16 Principles for AI-Driven B2B Outbound

Not legal advice. This article describes working principles and places them in a legal context. It states the law in general terms, is not legal advice, does not replace an assessment of your specific case, and creates no attorney-client relationship. Use at your own risk. CegTec is not a law firm. The per-channel requirements in detail are in the channel matrix for commercial outreach.

The central rule

Anyone running outbound with AI agents eventually faces one question that decides everything else. The wrong version is: “how do we get the advertising through somehow?” The right one is:

Earn the right to sell. How do we first create a legitimate, relevant business conversation?

This is not a politeness formula. It is the decision that determines whether a system does research and earns relevance from it — or buys reach and hopes enough of it sticks. The 16 principles below are that single rule, spelled out.

In several places they are stricter than the law. That is deliberate: the law is the floor, not the standard. Where a principle has a legal anchor, it is named.

In one place it runs the other way — the law is stricter than the principles suggest. Anyone assuming a research first touch is legally safer than a sales email is mistaken. Why is set out below under The legal catch in principle 1; if you take one thing from this page, take that.

The 16 principles

A. Intent — is the motive honest?

#	Principle	What it means concretely
1	Research first, sell later	The first interaction serves genuine information gathering, not the sale.
2	Truth over pretext	No pretend “research” with a sales pitch behind it.
10	No deception	No fake identities, no invented research, no manufactured triggers, no pretended relationships.
11	AI transparency where required	AI must not be used to obscure the sender or the purpose of the communication.

This group has one hard test: would you use the answer even if no business came of it? If not, it is not research, it is a hook — and then principle 2 applies.

Legally, 2, 10 and 11 anchor in two places, and both apply in B2B as well:

§5a(4) UWG: acting unfairly includes “failing to make apparent the commercial purpose of a commercial act, where it is not directly apparent from the circumstances, and where that failure is liable to cause the consumer or other market participant to take a transactional decision they would not otherwise have taken.” That is principle 2 as statutory text — and it applies regardless of whether a human or a model wrote it.

§7(2) no. 3 UWG prohibits advertising by a message “in which the identity of the sender on whose behalf the message is transmitted is concealed or disguised.” That is principle 10 as statutory text.

For AI systems interacting directly with natural persons, Art. 50(1) of Regulation (EU) 2024/1689 (the AI Act) adds: providers “shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system”, unless that is obvious. It has applied since 2 August 2026. No blanket duty to label AI-written sales copy follows from it — the provision addresses interaction with the system, not authorship of a text.

B. Data — what may be collected at all?

#	Principle	What it means concretely
3	Business context only	Role, company, market, technology, hiring, expansion. No private information.
4	Minimal data	Collect only what the specific purpose requires.
5	Source everything	Every relevant fact carries a source and a timestamp.
6	Purpose limitation	Use data only for the purpose it was collected for.

Principles 4 and 6 are almost verbatim Art. 5(1)(c) and (b) GDPR — data minimisation and purpose limitation. Principle 3 is the practical translation into sales work: the purpose is a business conversation, so the permissible context is the business one.

Principle 5 is the operationally most important of the group, and it has the same effect legally and substantively. Legally: where data was not collected from the data subject, the information duties under Art. 14 GDPR apply, and they cannot be met without documented provenance. Substantively: inside an AI pipeline, a claim without a source is indistinguishable from a hallucination. A signal without a timestamp is worse — it looks fresh and is not.

One note for implementation: source and date belong on the individual field, not on the record. Knowing only that a lead was “enriched in March” tells you nothing about whether the job posting the message cites is still open today.

C. Outreach — may and should this person be contacted?

#	Principle	What it means concretely
7	Channel-specific compliance	Email, phone, LinkedIn and postal mail are each assessed and handled separately.
8	No mass blast	ICP and concrete business context come before contact.
9	Human relevance	Every message must be plausibly relevant to this person in this role.

Principle 7 is where most systems fail — not from bad intent, but because a channel switch in a tool looks like a formatting choice. It is a legal choice: requirements differ substantially per channel, from consent-free to prior express consent. The full overview with statutory citations is in the channel matrix; how to build the consent you need instead of working around it is in building consent instead of risking cold outreach.

Principle 9 has a usable self-test: would this message work unchanged for the neighbouring role at the same company? If yes, it is not relevant but generic — and then principle 8 is breached too, however tightly the ICP was defined.

D. The exit — how does it stop?

#	Principle	What it means concretely
12	Easy opt-out	A “not interested” ends the communication immediately.
13	Global suppression	Opt-outs apply system-wide, across every campaign and channel.

Principle 12 demands more than an unsubscribe link: it covers free text too. Someone replying “no need” has objected — the sequence must recognise that and stop, not send step 3 the next day.

Legally this is the place where the principles fell behind the law rather than ahead of it. Art. 21(2) GDPR grants the objection right against direct marketing unconditionally, and Art. 21(3) GDPR makes the consequence mandatory: “Where the data subject objects to processing for direct marketing purposes, the personal data shall no longer be processed for such purposes.” No discretion, no grace period.

The decisive one is Art. 21(4) GDPR, routinely overlooked in sequences: the data subject “shall be explicitly brought to the attention of” that right “at the latest at the time of the first communication”, and that notice “shall be presented clearly and separately from any other information.” The notice about the objection right is therefore not a footer in message three but a duty in message one, set apart from the rest. The same timing applies to the provenance disclosure under Art. 14(3) GDPR where the data is used to contact the person.

On top of that, §7(2) no. 3 UWG requires a “valid address to which the recipient can send a request to stop such messages, at no cost other than transmission costs at base rates”.

Principle 13 is the line that separates systems. A suppression list per campaign is not a suppression list. The objection belongs to the person and the company, not to the sending channel, and it does not expire. In practice: one central suppression that every campaign checks before sending, and that still bites when the same person re-enters months later through a freshly bought list.

E. The machine — does it stay controllable?

#	Principle	What it means concretely
14	Auditability	Every action the AI takes is traceable.
15	Stop conditions	On uncertainty the AI halts and escalates to a human.
16	Learn from responses	The system optimises ICP, messaging and qualification — not send volume.

These three are the difference between automation and loss of control.

Principle 15 needs named stop conditions, otherwise it is a statement of intent. Workable candidates: the source for a signal is older than X days. The role could not be verified. The reply is ambiguous rather than clearly positive or negative. The channel requires a consent that is not documented. In all four, the right action is to halt and ask a human — not to estimate a probability.

Principle 16 is the objective function, and in practice it is the one most often set wrong. A system optimised for send volume will always find a way to send more. One optimised for qualified replies finds a way to aim better — and usually sends less doing it.

The legal catch in principle 1

This is where well-meant practice and German law part ways — and it is the most important limitation on this page.

“Research first, sell later” sounds as though a research first touch were legally safer than a sales email. It is not. The Federal Court of Justice reads “advertising” broadly enough that an outreach serving your sales is advertising even when it sells nothing:

“The concept of advertising covers, in ordinary usage, all measures by an undertaking aimed at promoting the sale of its products or services. Besides directly product-related advertising, this therefore also covers indirect sales promotion — for instance in the form of image advertising.”

— BGH, judgment of 10 July 2018 – VI ZR 225/17, para. 16 (BGHZ 219, 233), aligning the term with Art. 2(a) of Directive 2006/114/EC

In the same decision, official headnote b) reads: “A customer satisfaction survey in an email falls under the concept of (direct) advertising even where the email transmits an invoice for a product previously purchased.” A survey is advertising — and the unobjectionable part of the message does not strip it of that character.

The Cologne Court of Appeal worked the telephone case through and named the one exception:

“By unanimous view, §7(2) no. 2 UWG does not apply where the survey is conducted by a neutral institute for scientific purposes and does not directly serve the sales promotion of a specific principal.”

— OLG Köln, judgment of 12 December 2008 – 6 U 41/08

There, a bank had a commissioned market research institute survey its customers about service satisfaction. The court found an unreasonable nuisance — although a neutral institute made the calls, although the subject was service quality, and although nothing was sold. It sufficed that the survey served the goal of retaining those people as customers.

What follows in practice — three sentences that make the difference:

1. A research first touch is no legal free pass. If your research serves your sales, it is advertising, and the channel-dependent requirements of §7(2) UWG bite on message one, not on the message where you finally sell. Consent you do not have for the channel is not obtained by asking first instead of offering.
2. Genuine market research is a narrow path. The exception requires a neutral institute, a scientific purpose, and no sales promotion for a specific principal — all three at once. A sales team “doing market research” satisfies none of them.
3. Principle 1 is therefore a quality instrument, not a compliance instrument. It makes outreach relevant and honest. It does not replace consent. Confusing the two builds exactly the pretext principle 2 forbids — just with a cleaner conscience.

The conclusion is not to abandon research. It is to treat research and channel choice as two separate questions: research decides whether the message is relevant; the channel decides whether you may send it. The channel question is answered in the channel matrix.

How to audit your own sequence

The principles are built as a checklist. Take a running sequence and walk it in this order — the first three questions usually surface most of the findings:

1. Open message 1 and delete every sentence that sells. Is there a real question left whose answer you would use even without a deal? (Principles 1, 2)
2. Take the signal message 1 cites and look for its source and date. Not on the record — on the field. If either is missing, the personalisation is unevidenced. (Principle 5)
3. Send a test reply of “not interested” and wait for the next step. If it still arrives, 12 and 13 are documented, not implemented.
4. Take the same message and swap the recipient for the neighbouring role. If it works unchanged, it is generic. (Principles 8, 9)
5. For any contacted person, ask the system for the decision chain: why this person, from which source, over which channel, on what legal basis. If that takes more than a minute, principle 14 is missing.
6. Ask what your system optimises for. If a send count sits at the end of the dashboard, principle 16 is not met — whatever the concept document says. (Principle 16)

Work through those six and you usually end up with two or three concrete fixes instead of a debate about philosophy.

Why this is not a sacrifice

The most common objection to this list is that it costs reach. That is true of send volume and untrue of outcomes, and the reason lies in the question buyers actually ask.

Code your inbound replies by question type and most B2B funnels show the same ranking: “how does this work?” sits well out in front — ahead of price, and ahead of requests for proof. Buyers want the mechanism explained, not to be argued into a position. A system that researches first and derives the outreach from context can answer that question. One that maximises reach never gets into a position to be asked it.

Count it in your own reply corpus before adopting the ranking. The point is the order, not any particular number.

The principles are therefore not a compliance surcharge on a working system — they are a description of what a working system does anyway. “Earn the right to sell” is the short form.

---

Try it in practice: GTM Goat implements these principles as system constraints — source and timestamp at field level, stop conditions with escalation to a human, system-wide suppression. Start a free trial.

Sources

• Regulation (EU) 2016/679 (GDPR), in particular Art. 5(1)(b) (purpose limitation) and (c) (data minimisation), Art. 6(1)(f) (legitimate interests), Art. 14 (information duty where data is not obtained from the data subject) and Art. 21 (right to object; unconditional for direct marketing).
• German Act Against Unfair Competition (UWG), §7 (unreasonable nuisance, channel-dependent consent requirements). The per-channel breakdown with statutory citations is in the channel matrix.
• Regulation (EU) 2024/1689 (AI Act), Art. 50 — transparency obligations for certain AI systems, including systems interacting directly with natural persons. The transparency chapter has applied since 2 August 2026.
• BGH, judgment of 10 July 2018 – VI ZR 225/17 (BGHZ 219, 233), headnote b) and para. 16 — customer satisfaction survey as (direct) advertising; the concept of advertising including indirect sales promotion, aligned with Art. 2(a) of Directive 2006/114/EC. Official full text: bundesgerichtshof.de, ECLI:DE:BGH:2018:100718UVIZR225.17.0.
• OLG Köln, judgment of 12 December 2008 – 6 U 41/08 — telephone customer survey by a commissioned market research institute as an unreasonable nuisance; exception only for a neutral institute acting for scientific purposes without direct sales promotion for a specific principal. Official full text in the NRW case-law portal.
• The 16 principles themselves are CegTec’s own working basis for building and operating outbound systems with AI agents. In several places they deliberately go beyond what the law requires.