Your buyers are findable. In half of Europe, emailing them is not.
We build AI sales agents that research B2B prospects market by market, score them against your own account history, and write outreach in the local language — then route every single message through a compliance gate that knows which countries permit email and which do not. You own the code, the data and the accounts.
Is cold email to businesses legal in the EU?
It depends entirely on the country, and most tools sold in this category do not tell you that. GDPR is uniform across the European Union; the ePrivacy Directive is not. Each member state implemented it separately, and that is what decides whether unsolicited business email is lawful. There are three groups, not two. Consent is required outright in Germany and Austria (double opt-in), Italy, Spain, Poland, Greece, Denmark, Czechia and the Netherlands. A second group turns on who you write to rather than which country: Belgium, Finland and Portugal permit email to a generic company address such as info@ but not to a named individual, and Sweden and Ireland permit it where the message genuinely relates to that person's professional role. Genuinely opt-out are France, Hungary, Luxembourg, Slovenia, Estonia, Croatia and Latvia. A cold email that is lawful in France is unlawful in Germany — and the exposure sits with the sender, not the software vendor.
theagency47 · Updated August 2026Full country-by-country breakdown, with the governing statute for each →
Most AI SDR tools sell you something illegal in half your market.
The pitch is always the same: connect your list, the AI writes personalised emails, volume goes up. It works, in the sense that messages get sent. What it does not do is ask whether those messages are lawful in the country they are landing in — because the vendor is not the one who receives the complaint.
For a business selling into Europe, that is not a technicality. Germany's UWG requires the same consent for a B2B email to a company address as for a promotional email to a consumer. Italy's Garante reads the existing-customer exemption strictly. A single complaint can produce a formal warning, a fine, and a reputational problem that outlasts both.
We build the version of this that does not do that — and in the markets where email is blocked, we build the version that still produces business through the channels that are lawful there.
Six agents, one compliance gate.
Scout
Finds businesses city by city and category by category, then removes anything already in your CRM so an existing customer is never approached as a stranger.
Qualifier
Scores each prospect zero to one hundred on signals specific to your trade. Every score is explainable — you see which signal contributed what, and you can disagree with it.
Contact Resolver
Finds published business contacts, preferring role-based company addresses over personal ones, and records the source URL and legal basis for every single one.
Composer
Writes genuinely personalised outreach in the local language, referring to the specific business rather than merging a name into a template.
Conductor
Routes by channel according to the compliance gate, and manages sequencing, follow-up cadence and volume throttling so deliverability holds.
Analyst
Classifies replies, honours opt-outs immediately and without human involvement, escalates interested prospects to your sales team, and produces the daily report.
The compliance gate is fail-closed.
Every prospect passes through the gate before any message is sent. It classifies the contact and routes it by the rules of that specific market. Four things make it a control rather than a claim:
It fails closed
An unknown regime, an incomplete configuration or an evaluation error results in blocked, never in sent. The cost of one wrong send is disproportionate to the cost of one missed prospect.
There is no bypass
No sending path avoids the gate — not admin, not debug, not a manual override, not a retry queue. A single bypass would make the whole mechanism decorative.
Every decision is logged
Prospect, market, regime applied, outcome, legal basis, timestamp and the version of the rules in force at that moment. Append-only, and built to survive an external review.
You can audit it
You or an adviser you appoint can inspect the decision log and rule versions at any time. A control you cannot audit is not a control.
You lose the channel, not the work.
In Germany, Italy, Austria, Spain and Poland the system does not send cold email. It still performs one hundred percent of the research, qualification and preparation — and the output becomes a telephone briefing with a call script and, critically, recorded evidence of why that specific business would have an objective interest in your specific offer.
That last point is not decoration. German law permits B2B calls on presumed consent, but the courts read it narrowly: a match of industry sector is not enough. The qualification depth that produces that evidence is what makes the call lawful, which is why scoring matters more in Germany than anywhere else.
Other lawful routes the system uses: LinkedIn, physical mail, trade events, and consent-capture programmes that turn your existing customer base into a documented, timestamped permission asset over about a year.
So does an AI make the call?
Yes — automatically, and at a volume no team can staff. The voice agent dials, holds the conversation in the local language, handles interruptions, qualifies against the same scoring the rest of the system uses, books the meeting into your calendar and writes the outcome back to your CRM. It works a list on a schedule and it does not get tired at four in the afternoon.
The gate is consent, not capability. Automated calling is its own legal category in Europe. The ePrivacy Directive treats a system that dials and conducts the conversation without a human as something separate from a person picking up a phone, and requires prior consent for it. Germany is stricter still: an automated calling machine needs express prior consent under UWG §7(2) no. 3 with no business-to-business exception, and the existing-customer exception covers electronic mail only — not calls. So autodialling cold German prospects is not an aggressive growth tactic. It is a fine.
Which is why the consent asset is the actual product. Everything else the system does — discovery, qualification, compliant email in opt-out markets, LinkedIn, trade events, your own customer base — feeds one thing: a documented, timestamped permission list. Once a business is on it, the voice agent can call that business automatically, repeatedly, for as long as the consent stands. Most suppliers will sell you the dialler. The dialler is the easy half.
What runs autonomously from day one: every prospect who has consented, anyone who calls you back, your existing customers where you hold permission, appointment confirmations and post-meeting follow-up. What stays human: the very first approach to a stranger in an opt-in market — until they consent, at which point they move onto the automated track and stay there.
One more thing, since August 2026: Article 50 of the EU AI Act requires that a person be told they are speaking with an AI. Ours says so in its first sentence. Suppliers treat that as a cost. In practice, on a call the recipient has already agreed to receive, it disarms people faster than pretending would.
Four tiers. Published, like everything else here.
Outreach Spark
One market, one buyer segment, opt-out jurisdiction only. The fastest way to see whether this works for your trade before committing to the full pipeline.
Outreach Starter
One market, full pipeline, compliance gate, audit log and daily reporting. The standard build, and the right starting point for most businesses.
Outreach Pro
Two to three markets including opt-in jurisdictions, so the pilot validates both the email path and the telephone path rather than only one.
Enterprise
Multi-market programmes scaling to twenty markets and beyond, where cost per market falls substantially as configuration is reused.
Why the monthly is not simply a price per market
Two different things are being paid for, and it is worth separating them.
The platform is shared. Orchestration, the compliance gate, the audit log, reporting, security patching, model upgrades when providers release new versions — that work is done once and benefits every client running on it. A single-market business does not carry the whole of that fixed cost alone, which is why the monthly for Spark and Starter is a fraction of what a multi-market programme pays.
The rest is specific to your market and does not scale away: quality review of a sample of generated messages, deliverability and domain reputation monitoring, repairing local data sources when they change, recalibrating the scoring model against what actually converted, and monitoring legal developments in your jurisdiction.
At enterprise scale the structure changes to a fixed platform fee plus a fee for each market — because twenty markets consume dedicated capacity rather than a share of it. Monitoring twenty agents is genuinely twenty agents' worth of monitoring; automation reduces that, it does not remove it. Build costs compress steeply as configuration is reused. Operating costs compress far less, and any supplier showing you both curves falling equally steeply should be asked to explain the mechanism.
Third-party services — language models, data sources, contact verification, email infrastructure — are billed directly to you at cost, in accounts held in your name. You receive the supplier invoices. Nothing is marked up and nothing is hidden. Expect roughly €150 to €450 per market per month at working volume, depending on how many prospects you process.
You own it, and it survives us.
Custom code, configurations, prompts, scoring models and all prospect data are yours — fully, irrevocably, perpetually. The repository is yours from day one rather than handed over at the end; we commit into it. Third-party accounts are opened in your name from the start, so there is never a moment when your system lives somewhere you cannot reach.
If you end the maintenance agreement, the system keeps running. The honest caveat, which we would rather state than have you discover: without maintenance it will drift. Models are deprecated, data sources change structure, deliverability rules tighten. Expect three to six months of full performance, then gradual degradation. Any supplier claiming their system runs unattended indefinitely is either inexperienced or not being straight.
Industries where the buyers are findable.
The system is only as good as the evidence your buyers publish about themselves. Eight verticals where that evidence is reliably there — and where we can tell you, before you spend anything, roughly how many qualifying prospects exist in your markets.
Food & beverage exporters
Restaurants publish menus. A menu naming your category is not a hope that they might be interested — it is proof they already buy it, from someone else.
Hospitality suppliers
Linen, tableware, coffee equipment, amenities, kitchen equipment. Tens of thousands of small buyers per country against a sales team of two or three.
Recruitment agencies
Your clients advertise their need publicly and with a date on it. A job posting is the clearest buying signal that exists in any industry.
Industrial components
Factories list processes, tolerances, certifications and installed machinery because their own customers demand it. High account values, so one customer can justify the system.
Medical & dental suppliers
Clinics are registered and advertise their treatments, which tells you what they consume. Healthcare promotion rules sit on top of ePrivacy here, and we scope both.
Packaging & private label
Every brand photographs its own packaging, so you can see the current format before making contact — and catch rebrands while the specification is still open.
Logistics & freight
Exporters state which countries they ship to. Works when you have a genuine specialism; less so if you compete purely on rate, and we will say which you are.
Vertical B2B SaaS
Your buyers belong to a registered profession, so the universe is complete rather than approximate. You could build this yourself — the page explains when that is the better call.
Something else
The test is simple: can your buyers be enumerated, and do they publish evidence of what they need? Describe your market and we will tell you whether it passes — including if it does not.
Questions worth answering before a call.
How is this different from Artisan, 11x, Clay or Apollo?
Those are subscriptions. You rent an account, you send at volume, and the compliance risk is yours. What we build is a system you own, configured to your trade rather than generically, with a gate that blocks what it cannot vouch for and a log you can audit. If we disappear, it keeps running. The trade is a build cost against a subscription you would otherwise pay forever — and a compliance posture those tools do not offer at any price.
How long does it take?
Eight to thirteen weeks for a single market. Development is not the constraint: sending domains need three to six weeks of gradual warm-up before the first real message, and that is calendar time rather than effort. Anyone quoting six weeks for a compliant multi-market build is omitting either the warm-up or the legal work.
What do you need from us?
An export of your existing accounts with order value and frequency, even if untidy, and about thirty minutes with whoever knows which customers are genuinely good rather than merely large. Those two inputs do most of the calibration. We will not ask you to define a minimum order value or a scoring rule in advance — if a supplier requires that before starting, they are asking you to do the analysis and then charging you to implement it.
What is the constraint we should worry about?
Your own sales capacity, almost always — though less than it used to be. The system delivers a qualified prospect and a briefing, and once that prospect has consented the voice agent carries the call itself, which is the part that genuinely scales. What does not scale is the first approach to a business in an opt-in market that has not consented yet: that call is human, in the local language, and there are only so many of them in a week. We would rather activate markets at the pace your team can absorb than deliver volume that decays in a queue.
Tell us the market, and we will tell you what is lawful in it.
No deck. Describe the market and the buyer you are trying to reach, and you will get back a written assessment of what is permitted there, what channel we would use, and what it would cost. If the answer is that this is not worth doing for your trade, we will say so.