Marketing automation for SMEs: with or without AI

An SME that does its marketing repeats the same tasks every week: writing to its contacts, posting on social media, answering the same questions, producing the month’s figures, keeping the catalogue up to date. Marketing automation means handing these tasks over to tools, so that the team keeps its time for what calls for judgement. Since the arrival of language models, it is often presented as a matter of AI. That is only true for some of the tasks.

This guide opens our Digital marketing section. It goes through six tasks, from SEO in a new market to the product catalogue, and says for each what can be automated without AI, what AI really brings and what must stay reviewed by a person. The examples come from sites we operate: Palmora Property, a real estate listings site in six languages, and an online shop that we are getting back on track. For agencies, we cover separately what AI changes about the agency business.

What can an SME automate in its marketing?

Six tasks come up in almost every SME. For each, the table says what a rule does on its own, what AI adds when it is worth the cost, and what a person keeps control of.

The following sections take these tasks one by one, with what we run ourselves.

Do you need AI to automate your marketing?

No, not for most tasks. A follow-up after a quote, a welcome email, the scheduled publication of an article or Monday’s report all follow a rule. A rule costs almost nothing on each run and gives the same result every time; an AI model is paid for on every call, and its answer varies from one run to the next. Adding AI everywhere, without a need for it, multiplies costs and sends projects off course. We cover this split in Automation without AI: when a rule is enough.

AI is justified when you have to read or write free text: translating a page, summarising a contact’s history, extracting a size from a product description, preparing a reply to a question asked in a customer’s own words. Even there, we combine several models with steps that use no AI: each task goes to the cheapest model that does it well, and whatever a rule can check, a rule checks. The method is in The cost of AI in business: tokens, routing and forecasting.

That leaves review.

For a translation, the level of review depends on what the page commits you to, and we go into it in Translating your website with AI: automatic or reviewed?.

How do you open a new market with SEO?

Opening a market starts with being found in its language. Google sets out three rules on its page about managing multi-regional and multilingual sites. It recommends “using different URLs for each language version of a page”, rather than switching the language with a cookie. It asks you not to redirect visitors according to the language you assume they speak. And it reads a page’s language from its visible content: “We don’t use any code-level language information such as lang attributes, or the URL.” A /de/ URL is therefore not enough: the text itself has to be translated.

That is what we do on Palmora Property, built with Astro, in six languages and more than 3,000 URLs. For translation, switching to an AI model divided the cost by 60. That ratio holds for that site: it shows how large the effect of such a trade-off can be, not what you will save. The other saving owes nothing to the model: each text segment is translated only once, only new or modified segments are sent again, and 95% of segments come from a cache without calling the AI again.

Before translating a whole site, choose your markets and write down in advance what will prove demand, then translate the pages that sell first. We go through this choice in six steps in International SEO: choose your markets before translating.

When a market calls for hundreds of pages built on the same template, first read Programmatic SEO: when it works, when it hurts.

What can you automate on social media?

Without AI: scheduling posts in advance, relaying each new article or each new product, sorting incoming messages by topic. These rules fit inside the usual publishing tools, or inside an automation tool such as n8n, which we run on our own server.

With AI: turning one message into several lengths or tones, summarising an article into a post, suggesting a hook. The time saved is real, on one condition: a person chooses the version that goes out. Replies to comments, the brand’s tone in a conversation and any published figure stay human, because a mistake is seen in public and gets shared.

How do you automate emailing without legal risk?

Emailing is the most tightly regulated task. In France, the CNIL, the country’s data protection authority, states on its page about commercial prospecting by email sets out the rules, and they change depending on who is approached:

In both cases, every message must say who is sending it and make it possible to refuse the next ones “by a simple means”.

Once these rules are in place, the automation needs no AI: a welcome email after a sign-up, a follow-up a few days after a quote left without a reply, an unsubscribe handled without delay. These sequences live better in the CRM, where the contacts and their history are. We have used Twenty, an open source CRM, in production since early 2025, and we contribute to the project: our feedback is in Twenty CRM in production: our view, limits included.

AI helps at the margin: summarising a contact’s history before a follow-up written by hand, suggesting two email subject lines to compare. The recipient list, the proof of their agreement and any promise written in a message stay with a person.

To move on to sending, the full procedure, from the right to write to people through to measuring clicks, is in How to run an email campaign, step by step.

Can you automate customer support with AI?

Partly, yes, and partly without AI. The work is split across three levels:

The base of rules already handles many requests, always in the same way. AI, for its part, follows the principle of an assistant that searches your content before answering, presented in Enterprise RAG: an AI that answers from your documents. An obligation comes with it. Article 50 of the European AI Act asks providers of systems intended to interact with people to make sure that they “are informed that they are interacting with an AI system”, unless this is obvious to a reasonably attentive person. The other obligations, GDPR included, are in AI Act and GDPR: what an SME using AI must do.

The third level follows the same logic as the loop of an AI agent, where a human approves every action that commits the company.

Which reporting should you automate to know what pays off?

Reporting is the easiest task to automate without AI, and the one most often put off. A report every Monday, pulled straight from the tools (visits, contact requests, sales, rankings in Google), spares you half a day spent copying figures. Our SEO engine, for example, scans Google’s Search Console every day, without AI, and builds the work queue. Audience measurement can also stay in-house, with a tool such as Matomo, which we run on our own server.

AI can then explain the gaps in plain language: why contact requests fell this week, which page lost rankings. The figures, however, always come from the tools, never from the model, and the decisions remain yours. To link advertising to actual sales, see how to send your CRM sales back to Google Ads. The figures to track, how to connect the tools and the data retention pitfalls are covered in our guide to automated marketing reporting.

How do you automate product catalogue management?

A catalogue is first kept by rules: the import from the supplier, the mandatory attributes checked before publication, a feed that pushes the products to Google Shopping or a marketplace. AI steps in when the information exists but sits in free text.

On Palmora Property, 1,726 entries were rewritten with the help of AI, in 36 batches, with a single failure. And the shop’s catalogue was slimmed down on the basis of sales and Google’s Search Console, not on intuition. The method, from attributes to the feed, is in Google Shopping and marketplaces: preparing your store.

Where to start

Before recommending a tool, we start with an audit: how your marketing tasks flow, which tools are in place and which of them can be connected to one another, what holds the team back, its appetite for digital technology and its training needs. It is an investment that small organisations cannot always afford, and the leverage mostly comes into play beyond a certain team size. That is the role of our digital and AI maturity audit, which ends with a written GO or NO-GO, project by project.

If you are starting on your own, take the task that costs you the most time each week and already follows a rule, such as the follow-up after a quote or Monday’s report: it is the one that pays off fastest, without AI. To hand over advertising, SEO and the tracking of every contact through to the sale in one go, see our digital marketing in one block offer. And for an outside view on your marketing tasks, tell us about your situation.

Frequently asked questions

Sources

1. Managing multi-regional and multilingual sites (new tab)developers.google.com
2. Commercial prospecting by email (new tab)cnil.fr
3. Article 50 of the European AI Act (new tab)ai-act-service-desk.ec.europa.eu