An AI voice agent only creates value if it talks to your real phone system and writes into the CRM your sales team already uses: if the call doesn't come through cleanly and the outcome isn't logged where the salesperson looks every morning, the project stays a demo. Summum IA designs and implements the technical integration layer between the conversational agent, your company's telephony infrastructure — physical PBX, IP-PBX, or cloud provider — and the CRM that holds each customer's history, so that the voice conversation flows in and out of your management system without manual work. The EU AI Act (Regulation (EU) 2024/1689) also requires, from 2 August 2026, that any AI system interacting with people clearly disclose that it is an automated agent (Article 50); we build that transparency obligation into the integration layer itself, not as an afterthought.
Most voice agent projects that fail don't fail because of language model quality — they fail at the integration: the call is handled well but the outcome is logged nowhere, or logged in a system the salesperson never opens. That is why we split the project into two layers with distinct responsibilities: the conversational layer — what the agent says and how it decides — and the integration layer — how the call comes in, what data it looks up mid-conversation, and where the outcome is written once the call ends. This service covers the second layer: the engineering that connects your phone system, your CRM, and the agent.
On the telephony side we work with the platforms most common among Spanish SMEs and mid-market companies: 3CX (on-premise or cloud), RingCentral, Twilio Programmable Voice for custom builds, and any provider exposing a standard SIP trunk. In most cases there is no need to replace your phone system: the agent connects as another extension or through a dedicated SIP trunk, so calls route to it exactly as they would to a human, following the same time-of-day, queue and forwarding rules your company already has configured. On the CRM side, the integration is built against the HubSpot or Salesforce API when you already use one of them — or against any CRM with a documented REST API: the agent looks up the customer record before responding (history, last interaction, contract status) and, once the call ends, automatically writes the transcript, summary, detected intent and suggested next action as an activity or task in the corresponding record.
This entire integration layer is designed from the start for compliance with Article 50 of the AI Act: the notice that the caller is speaking with an AI system is issued at the start of every call from the telephony flow itself — it does not depend on the agent remembering to say it in each conversation — and the audit trail for every call (who called, what was discussed, what was written to the CRM) is documented in a traceable way inside the management system itself, ready for regulatory inspection without extra work on the day it is needed.
We review your telephony infrastructure (brand, model, deployment mode: on-premise, virtualized or cloud) and your CRM (HubSpot, Salesforce or another with an API), identify which extensions, queues and routing rules exist today, and which CRM objects (contacts, deals, tickets) the agent needs to read and write.
We document exactly which telephony events trigger the agent (an inbound call to a specific queue, a transfer from the IVR, an after-hours call), what CRM data it looks up before responding, and what it writes once each type of call ends — including handoffs to a human agent with the context already loaded.
We configure the SIP trunk or the platform's native connector (3CX, RingCentral, Twilio) so the agent receives and places calls with the same audio quality and the same business rules as a human extension, without touching your existing numbering or carrier contracts.
We connect the agent to the CRM API, test the full cycle — inbound call, record lookup, conversation, result write-back, salesperson notification — with real-world scenarios, and adjust the CRM fields and workflows that depend on this new automated data source.
The operational detail: what we deliver as part of the work and what we keep alive afterwards.
Telephony–CRM compatibility matrix
Technical document setting out which telephony platform and which CRM are integrated, by which method (SIP trunk, native connector, REST API), and what limitations apply to your specific setup.
Configured telephony connector
Working connection between the agent and your phone system (3CX, RingCentral, Twilio or a generic SIP trunk) that respects the time-of-day, queue and forwarding rules your company already has in place.
Configured CRM connector
Two-way integration with HubSpot, Salesforce or the REST-API CRM you use: the agent looks up the customer record before responding and writes the transcript, summary and next action once the call ends.
AI Act Article 50 transparency notice
A call-opening message disclosing that the caller is talking to an AI system, built into the telephony flow itself so it never depends on an individual conversation, in line with the obligation applicable from 2 August 2026.
Traceable audit log
A record of every call — origin, summarized content, data written to the CRM — documented and searchable for regulatory inspection or internal quality review.
Integration monitoring dashboard
A dashboard showing the technical health of the integration: correctly routed calls, connector errors, CRM write time and agent availability.
The technical integration is the layer that connects the voice agent with the business infrastructure that Summum Sistemas already implements: the same CRM Sistemas sets up is the destination for every conversation the Summum IA voice agent handles. The split is clear: Sistemas implements and maintains the CRM, IA builds the voice layer that operates on top of it.
Summum Sistemas implements the underlying CRM (generic or with a REST API) that the voice integration builds on: the management system first, then the conversational layer that feeds it with every call.
View service → sistemasWhen the chosen CRM is HubSpot, the voice integration consumes its API to read the contact record and log call activities, without duplicating the rollout Summum Sistemas already carries out.
View service → sistemasFor Salesforce customers, the voice agent connects to standard objects (Contact, Lead, Task) via the official API, reusing the data model Summum Sistemas has already configured.
View service →In most cases, no. The agent connects to your current phone system — 3CX, RingCentral, Twilio or any provider with a standard SIP trunk — as an additional extension or through a native connector, keeping your existing numbering, time-of-day rules and queues intact. Only if your provider offers no standard integration path do we evaluate, as a last resort, a partial migration of the affected line.
We work natively with HubSpot and Salesforce, and with any CRM exposing a documented REST API. The integration looks up the customer record before the agent responds and automatically writes the call transcript, summary and suggested next action into the corresponding record, with no manual work from the sales team.
We evaluate alternatives case by case: some systems offer webhooks or scheduled exports even without a full REST API, and in other cases the recommended first step is migrating to a CRM with an open API — something that can be coordinated with Summum Sistemas — before building the voice integration on a foundation that cannot support it reliably.
Yes. Article 50 of Regulation (EU) 2024/1689 (AI Act) requires that people be informed they are talking to an AI system whenever this is not obvious from context; this transparency obligation applies from 2 August 2026. We build it into the telephony flow itself, so the notice is always issued at the start of the call, regardless of how the conversation unfolds.
It depends on the platform and on whether the CRM already exposes an open API: an integration with 3CX or Twilio and a CRM with a standard REST API is usually completed within weeks, while older phone systems without a native connector or CRMs without a documented API add upfront analysis time. Our initial audit delivers a concrete timeline for your specific case.
The agent transfers the call within your own phone system, following the same queue and time-of-day rules you already have configured, and hands the human agent the context already retrieved from the CRM — reason for the call, recent history — so the customer never has to repeat information.