There's a conversation that comes up every time I visit an SMB: "We'd like to get started with AI, but we don't have a technical team. Where do we begin without launching a six-month project?"
The short answer: you don't have to. In 2026, 80% of the automations that really move the needle in an SMB are built with visual tools, without writing a single line of code. What's usually missing isn't the technology—it's the judgement to choose what to automate first.
This post is a curated list of five concrete automations that you can have up and running in your SMB before the week is out. Each one meets three conditions:
- It can be implemented in less than five days of actual work (not calendar days).
- It doesn't require a developer. It can be built with Make, n8n, ChatGPT Teams, Claude Projects, or similar tools.
- It has a measurable impact from day one: hours saved, errors reduced, or faster responses.
A word of warning before we start: "without a developer" doesn't mean "without good judgement". Each of these automations requires business decisions—what to classify, when to escalate to a human, what error tolerance to accept. The easy part is connecting the tools. The part that matters is designing the rules.
Let's dive in.
1. Automatic replies to frequently asked questions (WhatsApp and email)
The problem it solves
In most SMBs, between 40% and 60% of incoming messages are variations of the same five or six questions: opening hours, prices, location, availability, required documents. The team wastes two or three hours a day answering the same things over and over—and every delay in responding is a lead that goes cold.
How it works
An AI agent connected to your WhatsApp Business and your inbox reads each message, detects if it matches a known FAQ, and instantly replies with the correct information. If the query is ambiguous or beyond its scope, it escalates it to a human with the context already summarised.
What you need
A WhatsApp Business API account, an orchestration tool like Make or n8n, and a language model (Claude or GPT) with a simple knowledge base containing your FAQs. You don't need to train anything: you feed it a document with sample questions and answers, and it works.
Why this one is first
Because it's the automation with the best effort-to-impact ratio for a public-facing SMB. Two or three days of setup, visible results on day one.
2. Automatic classification and triage of incoming emails
The problem it solves
An inbox shared by several people becomes a bottleneck. Someone has to open each email, decide whose task it is, label it, and forward it. In a consultancy or professional firm getting 80-150 emails a day, that's easily two hours of pure administrative work daily.
How it works
Each incoming email is processed by an AI classifier that decides three things in seconds: category (invoice, query, complaint, sales, admin…), urgency (today, this week, informational), and owner (who should handle it). It labels the email, moves it to the correct folder, and, if it's urgent, triggers a notification to the person responsible.
What you need
A compatible email account (Gmail, Outlook, standard IMAP), Make or n8n for orchestration, and Claude or GPT for classification. The actual implementation time is one or two days—most of which is spent fine-tuning the categories with real examples from your inbox.
A common mistake
Trying to have the AI reply to the email. Don't. In this automation, the AI's only job is to classify and route. Replying is a later step, and mixing them from the start multiplies the errors.
3. Automatic transcription, summary, and minutes for meetings
The problem it solves
Meeting minutes are often written late, poorly, or not at all. In sales follow-ups, board meetings, or client sessions, what was said is forgotten within 48 hours, and the agreed actions get lost.
How it works
You record the meeting (on Zoom, Meet, or with a simple phone audio recording). You upload it to a shared folder. From there, an automated workflow does three things: transcribes the audio with a modern Speech-to-Text (STT) engine, summarises the content into structured minutes (attendees, topics discussed, agreements, tasks with owner and due date), and sends the minutes to participants via email.
What you need
A transcription service like Deepgram or Whisper, Claude or GPT for the structured summary, and Make/n8n for the glue. It can be set up in a day. The only relevant design decision is the format of the minutes—once agreed upon with the client, it's replicated.
A word of advice
Start with internal meetings before using it with clients. The AI might misunderstand a proper name or a figure; you want to have detected and corrected those patterns before sending out official minutes.
4. Data extraction from PDF invoices and delivery notes
The problem it solves
In consultancies, professional firms, and distribution companies, someone spends hours every week opening PDFs, copying data (invoice number, subtotal, VAT, supplier, date), and pasting it into a spreadsheet or management software. It's a job nobody wants to do, and it's prone to errors.
How it works
When a PDF arrives in a designated folder—Drive, Dropbox, a specific email address—a workflow picks it up, applies OCR if needed, and extracts the relevant fields with a language model. The result is written directly to the spreadsheet or management system. If the extraction confidence is low, the PDF is flagged for human review.
What you need
An OCR service (integrated into most modern orchestrators), Claude Haiku or a similar model for structured extraction, and a target spreadsheet or database. It takes two or three days to get it working with 90% of your usual formats.
Where it usually fails
On unusual formats from small suppliers who use highly customised templates. The right approach isn't to aim for 100% coverage—it's to accept that 5-10% of cases will go to human review and to design the workflow to make that easy.
5. Generation of draft replies for sales enquiries
The problem it solves
A lead asks a question on WhatsApp, your website, or by email, something like "hi, I'd like some information about X". A quality response—personalised, with the right points, with the correct call to action—takes time. And when you take too long, the lead goes cold.
How it works
When a sales enquiry comes in, an AI agent reads the content, identifies which service or product the prospect is asking about, and generates a personalised draft response that is placed directly in the outbox as a draft. The salesperson reviews it, makes adjustments in 30 seconds, and sends it.
The key difference: a draft, not an automatic send
This is important. The AI does not send anything. It only prepares a draft. The final decision remains human, which protects you from errors in tone or content that can be costly in consultative selling.
What you need
Access to your email, Claude or GPT with a well-crafted prompt that includes your catalogue, your pricing, and your tone of voice. This takes one or two days. The real time is spent fine-tuning the prompt with examples of good responses you've sent in the past.
What all five have in common
If you've noticed, there's a pattern:
- None of them aim to replace the human in the decision-making process. They all classify, prepare, or summarise so that the human can make better, faster decisions.
- None require complex integrations with your ERP or CRM. They work with the tools you already use: your email, your WhatsApp, your Drive folders.
- All have a clear "fall-back mode": if the AI isn't sure, it escalates to a human. This makes them robust from day one.
AI automation in an SMB isn't magic. It's operational engineering with modern tools. And in 2026, the tools are good enough for a consultant with good judgement—not a development team—to get them working in days.
Where to start
If I had to recommend an order for an SMB starting from scratch, it would be this:
- Start with #1 or #2 (automatic replies or email classification). They are the most visible to the team and generate the "wow" factor that unlocks the budget to continue.
- Then move on to #3 or #5 (meeting minutes or sales drafts). Both are inexpensive to implement and save time immediately.
- Leave #4 for when you're up and running (invoice extraction). It's very useful but requires more iteration to reach 90% accuracy.
Don't try to implement all five at once. Most SMBs that fail with AI don't fail because they choose the wrong tool—they fail because they try to do too much in parallel and don't finish any of them properly.
A final thought
These five automations are just the starting point. The real value of AI in an SMB comes when they are combined: when the classified email feeds the sales draft, when the extracted invoice triggers an alert to the manager, when the meeting minutes update the CRM.
But to get there, you have to start with the simple things. And the simple things, today, are more powerful than most SMBs imagine.
Want to get started but don't know where?
At Isenzo-IA, we work with SMBs and professional firms in the Baix Llobregat region and surrounding area to design and implement these types of automations. No endless meetings and no technical jargon—just a conversation, a diagnosis, and a realistic plan.
If any of this has resonated with you and you'd like to explore it with no obligation, write to me at rafael.vidal@isenzo-ia.com or call me on +34 692 439 218. I'd rather have an honest 20-minute chat than have you fill out a form.
— Rafa Vidal, founder of Isenzo-IA
