What an automated outreach can do today
Let's start with the numbers, because they are the hardest to argue with. A cold email platform aggregated the data of 53.1 million emails sent and 60,000 sequences from the first half of 2026:
- average reply rate: 3.7%, average open rate: 21%;
- bounces: 1.53% on a verified list, 2.55% on an unverified one;
- follow-ups produced 44% of the positive replies, and the first follow-up alone produced 26% of the positive replies;
- the best campaigns brought 2–3 meetings per 100 cold emails.
This is vendor data, so it comes from a self-selected sample: the people who send with a tool like this are the ones who take it seriously to begin with. The order of magnitude is still worth accepting, because it is measurement on their own platform, not an estimate.
And now the number that turns the whole thing on its head: campaigns sent to target groups under 200 people produced roughly twice the reply rate of those sent to groups over 500. In the platform's own breakdown that is 15–20% against 8% — far above the 3.7% average, so that breakdown clearly covers the better half of the campaigns. The ratio is the usable message, not the levels.
So the promise of machine outbound — "reach ten times as many people for the same work" — ruins precisely the thing that makes it work. A small, hand-picked list is not a constraint, it is the active ingredient.
Where it breaks first: the sending domain
This is the most expensive one, because it is not easily reversible.
The large mailbox providers measure complaint rates. At Gmail the target is below 0.10%, and 0.30% should never be reached. Enforcement at the same provider got stricter from November 2025: non-compliant traffic runs into temporary and then permanent rejection.
If an automated campaign goes out cold to several thousand people, the complaint rate crosses the threshold fast. What you lose with it is not the campaign but the deliverability of your company email: your quotes, your invoices, your letters to customers. That is why the trade sends from a separate domain — but that treats the symptom, not the cause.
The technical conditions of deliverability are covered by the requirements of newsletter sending post, point by point.
The legal line: where it breaks
Here is the other break point, and a surprising number of people have it wrong. The EU foundation is shared: under Article 13 of the ePrivacy Directive (2002/58/EC), direct marketing by electronic means may not be sent without prior consent. The exception is the existing customer you are writing to about your own similar product — and even there, a free opt-out has to be offered at collection and in every single message.
Article 13(5) of the same directive, however, imposes the consent rule expressly on natural person subscribers only; the protection of everyone else, companies included, it leaves to the member state. Which is why the position is not the same in two countries — and why the most expensive mistake is carrying the Hungarian rule over to a Spanish list, or the other way round. Open the one that applies to you:
Hungary — Section 6 of the advertising act covers natural persons only
Sending cold advertising to a natural person by electronic means is not allowed without prior, unambiguous and express consent. This is what Section 6 of the act on business advertising activity states. The same section records that consent may be withdrawn at any time, without giving reasons and free of charge — and also that addressed postal advertising mail may be sent without prior consent, but only if the recipient can prohibit further mailings at any time. The legal position of the postal and of the electronic channel is therefore not the same.
Outreach sent to a company as the recipient is not covered by that rule. This does not mean the field is clear, though: the work email address of a named employee is personal data, so the data protection regulation applies all the same. Between an info@ address and a kovacs.anna@ address the difference is not a marketing one but a legal one.
On the electronic channel the acting authority is the media and communications authority, the NMHH, solely on the request of a natural person, within one year of the advertisement being published. Its competence does not extend to a foreign sender or to an advertisement in a foreign language.
Spain — companies are protected too, and the data protection authority issues the fines
The Spanish rule is Article 21 of Law 34/2002 on information society services, the LSSI: it prohibits sending advertising or promotional communications by email or an equivalent electronic means where they were not previously requested or expressly authorised by the recipients. Enforcement practice treats SMS as covered as well.
The decisive difference from the Hungarian position sits in the statutory definition: in the annex to the LSSI, a "recipient" is a natural or legal person. The Spanish prohibition therefore reaches the company address as well, regardless of whether personal data is involved. An info@ address is not a loophole here.
The existing-customer exception in Article 21(2) is narrower than most people assume: it requires a prior contractual relationship, contact details obtained lawfully, and a similar product of your own company. On top of that, every message has to carry a valid email address for opting out — without it, sending is prohibited in itself.
The acting authority is not the telecoms regulator but the data protection authority: Article 43 of the LSSI assigns the Article 21 and 22 infringements to it expressly. Fines run up to 30,000 euros in the minor band and between 30,001 and 150,000 euros in the serious one. And there is one duty with no Hungarian counterpart: under Article 23(4) of the LOPDGDD, advertising exclusion lists have to be checked before direct marketing — unless the recipient gave consent to you specifically.
Four cases that show what counts
- Hungarian regulatory case, 2008. A company sent electronic advertisements without prior consent over two days. The authority found an infringement of the advertising act and of the act on electronic commerce, with a fine of 200,000 forint. The amount looks laughable today — the lesson does not.
- Hungarian data protection case, 2024. Between July and November 2019 the complainant asked five times to be unsubscribed; the company confirmed the deletion in October 2019, and yet the complainant kept receiving newsletters until May 2021. The authority found an infringement of the rules on consent, on handling requests and on data protection by design. Unsubscribing is not a request but an obligation — and you have to be able to prove that you performed it.
- Spanish regulatory case, 2024. An optician chain sent advertising SMS messages without prior consent and with no way to unsubscribe shown, including to a recipient whose unsubscribe request it had previously confirmed. The data protection authority set the fine at 10,000 euros; the company ended up paying 6,000, having used the two separate 20% reductions for admitting liability and paying voluntarily. The electronic channel is not only email — and a missing unsubscribe address is an infringement in itself.
- French case, 2023. A data broker was fined 75,000 euros because the consent collected on prize draw and product testing forms was not valid: the accept button stood out visually, the small print pushed towards accepting the data transfer, and it was not clear that the data would go to partners for the purpose of commercial outreach. If you send to a purchased list, this is your risk too.
What is new from August 2026: the bot has to introduce itself
The transparency rule of the EU AI act has applied since 2 August 2026, and it bears directly on outbound.
The substance: if an AI system interacts directly with a natural person, that person has to be told they are talking to a machine. The disclosure has to happen "in a clear and distinguishable manner, at the latest at the time of the first interaction" — and with every new person, not once at deployment. The "it is obvious, so there is no need to say it" exception has to be read narrowly.
The European Commission's guidance expressly offers as an example the case where an AI voice agent makes sales calls or sends emails on behalf of a business. So this is not a grey zone.
What it does not prescribe: there is no separate duty to mark that the text of a marketing email was written by AI. The labelling duty for text covers content that informs the public on matters of public interest, not commercial mail.
The fine ceiling for this obligation is 15 million euros or 3% of global annual turnover — in Hungary the implementing government decree fixes it at 5.7 billion forint. At a small company that will obviously not be the number, but the order of magnitude shows where the legislator placed the question.
"The bot said it" is not a defence
There is a 2024 Canadian decision that states this precisely. An airline's chatbot wrongly told a customer that a discount could be claimed retroactively. The company's defence was that the chatbot is a separate legal entity, responsible for its own actions.
The tribunal rejected that: the chatbot is part of the company's website, and the company is responsible for everything that appears on its website — whether it is static text or a chatbot. The amount awarded was small, the principle is not: no AI-specific legislation was needed for it, ordinary liability law was enough.
That is the most important sentence to hold on to before deploying an AI salesperson. Whatever the bot asserts about your prices, your deadlines and your guarantee, you asserted it.
What it ruins irreversibly
Four things, in order from the most expensive:
- The reputation of the sending domain. Rebuilding it can take months, and in the meantime your ordinary business email stutters too.
- The brand. Someone whose first impression of you comes cold, in machine-written copy, with the wrong name on it, is someone you will not get to address a second time.
- Your database. A mass unsubscribe cannot be undone, and the withdrawal has to be honoured.
- Your own team's judgement. If the metric is the number of emails sent, the whole process will optimise for that.
What does work, though
- A small list, with real research. According to the platform data this is not a moral question but an efficiency one: campaigns under 200 people brought twice the reply rate.
- AI in the research and the drafting, not in the sending. Gathering company data, finding a relevant reason, a first draft — that is the machine's job. "Who do we send it to" and "this can go out" belong to the human.
- Follow-ups, but few of them. 44% of the positive replies come from there; the first follow-up gives the larger part.
- A verified list. Bounces run at 1.53% on a verified list and 2.55% on an unverified one — and that eats your domain reputation directly.
- By post the rule is different. In Hungary, if your offer can carry it, the legal position of addressed advertising mail is more favourable.
On the finer points of how you address people — how much of what you know you should show — the limit of personalization post goes into detail.
Common misconceptions
- "You may send cold to a company address." The separate rule in the advertising act does speak about the natural person, but the work address of a named employee is personal data, so the data protection rules are alive anyway.
- "More emails, more chances." The measured data says a bigger list gives a smaller reply rate, and meanwhile you are gambling with your domain.
- "We will sort out the unsubscribes later." The regulatory case was about exactly that: five requests, a confirmed deletion, and emails a year and a half later.
- "No need to say it is a bot, it is obvious anyway." The exception has to be read narrowly, and the official guidance brings up the sales call as its very example.
- "If the bot says something, we are not liable for it." That is precisely the defence a 2024 decision rejected.
What to take away
Automated outbound does not break because the copy is weak. It breaks because the volume itself is the mistake: it lowers the reply rate, burns the domain, and opens a legal surface where a missing consent and an unperformed unsubscribe can be proven. Use AI to address fewer people better — and say so when a machine is talking.