Skip to main content
Audact
See the offer

Sector · Financial services

Financial services: decided before the line opens, said within the approved set.

Pilot intake open

Financial communication is the most heavily regulated voice channel in Europe, and the rules stack: DORA on the client's operational resilience, MiCA on what may be said about crypto-assets, the GDPR on the data, Article 50 on telling the person they are speaking to an AI, and national rules on whether an outbound call may be placed at all. Two mechanisms carry almost all of it: an eligibility decision before the call is placed, and enforcement of the approved set of statements while it runs.

What the law requires

A person interacting with an AI system must be told so, unless it is obvious from the circumstances.
Regulation (EU) 2024/1689, Art. 50(1)
Contracts for ICT services used by a financial entity must carry a fixed set of provisions, and major ICT incidents must be reported to the competent authority.
Regulation (EU) 2022/2554, Art. 30 and Art. 19
Communications about crypto-assets must be fair, clear and not misleading, and marketing must be identifiable as marketing.
Regulation (EU) 2023/1114, Art. 7 and Art. 66
Unsolicited calls to natural persons in the Netherlands need a basis, and the caller must identify itself and offer the right to object.
Telecommunicatiewet, art. 11.7
A German collection service must give the consumer a defined set of information when it first asserts a claim.
§ 13a RDG (DE)
In the Netherlands the extrajudicial collection costs chargeable to a consumer are capped by decree, so an agent that quotes a figure must quote the right one.
Besluit vergoeding voor buitengerechtelijke incassokosten, on the basis of art. 6:96(5) BW (NL)

What Audact does about it

An eligibility check before the line opens: the number is tested against consent state, opt-out lists and the contact rules for that jurisdiction, patent-pending (UK IPO).
In controlled testing · Germany: national policy layer active. Netherlands: in development. Belgium and Austria: planned.
Approved-message enforcement: figures, terms and claims the agent may state are validated against what you approved, and anything else is refused.
In controlled testing
The Article 50 disclosure delivered where the gate is enabled for that jurisdiction.
In controlled testing
A receipt per governed segment: a hash of the output the gate released (on a block, the fallback reply the caller receives), what was refused and what was relied on.
Working today
Vulnerable-person handling, so a caller in difficulty is escalated rather than pressed.
In development
DORA and MiCA policy packs, and voice authentication against vishing.
In development
DATEV write-back for accounting workflows.
Planned

Two mechanisms

Whether you may call, and what you may say.

Almost every rule in this sector resolves into one of those two questions. Whether you may call is decided before the line opens: consent state, opt-out lists, contact rules for that jurisdiction, and the time of day where that is regulated. A call that fails the check is not placed, which is a different thing from a call that is placed and later found to have been wrong.

What you may say is decided while the call runs. Amounts, terms, risk warnings and the wording of a claim are validated against the approved set; anything outside it is refused rather than improvised. In print that review happens once, before publication. In a voice channel there is no before, which is why the check has to sit in the path rather than beside it.

Collections

Debt collection is the hardest test of the same stack.

Outbound collection combines everything at once: an unsolicited call, a consumer who did not ask for it, a regulated set of information that must be given, a capped amount that must be quoted correctly, and a real chance that the person on the line is in financial difficulty. It is also where a generic agent does the most damage the fastest.

The mechanisms do not change for it. Eligibility is decided before dialling, the amount and the notice come from the approved set, vulnerable-person handling escalates instead of pressing, and the receipt records what was relied on and a hash of the output the gate released. What changes is how little room there is for the agent to improvise — which is the point.

Questions

Does DORA apply to my agency?
Not directly, unless the agency is itself a financial entity under Article 2. It reaches you through the contract with your financial client: Article 30 requires a fixed set of provisions in that agreement, and the client needs evidence from you for its own register and reporting. This page is general information, not legal advice.
Can the agent quote an amount?
Only an amount from the approved set. Figures are validated before they are spoken rather than checked afterwards, which matters most where the chargeable amount is capped by law and a wrong figure is itself the breach.
What happens when the person on the line is clearly struggling?
The interaction adapts and escalates to a person instead of continuing the script. That behaviour is recorded on the receipt for the call, which is what makes it answerable if the interaction is questioned later.

The decision that happens before the dial

Eligibility is settled before the call is placed, not reviewed after it. See how the outbound engine tests a number against consent, opt-out and the rules for that jurisdiction.

See the outbound eligibility API

Related