AI Automation for Small Businesses: 5 Real Examples With Costs

Most AI advice for small businesses is either "AI will replace your whole team" or "it's all hype, ignore it." Both are wrong. There's a narrow band of automations that reliably pay for themselves within months, and a much wider band that quietly burns money.

I build these systems for a living, and I run them in my own products. Here are the five patterns I keep coming back to, what they actually cost, and how to pick your first one.

1. Document and receipt processing

This is the most boring one, and it's usually the best return. Invoices, receipts, delivery notes, application forms - anything a human currently reads and retypes into a system.

Modern vision models read messy real-world documents better than the OCR tools of five years ago by a wide margin. In my own iOS app, MyHomeStock, AI receipt scanning captures a receipt in under 5 seconds - line items, prices, quantities - straight into inventory. The same pattern works for accounting inboxes, supplier invoices, or expense reports.

Realistic cost: a focused pipeline (email or upload in, structured data out, human review for low-confidence cases) starts around €1,500 and takes one to three weeks. Running costs are typically a few cents per document.

If someone on your team spends two hours a day retyping documents, this pays for itself in the first quarter.

2. Lead qualification and routing

Not every enquiry deserves the same response. Some are ready to buy, some are tyre-kickers, some are spam. An AI layer can score each incoming lead against your criteria, enrich it with context, and route it: hot leads get a fast personal reply, everything else gets a polite automated one.

I run this on my own site. Every contact form submission gets AI lead scoring, and the promising ones flow into an AI-assisted proposal drafting step - so by the time I sit down to reply, half the work is done. It's a big part of why I can respond to serious enquiries within hours instead of days.

Realistic cost: €1,500 to €3,000 depending on how many systems it touches (forms, CRM, email). Two to three weeks of work. The value scales directly with your lead volume and your response time before automation.

3. Customer-question chatbots grounded in your own data

The keyword here is grounded. A generic chatbot bolted onto your site is worse than nothing - it hallucinates answers and annoys people. A chatbot that only answers from your actual documents (pricing, policies, product specs, FAQs) using retrieval-augmented generation is a different animal.

Done properly, it deflects the repetitive 60-70 percent of questions - "do you ship to X", "what's included in Y" - and hands the rest to a human with full context. The build work is mostly in the unglamorous parts: cleaning your source content, setting refusal behaviour for questions it shouldn't answer, and logging everything so you can see where it fails.

Realistic cost: €2,000 to €5,000 for a production version with your content indexed, escalation paths, and an admin view of conversations. Budget a few hours a month to keep the source content current - a chatbot grounded in stale data is confidently wrong.

4. Content operations

Notice I said content operations, not content generation. Fully AI-written content is a race to the bottom, and search engines are getting better at ignoring it.

Where AI genuinely helps is the machinery around content: turning one long piece into social posts and a newsletter section, generating metadata and internal-link suggestions, translating and localising, maintaining consistency across hundreds of pages. I run DevOpsNess, a 400+ article engineering publication, solo - that's only possible because the repetitive parts of the pipeline are automated while the actual thinking stays human.

Realistic cost: €1,500 to €4,000 for a pipeline tailored to your workflow. Worth it if you publish regularly across multiple channels; not worth it if you post twice a month.

5. Internal report generation

Every business has someone who spends Friday afternoon pulling numbers from three systems into a document nobody formats consistently. Weekly sales summaries, project status updates, inventory reports.

An automation that queries your actual data, has AI write the narrative summary ("orders up 12 percent, driven mostly by X"), and delivers it to Slack or email every Monday morning is one of the cheapest wins on this list - because the data is already structured and the output format is predictable.

Realistic cost: often €1,500 to €2,500, one to two weeks. The hard part isn't the AI, it's getting clean access to your data sources.

Where AI is not worth it

A few honest exclusions. Don't automate a process you do fewer than a handful of times a month - the maintenance costs more than the manual work. Don't put AI anywhere a wrong answer is expensive and unreviewable: legal commitments, medical advice, final pricing decisions. Don't buy AI to fix a process that's broken for human reasons - automating chaos gives you faster chaos.

And skip anything pitched as "AI strategy" without a specific workflow attached. If nobody can name the task being automated, there's no project.

How to pick a first project

Use three filters. High volume: the task happens daily or weekly, not quarterly. Low stakes per item: a single mistake is cheap and a human can catch it in review. Clear before/after: you can measure hours saved or response time improved within a month.

Document processing and report generation pass all three for most businesses, which is why I usually recommend starting there. Chatbots and lead routing come next once you've built trust in how these systems behave.

My AI automation projects start at €1,500, and full AI/SaaS platforms from €5,000 - all fixed-price, with a free mockup before you commit anything. Full details are on the pricing page.

If you're wondering whether a specific workflow in your business is automatable, describe it to me and I'll tell you straight - including if the honest answer is "don't bother."