incorrect price, date or policy with complete confidence. An incorrect answer sent automatically, in your name, costs more than the time saved. For that reason, our contact-form automation produces drafts, and a person sends the replies.
How should personal information be handled?
An automation often passes client names, email addresses and phone numbers through several services, some of them located outside Canada. In Quebec, Law 25 requires businesses to assess the risks before personal information is transferred outside the province. This does not prevent you from using these tools. It means you should know where the data goes and send only what each step actually requires. A model that sorts emails does not need the client's phone number.
What trends are emerging in automation?
AI "agents" are currently receiving a great deal of attention: software that does not only draft content but also carries out actions on your behalf. We recommend caution. An agent placed in a fragile workflow still produces a fragile workflow, with more ways for it to go wrong.
For most small businesses, the current gain is far more modest, and far more reliable: five ordinary workflows, set up one at a time, each with its record and its fallback. This approach can readily give an owner a few hours back every week.
If you would like support in selecting which tasks to delegate and configuring them so they do not fail unnoticed, our team offers this kind of automation work. Otherwise, we suggest starting with the task that causes you the most frustration; it is usually the right place to begin.