A new lead messages your business at 9 p.m. You reply at 9 a.m. the next morning, twelve hours later, and by then they have already messaged your competitor. That gap is the cheapest problem in a small business to fix, and it is the perfect first automation. This is AI automation for small business in its simplest form: when a new lead comes in, software drafts a personal reply in your tone, and you approve and send it in one tap. The AI drafts. You send. Nothing goes out without a human looking at it first.
This is the inbox version of the missed call problem. The business that answers first usually wins the job, and the business that answers hours later is explaining itself. It is the same problem my AI voice agent for a dental clinic solves on the phone, catching the call instead of letting it ring out. Here the cost to start is nothing, because this scenario runs on Make's free tier, if you build the trigger the right way.
Contents:
- Quick verdict: should you build this first?
- What is AI automation for a small business, in plain words?
- What is Make, in plain words?
- How much does this automation cost on the free tier?
- How do you build the scenario, step by step?
- How do you write a prompt that does not invent things?
- What mistakes break this automation?
- Who should build this, and who should skip it?
- What should you automate second?
- Frequently Asked Questions
Quick verdict: should you build this first?
Yes, if leads reach you through a form or inbox and replies currently wait hours. Build this one scenario, run it for two weeks, and only then think about a second one. Skip it for now if almost all your enquiries arrive by phone, or if you get fewer than a handful of leads a month. At that volume, replying by hand is already fast enough, and your time is better spent getting more leads first.
The one rule that decides whether this automation helps or harms you is simple: the AI drafts, the human sends. Keep that rule, and the automation saves you the blank page ten minutes on every lead. Break it, and one wrong auto sent reply will cost you more time in apologies than the automation ever saved.
What is AI automation for a small business, in plain words?
AI automation for a small business means one boring, repeated job is done by software, with a human still in charge of the decision. It does not mean a robot runs your business. In this post, the repeated job is writing the first reply to a new lead. The software does the drafting. You do the judging.
A scenario is just a short chain of steps. Step one sees the new lead. Step two writes a draft. Step three puts that draft in front of you. Step four is you, sending it. If a step cannot be explained in one sentence, the scenario is too complicated for a first build.
What is Make, in plain words?
Make, formerly Integromat, is a tool that connects apps that do not talk to each other. When something happens in one app, such as a new form entry, Make runs a scenario, a short chain of steps, that does something in another app, such as drafting a reply. You build scenarios by dragging steps onto a canvas, not by writing code.
If you are still choosing a platform, read my Make vs Zapier for small business automation comparison first. The short version is this: Make's free tier can run a real multi step workflow, which is why this guide builds on Make. The rest of this post assumes Make, and every step below works on its free plan.
How much does this automation cost on the free tier?
Check the Make pricing page before you build, because prices and limits change. When checked for this post, the free plan was $0 with 1,000 credits a month, 2 active scenarios, and a 15 minute minimum interval for scheduled checks. Make renamed its old operations to credits in August 2025, and for ordinary apps 1 operation equals 1 credit. So one module doing one thing costs about 1 credit.
That last fact decides how you build the trigger, and most guides get it wrong. The Make help page on credits says a trigger module uses 1 credit per run even when it finds no new data. A scenario that polls a form every 15 minutes would run about 96 checks a day, which is close to 2,880 checks in a 30 day month, before a single lead arrives. That alone would eat the whole free tier. So do not poll. Use a webhook trigger instead, which only runs when a lead actually arrives. A webhook is a web address Make gives you for the scenario, which your form calls the moment someone submits it. No submission, no run, no wasted credits.
With a webhook trigger, the maths is honest. One lead uses roughly a handful of credits: one for the trigger, one or two for the AI draft, one to deliver the draft to you, one to log it. Call it 3 to 5 credits per lead. On 1,000 free credits a month, that is roughly 200 to 300 leads a month before the free tier runs out, far more than most small businesses receive. This is arithmetic from the free tier figure above, not a promise from Make.
One cost sits outside Make, and you must check it separately. The AI step you connect has its own account and token rules. Make's help page says a third party AI module uses 1 credit per operation, and you pay your AI provider for tokens on top of that. Check that provider's own pricing page before you connect it, because it is separate from Make's free tier. If your volume ever passes the free tier, you have a lead problem most businesses would love, and the cheapest paid tier is the next step, not a rebuild. Check the paid price on the Make pricing page on the day you decide, because it varies by region and by the credit amount you choose.
How do you build the scenario, step by step?
Build it in this order, and test each step before you add the next one.
- Create the webhook trigger. In Make, create a new scenario and add a webhook module as the first step. Make will give you a webhook address. Put that address in your lead form as its submission address, or connect it through your form tool's webhook setting. Submit a test lead yourself, and check that Make sees it. If the test lead does not arrive, stop here. Do not build the rest on a trigger that does not fire.
- Add the AI draft step. Add an AI module as the second step, and feed it the lead's details: name, what they asked about, and any service or town they mentioned. Use the prompt in the next section, and keep it strict. Run the scenario once with your test lead, and read the draft it produces.
- Add the human approval step. Add a step that sends the draft somewhere you already look, an email to yourself, a row in a spreadsheet, or a message in your team chat, with the lead's details attached. The draft does NOT go to the lead. Nothing is sent until you read it.
- Add the log step. Add a step that writes the lead to a simple sheet with the date, what they asked about, and whether you replied. A month later, that sheet tells you which enquiries turn into jobs, which is the data every "I need more leads" conversation should start from.
- Send it yourself. Read the draft, fix anything that sounds off, which takes ten seconds and usually nothing, and send it from your own email or messaging app. The lead gets a reply that sounds like you, because a human, you, sent it.
Test the whole chain with two or three made up leads before you switch it on for real leads. Check that a lead with a strange question still lands in front of you for approval, and that no draft ever goes straight to the lead.
How do you write a prompt that does not invent things?
The prompt is the whole job, and mine reads roughly like this: "You are writing a first reply for [business name], a [trade] in [town]. The lead asked about [their message]. Write a warm, short reply that answers their actual question, mentions one relevant service, and ends by asking one follow-up question. Do not invent prices, dates or promises. Keep it under 80 words."
The "do not invent" line is not decoration. It is what stops the AI promising a discount you never offered. The same rule runs through my free AI tools for social media content workflow: the AI drafts, I decide, and no fact, price or claim goes out unless a human has verified it.
Three more prompt rules keep the drafts usable. First, name the business, the trade and the town in every prompt, or the draft will sound generic. Second, ask for one follow-up question only, or the reply turns into a questionnaire. Third, cap the length, or the draft will ramble past what a busy lead will read.
What mistakes break this automation?
- Letting the AI send directly. The day an auto sent reply promises something wrong, you will spend more time apologising than the automation ever saved. Draft, approve, send, in that order, always.
- A prompt with no guardrails. An AI told only to "reply nicely" will invent opening hours, prices and availability. The prompt must name what it may not invent.
- Polling instead of a webhook. A 15 minute check burns a credit every time it runs, even with no new lead, and can eat the free tier before a lead arrives. Use a webhook trigger, as explained above.
- Automating a form nobody checks. If your website form sends leads to an inbox nobody opens, this scenario just drafts replies to a pile nobody reads. Fix the inbox first.
- Starting with five automations. Build this one, run it for two weeks, and only then add a second. The review request follow up is the natural next one, and it is the same pattern with a different trigger.
Who should build this, and who should skip it?
Build it if leads reach you through a form or inbox and replies currently wait hours, and if you can spare one evening to set it up and one week to watch the drafts before trusting them. Skip it for now if almost all your enquiries arrive by phone. The phone version of this problem is the voice agent, not an inbox draft. Skip it too if you get fewer than a handful of leads a month. At that volume, replying by hand is already fast enough.
What should you automate second?
Once this scenario has run cleanly for two weeks, the natural second scenario is a review request follow up: after a job is done, a message asks the customer for a review. That is also a fix in my local SEO checklist that helps a small business show up on Google Maps, because fresh reviews help a business show up when nearby customers search. Build it the same way: trigger, draft, human approval, send. Do not build five scenarios in week one. One working scenario beats five half built ones.
Frequently Asked Questions
Does the AI reply to my customers without me?
No, not in this setup, and that is deliberate. The AI writes a draft and waits. You read it, fix it if needed, and send it. Nothing reaches a customer without a human approving it.
How much does this automation cost?
The Make side runs on the free tier's 1,000 credits a month, which covers roughly 200 to 300 leads a month at 3 to 5 credits per lead, as long as you use a webhook trigger and not a 15 minute poll. The AI step you connect has its own account and token rules, so check those on its pricing page before you connect it, because they are separate from Make's.
Do I need to know how to code to build this?
No. Make scenarios are built by choosing modules and filling in fields, and the prompt is plain writing. The skill this needs is writing a clear prompt and keeping a clear approval habit, not code.
What happens when a lead asks something unusual?
The draft will either answer from what the lead wrote or produce something you will catch at the approval step, which is exactly why the approval step exists. Unusual leads are the reason the human stays in the loop, not a reason to avoid the tool.
Why does my free tier run out when I have few leads?
Almost always because the scenario polls on a schedule instead of waiting for a webhook. A trigger check uses a credit even when it finds nothing new, so 15 minute checks alone can use thousands of credits a month. Switch the trigger to a webhook, and the scenario only runs when a lead actually arrives.
Written by Amjad Rasool, SEO Expert and Social Media Expert. I help small businesses grow with content systems, local SEO and AI tools. About Amjad Rasool
Which lead do you reply to slowest, the evening enquiry, the weekend enquiry, or the one that needs a price? Tell me in the comments and I will cover the winner in detail next.
){kind=link}
0 Comments