Skip to main content
Audact
See the offer

Integration · DATEV

Client outreach that files itself in DATEV.

Planned · no date set

Outbound client outreach, inbound advisory triage and bookkeeping data capture for accounting firms, with the result written back through DATEVconnect instead of retyped. The governance runs in the call: the disclosure, the outbound eligibility check, and the boundary between answering a question and giving advice.

What the agent produces per call

{
  "event": "call.completed",
  "outcome": "documents_requested",
  "outcome_source": "conversation",
  "confirmed_by": null,
  "receipt_id": "rcpt_01HX7K9M3P4Q5R6S7T8U9V",
  "article_50": { "disclosed": true, "clause": "1", "at_turn": 1 },
  "eligibility": { "decision": "allow", "checked_at": "2026-08-15T09:58:12Z" },
  "write_back": {
    "system": "datevconnect",
    "client_ref": "sha256:e3b0c44...",
    "activity": "Belegnachfrage Q3",
    "due": "2026-09-15"
  }
}

Worked example. Acme GmbH and Nova Industries are fictional; no customer data appears here.

outcome is what was requested in the call; confirmed_by stays null until the system of record confirms it.

What applies to the call

The client must be told they are speaking to an AI system, at the latest at the first interaction.
Regulation (EU) 2024/1689, Art. 50(1) and 50(5)
Telephone advertising to a consumer without prior express consent is an unacceptable nuisance.
Gesetz gegen den unlauteren Wettbewerb (DE), § 7(2)
Financial entities must manage and report ICT-related incidents.
Regulation (EU) 2022/2554 (DORA), Art. 17, applicable since 17 January 2025

What Audact governs in the call

Article 50 disclosure at the top of the call, recorded with its position.
In controlled testing
Outbound eligibility before every dial: consent, registries, quiet hours.
In controlled testing · Germany: national policy layer active. Netherlands: in development. Belgium and Austria: planned.
Scope boundary between an answer and tax advice, routed to a person.
In development
Write-back of activities, client records and document references via DATEVconnect.
Planned

Sector

The chase for documents is the whole job.

An accounting practice spends a measurable part of the year asking clients for the same receipts. That is a conversation with no judgement in it and a great deal of repetition, which is exactly the shape a governed agent handles well.

The line that matters is the one between asking for a document and answering a tax question. The first is administration; the second is advice, and advice belongs to a person with a licence. The gate holds that line during the call rather than in a briefing afterwards.

Setup

Four steps, once the integration ships.

  1. Connect your DATEV tenant through the DATEVconnect APIs.
  2. Map agent personas to client engagements and bookkeeping categories.
  3. Configure the disclosure wording and the advice boundary per language.
  4. Run in shadow mode (writes are mirrored, not committed), then switch to active.

Questions

Does client data go into the model?
Identifiers do not. What the agent needs to hold a conversation is passed as references; what it produces comes back as structured fields, and the identifiers stay in the compartment with their own key.
Can the agent give tax advice?
No. Each sentence segment the model produces is checked before it is sent to speech; a segment already spoken is not withdrawn. A transfer to a person, triggered by the agent, is in controlled testing; automatic hand-off on red-flag signals is in development. The receipt records where the boundary was drawn.

Do not wait for the integration

Every integration will be built on the same six endpoints. They run inside the managed platform; no public API is served on api.audact.ai yet, apart from the published Merkle roots. Where a sector system exposes an API, the shorter route is to call it from your side and let Audact govern the conversation.

See the API

Related