Friday, 11:20 AM
On the owner's desk sits a note that says "call back yesterday's lead." On the phone, three Google reviews are waiting without a reply. In the inbox, buried among eighty emails, is a reminder from a supplier. And the spreadsheet of "who owes us money" hasn't been opened in three weeks.
None of these tasks is hard. Each one takes fifteen minutes. And that is exactly why none of them gets done.
This is the "I'll deal with it later" drawer. And it is exactly where AI automations earn their keep.
Most people looking for tasks to automate in a business start with the tool: which software, which connection, which flowchart. We suggest starting with the drawer. This guide covers ten small, recurring tasks, and for each one four things: how it looks today, what an AI agent does with it, what stays with you, and how you know it actually works. At the end there is a checklist to help you pick three to start with, and a calculator where you enter your own numbers.
In short
- AI automations are worth the effort for small, recurring, well-defined work: lead follow-up, inbox sorting, a weekly summary, reminders, draft replies, website checks.
- In every task the agent prepares and carries out, and you approve anything that goes out or costs money.
- Don't start with ten. Pick three: one that hurts, one that is easy to check, and one that connects to a system you already have.
- Every task needs a simple metric you can check yourself, otherwise there is no way to know whether it works.
What makes a task a good candidate for AI automation?
Think of a new employee on their first day. You don't hand them the keys to the safe. You first give them a drawer of small things: "Go through this list and prepare a draft for me." A task like that suits an agent too. Four tests:
- It repeats. It happens every week or every month, not once a year.
- It has a definition of "good." You can explain in one sentence what a good result looks like, or show an example.
- The result can be checked fast. A minute of reading, not an hour of investigation.
- A mistake is cheap. A draft with an error just needs fixing. A message sent to a customer without you seeing it is another matter.
A task that fits
For example: a weekly summary
- Repeats in the same structure
- The result is read and checked in a minute
- Nothing goes out without approval
- Easy to compare with work you did yourself
A task not to start with
For example: negotiating with a customer
- Every case is different
- Success depends on judgment and relationships
- A mistake can hurt a customer or cost money
- Hard to check what "good" is
The agent itself is not a type of software. It is a language model connected to the business's tools that does work in them according to permissions. All of that is explained in AI agent for business: what it really does, and if you wonder how it differs from a classic automation tool, we covered that in business automation: a train, a triage station, or a taxi.
Ten tasks to automate in a business, by area
The scenes on each card are made up and are there for illustration only. Which connections are actually possible depends on the tools you have and the permissions of each system, so every task starts by checking what can be connected.
Customers and sales
1. Lead follow-up
- Today: A lead comes in through the form on Thursday evening. The first reply goes out on Sunday, if someone remembers.
- The agent: Logs every inquiry in a spreadsheet or CRM, drafts a first reply based on the topic, and reminds you on Tuesday whom you haven't answered yet.
- With you: A message in the business's name goes out with your approval, and every conversation with a hot lead is yours.
- How to measure: How long it takes from inquiry to first reply, and how many inquiries are left with no follow-up after a week.
2. Draft replies to reviews
- Today: A three-star review waits for two weeks, because nobody ever finds time to word a reply.
- The agent: Drafts a reply to each review in your tone, and flags reviews that mention facts that need checking.
- With you: You publish. A reply to a harsh review, especially one with accusations, is written or checked by you.
- How to measure: How many reviews are left without a reply, and how many drafts were published almost as written.
3. Inbox sorting
- Today: A pile of emails waits every morning. Most are newsletters, and one important one is drowned in the middle.
- The agent: Labels urgent, service, suppliers and noise, and prepares a draft for questions that keep coming back.
- With you: Sending, and anything involving money or a commitment. The agent deletes nothing.
- How to measure: How many urgent emails were missed (the target is zero), and how long a pass through the inbox takes you now.
Store and website
4. Uploading products
- Today: A shipment arrived, and the photos are on the phone. Three products take a whole evening.
- The agent: Uploads products as drafts with a description, category and photos, and flags what is missing, such as an unclear price.
- With you: The price and the publishing. No product goes live without approval.
- How to measure: How many products went up as a valid draft with no fixes, and how much time passed between the shipment arriving and publishing.
5. Website health checks
- Today: You find out the form doesn't send only when a customer calls to tell you.
- The agent: Checks the form, the payment page, plugins waiting for an update and the WordPress Site Health screen, and reports in a message.
- With you: A significant update happens with your approval and only after a backup.
- How to measure: How much time passes between a failure and the moment you know about it.
6. Checking supplier prices
- Today: A supplier sent a new price list as a PDF. Who sits down and compares it line by line?
- The agent: Compares it with the previous list, flags what changed and by how much, and shows which prices on the site are worth checking.
- With you: Every price change on the site or with a supplier.
- How to measure: In a spot check: how many of the flagged changes were correct, and how many changes were found too late.
Content and reports
7. Weekly analytics summary
- Today: You open Google Analytics when you remember, glance at a graph and close it with no conclusion.
- The agent: Every Sunday morning sends one page: what changed, where visitors came from, what stands out, and one question or suggestion.
- With you: The decision whether to act. The agent doesn't change a campaign or a budget on its own.
- How to measure: Did you read it, and did it lead to at least one decision a month.
8. Repurposing content
- Today: You wrote a good article. It sits on the site and nobody comes across it.
- The agent: Turns one article into a post, a newsletter and a few short quotes, and keeps the facts as they are in the source.
- With you: Approval and publishing, on every channel.
- How to measure: How many pieces came out of each source, and how many were approved without a full rewrite.
Finances and your time
9. Invoice and receipt reminders
- Today: A spreadsheet of who owes you and who hasn't sent a receipt, and somebody has to remember to go through it.
- The agent: Goes through the list, flags invoices past their due date, drafts a polite reminder and asks for missing receipts.
- With you: Sending the reminder and any decision about collection. Bookkeeping and tax reporting stay with your accountant.
- How to measure: How many invoices are overdue at the start of the month versus the end, and how many reminders needed correcting.
10. Meeting prep
- Today: Five minutes before a meeting, you search your phone for the client's last email.
- The agent: The evening before, prepares one page: who the client is, what was agreed, what is open, and three questions to ask.
- With you: The meeting itself and everything said in it.
- How to measure: Did you use the page, and how many times did you have to look something up mid-meeting.
Three notes on the cards. Google reviews: according to the Google Business Profile help center, a verified business owner can respond to reviews, so the agent prepares the draft and you publish it. Website checks: WordPress has a built-in screen, Site Health, under "Tools," that shows critical issues and recommended improvements. That is exactly the kind of information an agent can read and summarize for you in a message. You'll find the regular maintenance itself in managing a WordPress site with an AI agent. Finances: a payment reminder is a draft. Anything related to reporting, tax and bookkeeping stays with your accountant.
Which tasks to start with? An effort-versus-value matrix
Not every task is equally easy to start with. The sorting below is our estimate, not a measurement, and it varies from business to business: the more open your tools are to connections, the further left a task moves.
Setup effort versus value, by our estimate
Start here: low effort, clear value
- Weekly analytics summary
- Draft replies to reviews
- Meeting prep
- Repurposing content
Worth setting up: needs a system connection
- Lead follow-up
- Uploading products
- Website health checks
- Invoice and receipt reminders
Later: needs more trust or order
- Inbox sorting, because it means access to everything private
- Checking supplier prices, because every supplier's files look different
Why is the analytics summary in the green column? Because it changes nothing. It only reads and reports. There aren't many ways to hurt a business with a summary. Inbox sorting, by contrast, is perhaps the most tempting task on the list, and still we put it last: it touches private conversations, customers and contracts. You start it after the agent has proven itself somewhere less sensitive.
How do you choose your first three tasks?
Mark three tasks from the list, no more. The counter counts, and if you reached five, go back: the goal is to decide.
My first three tasks
And if you're not sure, here is a shortcut. Start with the pain you feel most strongly, not the task that sounds most impressive:
| If your pain is... | Start with... | Then add |
|---|---|---|
| Inquiries falling between the cracks | Lead follow-up | Draft replies to reviews |
| "I have no time for the website" | Uploading products | Website health checks |
| "I don't know what's going on" | Weekly analytics summary | Repurposing content |
| Money that is delayed | Invoice reminders | Checking supplier prices |
| A flooded inbox | Meeting prep and content repurposing, and only then inbox sorting | Inbox sorting |
We recommend at least one "read-only" task, like the weekly summary, next to one task that prepares a deliverable, like products as drafts. Read-only gives you confidence from day one, and a ready deliverable gives you the value you can feel.
How many hours are yours? A calculator with your numbers
We don't promise a savings figure, because we don't know your business. Here you enter how long each task takes today, and the calculator just adds up what you marked. This is not the saving: you will still check and approve. But it is an honest starting point for a conversation about what to start with.
How many hours a week does the task take you today?
0 hours a week
0 hours a month
Mark the task, and move the slider to match the time it takes you. If for everything you marked your role shifts from "doing" to "approving," that is a good sign. If a task takes only a few minutes a week, ask whether it is really the place to start.
How do you know it works? Three metrics you can check yourself
The most common mistake is to feel that something works without checking. Each card above has a metric specific to its task, and beyond it there are three general metrics that suit any task:
| Metric | What to check | Warning sign |
|---|---|---|
| Share of drafts approved almost as written | Each week, count how many results you approved without rewriting | If you rewrite most of them, the agent is missing examples or instructions |
| Approval time | How many minutes it takes you to go over a result | If it takes almost as long as doing it yourself, the task doesn't fit or isn't well defined |
| Things that were missed | Inquiries, emails or invoices that weren't handled | Every such case is a reason to stop and understand why |
Start with a short trial period, checking every result closely. When the metrics are stable, you can reduce the checking. If they aren't stable, that is not a failure: it is information about what to define again. You'll find the full starting steps in building an AI agent: step by step.
Three mistakes we see when people start with AI automations
Choosing the most impressive task instead of the most painful one. A task that looks great in a demo but doesn't bother you won't become a habit. The most useful agent is the one that handles what you put off every week.
Handing over everything at once. Ten new tasks, with no idea which works and which doesn't. Three tasks with clear metrics teach you more than ten that are hard to follow.
Forgetting the approval. The temptation is to let the agent send, publish and change things without stopping. In Platin Digital's view, publishing, messages to customers, price changes and budget changes stay with you. Speed isn't worth paying for with a customer's trust. We write about this at length in permissions, approvals and security.
By the way, classic automation tools like n8n do a good job on fixed steps, and if you've already built something that works on them, you don't have to take it apart. We just think that in a small business, where tasks keep changing, it is easier to manage an agent by message than to maintain flowcharts. The full comparison is in n8n or an AI agent.
Why now
Google searches in Israel for AI automations are still modest, but they are rising, and whoever starts now is learning on a small business, not on a giant enterprise system.
Year-over-year change in Google searches, Israel
Source: Google Ads Keyword Planner, Israel, September 2025 – August 2026, compared with the previous year. Google's rounded estimates.
This is not a promise that whoever starts now will profit. It is a sign that the question "how do we bring this into the business" is already being asked, and that you can start small, with three tasks, without betting everything.
How to start this week
- Mark three tasks in the checklist above, and write down each one's metric.
- Tell us, or whoever sets up your agent, which tools you have: website, store, calendar, email. Check what can be connected and with what minimal permissions.
- Collect three "good" examples for each task: a reply you liked, a summary that helped, a product description that worked.
- Define what waits for approval. Everything that goes out or touches money.
- Test for two to three weeks, and only then add another task.
If you'd like to do this with someone who has already set up agents for businesses, we're happy to help. Platin Digital is an OpenAI Select Partner, and we build agents with whichever models and tools fit each business best, from OpenAI, Anthropic, Google and others. You can start with a short conversation and choose the three tasks together.
Frequently asked questions
What are AI automations?
AI automations are recurring business tasks carried out with the help of artificial intelligence, so you don't have to do them by hand every time. Unlike classic automation of the "when X happens, do Y" kind, an AI agent understands free-form text, prepares a result and performs actions in the systems connected to it, and you approve before anything goes out.
Which tasks are best to start with?
Recurring tasks with a clear definition of "good," that you can check in a minute, and where a mistake won't be costly. Examples: a weekly analytics summary, draft replies to reviews, meeting prep and lead follow-up. It's best to leave the more sensitive tasks, like inbox sorting, for later.
Can the agent send messages to customers without my seeing them?
That depends on the setup, and with us the default is no. A message to a customer, publishing and a budget change wait for your approval. You can gradually widen specific tasks, after you see the results are stable.
How much time will it save me?
There's no way to know without knowing the business, so we don't give a number. Use the calculator above to see how long the tasks take you today, and after a few weeks measure how long they take with an agent, including approval time.
Do I need special tools to get started?
Usually not. The agent connects to tools you already have, like the website, the spreadsheet, the calendar and email, subject to the connections and interfaces each system allows. The first step is to check what can be connected, and the first tasks are chosen accordingly.
What is the difference between this and regular ChatGPT?
Regular ChatGPT drafts text that you copy and upload yourself. An agent is connected to the business's systems and does the work in them, leaving the approval to you. The full explanation is in the article on an AI agent for business.
