Five Concrete AI Automations for Your Business
By Anis Hammouche·May 15, 2026
Five jobs come back often in a small business and can be automated with AI: sorting inbound requests, entering invoices, answering frequent questions, turning voice notes into reports, and reshaping one piece of content into several formats. Pick one, on a narrow scope, and measure the result before adding another.
The habit that wastes time is starting with the tool: you compare subscriptions, sign up for three demos, and six weeks later nothing runs. Start instead from the job that eats most of your time or costs you money, then look for what solves it.
For each case I picked a different trade. These people and their situations are made up, so you can see yourself in them. The setup times are my estimates, not measurements. Each one covers the same steps: connect the tools, test on a few real cases, fix, then let it run while you read the first results. I assume existing tools (no custom code), a few hours a week on your side and a narrow scope. If you code your own solution or your data is a mess, plan for more.
| Case | Who it's for | What you get back | My estimate | Risk if the tool gets it wrong |
|---|---|---|---|---|
| Sorting inbound requests | Freelancer, small agency | Reading time you can't bill | 2 to 3 weeks | Medium: a good request ranked too low |
| Entering invoices | Self-employed, small company | Hours of data entry and filing | 1 to 2 weeks | High: a wrong amount or VAT reaches your books |
| Answering frequent questions | Online seller, craftsperson | Answers already written, sent without retyping | 3 to 4 weeks | Medium: a promise you never made |
| Writing reports | Craftsperson, technician, consultant | The evenings spent retyping notes | 1 week for transcription, 2 to 3 for the rest | Medium: a safety note left out |
| Reshaping content | Consultant, freelancer | One long text reused in several formats | 1 week | Low: you read it before publishing |
Freelancer or Small Agency: Sort Inbound Requests
Léa, a freelance web developer, gets quote requests through her site's form, on LinkedIn and by email. Each one sends her digging: who is writing, what the company does, whether the budget holds. Say twenty requests a month at fifteen minutes each: five unbilled hours, taken out of her evenings.
The automation connects the form (or the inbox) to an AI model that:
- reads the request and pulls out the essentials: need, deadline, stated budget;
- looks up what the company makes public (its site, its LinkedIn page);
- sorts the request: handle fast, needs scoping, outside your scope;
- creates the record in your CRM with a three-line summary.
What You Need to Sort Requests
- A web form or a dedicated inbox
- A CRM (HubSpot, Pipedrive, Notion: any one with a connector or an API)
- A workflow tool: n8n, Make or Zapier. If you're torn between the three, read n8n, Make or Zapier: which one to pick when you're starting out
- Optionally, an enrichment tool such as Apollo
Self-Employed or Small Company: Enter Invoices and Receipts
Karim runs a four-person communications agency. Supplier invoices arrive by email, expense receipts as photos, and every month someone enters, files and reconciles them by hand. Count sixty documents a month at five minutes each: five hours, and it's the kind of job you put off until the last day.
AI extraction tools (OCR paired with a language model) can:
- read an invoice and pull out the data: supplier, date, number, amount, VAT;
- match the invoice against a purchase order to spot a gap;
- suggest the right accounting category;
- show you only the cases they didn't recognise with confidence.
What You Need to Enter Invoices
Tools such as Pennylane or Dext include this reading of receipts. If you already have accounting software, check what it offers before adding another tool. As a precaution, read the first invoices one by one: you'll quickly see where the tool slips, on VAT or on duplicates for example.
Electronic invoicing between businesses has been rolling out across France since September 1, 2026. Before choosing a tool, check on the official tax administration page (in French) what applies to your business.
If your main problem is getting clients to pay rather than filing invoices, read Automating Unpaid Invoice Reminders instead.
Online Seller or Craftsperson: Answer First-Level Questions
Sonia sells candles online, alone with a part-time helper. Her customers ask the same questions every time: where is my order, how do I return an item, whether delivery to Belgium is possible. Take forty messages a week, half of them repeating an answer that already exists: twenty retypes, and the evening messages wait until the next day.
A first-level assistant can:
- answer frequent questions from a written base (shipping, returns, product care);
- hand the conversation back to you when the question goes beyond that;
- collect context before the handover: order number, email address;
- reply outside your working hours.
What You Need to Answer Frequent Questions
- A base of answers, even a basic one: an FAQ is enough to start
- A contact channel: email, site chat, or both
- Either the customer service tool you already use and its AI features (Intercom, Freshdesk and Zendesk offer some), or an n8n workflow connected to your FAQ
Craftsperson, Technician, Consultant: Turn Voice Notes into Reports
Marc, a self-employed heating engineer, records a voice note in his van after each job. In the evening he retypes a report for the client: work done, parts replaced, what still needs doing. That time isn't billed anywhere.
The chain has three steps: transcribing the audio into text, shaping it with a model, then filing the result in your tracking tool.
What You Need for Reports
- A transcription tool: Otter.ai, Fireflies, or the transcription built into Teams or Google Meet for meetings
- A model such as ChatGPT or Claude to put the text into shape
- A tracking tool (Notion, Excel, Asana) and a small n8n workflow that sends the report there once you've read it
Here is a starter prompt to adapt:
Here is the transcript of a voice note recorded after a job at a client's.
Turn it into a report with these headings: work done, parts or equipment used, issues found, to schedule (with the date if I gave one).
Add no information that isn't in the transcript. If a heading is empty, write "nothing to report".
Transcript:
[paste the transcript here]
The sentence "add no information" is a precaution: a model sometimes fills in what's missing, the mechanism described in AI Invents Figures and Sources. Read each report before sending it, especially safety notes. An alert on a word like "leak" can help you miss nothing, but it's not for the AI to decide alone what is urgent.
Consultant or Freelancer: Reshape One Piece of Content into Several Formats
Inès, a recruitment consultant, writes one long article a month. She has no time to cut it up into LinkedIn posts, a newsletter and a summary for her site, so it sits on her blog.
AI can reshape an existing text:
- a long article becomes a LinkedIn post, a newsletter extract and a summary for your site page;
- a spec sheet becomes a product description;
- an offer becomes several text variants to test.
What You Need to Reshape Content
A general assistant (ChatGPT, Claude or Gemini) is enough to start, with a prompt template you reuse for each article. You can then plug it into a workflow so the cutting-up starts by itself when you publish. Plan for more time if you want to add automatic validation and publishing.
Before You Plug Anything In: Your Clients' Data
An invoice, a quote request or a voice note holds personal data: names, addresses, phone numbers. Before sending them to a model, check where the tool hosts them, how long it keeps them and whether it reuses them for training. The terms of use say so, and the French data protection authority sets out its expectations in its questions and answers on using a generative AI system (in French). For a document that would embarrass you if it leaked, anonymise it or pick a tool hosted in a setting you control.
How to Choose Where to Start
These five cases aren't worth the same to you. Four questions to sort them:
- Your direct margin: in my view, sorting requests and entering invoices pay off fastest, because they touch what you bill or what you spend.
- How often the job comes back: the more often, the more the gain adds up.
- The risk if it goes wrong: the table at the top rates it on three levels. A misread invoice costs more than a clumsy post, and needs stricter checking.
- The state of your data: a messy CRM or an outdated FAQ needs cleaning before you plug anything in.
My recommendation: one case, a limited scope, a clear measure. Note how long the job takes you today, set up the automation, then let it run for two weeks before comparing. That's my opinion, not a rule: on a narrow scope, two weeks of hindsight is often enough to see whether it holds.
Frequently Asked Questions
Which automation should I start with? The one that eats the most time each week and whose mistakes cost little. For most small businesses, inbound requests or content are good starting points: the risk runs from medium (inbound requests) to low (content) as long as you read before acting.
How long before I see a result? Count the setup time from the table, then two weeks of use before comparing with the time the job used to take. These durations are my estimates: they move with your tools and your data.
The next step is 12 AI automations for your business: where to start concretely, which applies this sorting to a longer list. I write about automation in my newsletter, you can sign up right below.
