You pick your first automation by asking three questions: how often the process happens, how long it takes each time, and how much it costs you when it's delayed. In most of the small businesses we meet, the answer points to the first reply to a lead, and not to the most impressive thing you could build. In this guide we'll go over the method, ten processes worth knowing with a before and after for each, three automations we built ourselves, the difference between n8n, Make and Zapier, and what Israeli law requires of a marketing message. We checked the tools' prices and terms on November 21, 2026, directly on the companies' websites.
How to choose what to automate first
You can check any process in your business with three questions. Give each answer a score from 1 to 3, and multiply the three scores:
| The question | 1 point | 2 points | 3 points |
|---|---|---|---|
| How often does it happen? | Once a month | Every week | Every day, or several times a day |
| How long does it take each time? | A minute or two | 5 to 15 minutes | More than 15 minutes |
| What happens when it's delayed? | Nothing, it can wait until tomorrow | A customer waits and gets annoyed | A lead, an order or a payment gets lost |
The product matters more than any single question. A monthly report that takes two hours gets a low score, because it happens once a month and no customer is waiting for it. A reply to a lead that takes two minutes gets a high score: it happens every day, and a lead who waits until the evening can contact someone else in the meantime.
And there's one condition you check even before the score: the process needs a clear rule. "When a lead comes in, it's recorded and gets a message" is a rule. "When a customer is angry, reply the right way" is judgment, and there an automation can at most prepare a draft for you.
10 Small Business Automations, Before and After
These are the processes we come across most, roughly in the order we'd start with. Each row shows what happens today, and what happens once it's automated.
| The process | Before | After |
|---|---|---|
| First reply to a lead | The enquiry from the website lands in your inbox, and someone replies when they're free | The lead is added to the CRM, gets a WhatsApp message or an email right away, and you get an alert |
| A reminder before an appointment | You call or text every customer the day before | A reminder goes out with an option to confirm or reschedule, and the calendar updates itself |
| Invoice after payment | At the end of the month you issue invoices from a list | Every payment issues an invoice and sends it to the customer |
| A payment that didn't arrive | You go through the list and send awkward messages | Anyone who didn't pay on time gets a polite reminder, and you can see who still owes |
| Order updates in a store | "Where's my order?" on WhatsApp, and someone checks and replies | The customer gets an update at every stage, and a bot answers status questions |
| Abandoned cart | The customer disappears, and nobody knows | Anyone who left an email and opted in to marketing gets a reminder about what's still in the cart |
| Weekly report | You collect numbers from several systems into a spreadsheet | Every Sunday an email arrives with the enquiries, the sales and where they came from |
| Review request | You forget to ask | A few days after the service or delivery, a request to write a review goes out |
| Regular content for the site | There's no time to write, so nothing gets written | AI writes to a fixed structure and rules, and the content is published on a schedule |
| Customer questions on the site | A question asked at night waits until morning | An AI chat answers from the catalog and your policies, and passes you any conversation that needs a person |
Notice that no row in the table replaces a decision by the business owner. The automation does what you've already decided should happen, every time and without forgetting. The decision itself, what to reply to a lead and when to send a payment reminder, is yours to make, once.
Three Automations We Built, and What We Learned
A horoscope that writes and publishes itself
On Moshe Zohar's site, moshetarot.co.il, a daily horoscope is published every day for each of the 12 signs, and a weekly horoscope once a week. AI writes them based on the planetary transits at the time, and the system publishes them on its own. Each sign has a daily page with the same sections every day: a general horoscope, career, relationships, singles, health, advice for the day, and what's happening in the sky that day. At the bottom of the page are links to all the other signs.
What made this possible is the fixed structure. The AI doesn't decide what to write about: it gets the state of the sky for that day and a template with the same sections, and writes inside it. Content without that kind of structure, like an opinion piece or a reply to an angry customer, we wouldn't let AI publish on its own.
An AI sales agent in the LiNo Coffee store
In LiNo Coffee's WooCommerce store, an AI sales agent works in the chat. It asks customers what they like in a coffee, recommends from the catalog and adds to the cart right inside the chat. When a customer asks where their order is, the bot checks it against the store and replies. A conversation that needs a person goes to the store owner's Telegram, with a summary of what was said.
Our lesson from the project: the bot answers only from what's in the store. For an order, it checks the actual order instead of guessing, and when it doesn't know, it passes the conversation on. You can read more about AI chatbots for your website.
"Zman": a tool for scheduling WhatsApp messages
We built a tool that schedules messages in WhatsApp Web: you write ahead, pick a recipient and a time, and the message goes out on time. Most of the work actually went into what the tool refuses to do: automatic sending only to saved contacts by default, a random gap of 45 to 120 seconds between messages, quiet hours between 10:00 PM and 8:00 AM, and no more than 3 automatic messages to the same person a week. We listed all the limits, and why each one exists, in our guide to scheduling WhatsApp messages.
n8n, Make or Zapier: Which Tool to Build With
At their core, all three tools do the same thing: they wait for an event, like a submitted form, a completed payment or a time of day, and then run a chain of actions. The practical difference between them is how they count what you pay for, and who runs the server.
| n8n | Make | Zapier | |
|---|---|---|---|
| What you pay for | Each workflow run, no matter how many steps it has | Credits. Most actions in a workflow cost one credit each | Tasks. Every successful action counts, the trigger doesn't |
| What's free | The Community edition, which you run on your own server | 1,000 credits a month | 100 tasks a month, only in workflows with a trigger and one action |
| First paid plan | Starter: €20 a month billed annually, 2,500 runs | Based on the number of credits per month | Pro: $19.99 a month billed annually, 750 tasks |
| Best suited for | Long workflows, lots of runs, and anyone who wants the data on their own server | Short and medium workflows in a visual interface | Quickly connecting two well-known systems |
An example that shows the difference: a lead workflow with five steps, receiving the form, adding it to the CRM, a message to the customer, an alert to you and a row in a spreadsheet. In n8n that's one run. In Make it's roughly one credit per step. In Zapier the trigger isn't counted, and each of the four actions is a task. With 300 leads a month, that's 300 runs in n8n, about 1,500 credits in Make, and 1,200 tasks in Zapier.
We work mostly with n8n and Make. n8n is our default for long workflows and high volumes, and Make for short workflows the client wants to see and understand themselves. Zapier is handy when you need to connect two well-known systems quickly. And when no tool does the job well, as with the horoscope or the message scheduling tool, we write code.
AI Automation for Business: Where AI Really Helps
Inside a workflow, ChatGPT and Claude are good at three kinds of work: classifying an enquiry (a price question, a complaint, a booking request), summarizing (a long conversation into three lines for the owner), and drafting to a fixed structure. They're less suited to anything that has to be 100 percent correct, like a price, stock or an order status. The bot needs to check those in the system itself. This is how we build every workflow with AI:
- The AI answers only from information you gave it: the catalog, the policies, the opening hours.
- An answer about money, an order or stock is checked against the system, not written from the model's memory.
- When the bot isn't sure, it hands over to a person with a summary of the conversation.
- In the first month we read the conversations, and adjust the instructions based on what we see.
What Goes Wrong with Automations, and How to Avoid It
Silent failure
The connection to another system breaks, the workflow stops running, and nobody notices until a customer asks why nobody got back to them. Set up an alert for every workflow when it fails, to the email or WhatsApp of the person responsible for it. Without an alert, you have no way of knowing.
Duplicates and loops
The same lead is recorded twice, or a workflow triggers itself again and again. Before creating a new record, check whether it already exists, by phone number or email.
A change in another system
A vendor changes a field in its form or its interface, and the workflow keeps running with empty data. The alert catches what failed. A periodic check catches what's still running but no longer doing what it should.
Automating a broken process
If leads get lost today because nobody is responsible for them, automation will only send them to the same place faster. First decide who handles them, when and how, and only then build.
Messages to people who didn't ask
That's already a legal matter, so it gets its own section.
Marketing Messages: What the Spam Law Requires
In Israel, the rule is in Section 30A of the Communications Law (Telecommunications and Broadcasting), which was added to the law in 2008 and is usually called the "spam law". It prohibits an advertiser from sending "advertising material" by fax, automated dialing system, electronic message or short message (SMS) without the recipient's explicit prior consent, in writing. Consent can also be given in an electronic message or a recorded call.
Under the law, "advertising material" is a message distributed commercially to encourage buying a product or service, or spending money in some other way. An "electronic message" is a message sent over the internet that can be stored and read on a computer. A marketing email falls under the definition, and we treat a marketing message on WhatsApp with the same caution. By the definition, an appointment reminder, an invoice or a shipping update isn't meant to encourage a purchase. An appointment reminder that ends with "and we have a sale on..." is.
What the law requires of a marketing message sent lawfully:
- Explicit consent, in advance and in writing, from the person receiving it. On our forms that's a separate checkbox the visitor ticks themselves.
- The word "advertisement" at the start of the message, and in an email, in the subject line.
- The advertiser's name, address and contact details. In an SMS, the name and a way to get in touch to ask to be removed are enough.
- A simple way to say you don't want any more messages, at any time and free of charge. In an email, a web address where the request can be sent.
There are two exceptions. The first: a customer who gave you their details when they bought from you, or while negotiating a purchase with you. You may send them advertising about a similar product or service, if you told them at the time that their details would be used for advertising, gave them a chance to refuse, and they didn't refuse. The second: a one-time approach to a business, offering it to agree to receive advertising, isn't considered a violation.
And what happens when you break the law? When a message was knowingly sent in violation of the law, a court may award compensation of up to NIS 1,000 per message, without the recipient having to prove any damage. The same amendment also added the option of a class action against an advertiser on these grounds. An old list of 500 phone numbers where nobody opted in is a risk no promotion is worth.
WhatsApp has its own rules too. Under the WhatsApp Business Policy, you contact a person only if they gave you their number and agreed to receive messages from you, and you honor every request to stop. In other words, a marketing message on WhatsApp has to meet both sets of rules, the law's and WhatsApp's.
Checklist Before You Turn On an Automation
Build It Yourself or Get Help
A workflow of a form, a spreadsheet and a message you can build yourself in Make or Zapier, without writing code. It's worth getting help when the workflow touches money, when it connects several systems, when you need AI that talks to customers, or when you don't have time to maintain what you built. Our Business automation page shows the projects we've built and how a project with us works, from scoping to the month of support after launch.
FAQ page
What is business automation?
A process that runs by itself after an event, with no one having to remember to do it: a lead added to the CRM and sent a message, an invoice that goes out after a payment, a reminder before an appointment, or a report that arrives every week.
What should I automate first?
Whatever repeats most, takes time and costs money when it's delayed. Give each process a score from 1 to 3 on each of the three questions, and multiply. In most small businesses, the winner is the first reply to leads.
n8n or Make: which suits a small business?
For a short workflow you want to understand and change yourself, Make is easier. For long workflows or lots of runs, n8n works out cheaper, because it charges per run and not per action, and you can also run it on your own server.
Can I send automated marketing messages on WhatsApp?
Only to people who agreed in advance to receive them, and through the WhatsApp Business Platform, with templates that WhatsApp approves. WhatsApp's terms prohibit bulk and automated messaging from the regular app. We covered this in our guide to scheduling WhatsApp messages.
Can AI answer customers on its own?
Yes, when it answers from information you gave it, checks orders and stock against the system, and hands over to a person when it isn't sure. That's how the sales agent we built for the LiNo Coffee store works.
Who owns the automation once it's built?
You should. The accounts with the vendors are opened in the business's name, and with us, once you've paid in full, the workflows we built belong to you.
Is an appointment reminder considered advertising?
By the definition in the law, advertising is a message meant to encourage a purchase or spending money. An appointment reminder on its own isn't meant for that. If you add an offer or a sale to it, it may already count as advertising, and all the rules apply to it.
Sources
- Communications Law (Telecommunications and Broadcasting) (Amendment No. 40), 5768-2008 Section 30A, Book of Laws 2153, June 1, 2008
- n8n: Pricing Checked November 21, 2026
- n8n: Self-hosting and the Community edition
- Make: Pricing and credits Checked November 21, 2026
- Zapier: Pricing and tasks Checked November 21, 2026
- WhatsApp Business Messaging Policy Updated September 23, 2026
- Moshe Zohar: Daily horoscope, Taurus An example sign page
Have a question?
Leave your name and phone number, and we'll get back to you.










