Telco Voice service spent the last 15 years disappearing inside unlimited bundles. Operators kept investing in the network, but stopped treating the call itself as a product as the innovation moved into data plans.
So, other players started to conquer that Voice “White Space”. Twilio made communications programmable, contact-center platforms captured the workflow, and WhatsApp and other apps captured the customer interaction. Telcos were only paid for access and termination while software companies owned the developer and the business logic.
But with AI, Telco voice is sexy again, and some Telcos are realizing this amazing opportunity. AI can understand normal speech, maintain context, and complete an action during a call. An AI-powered conversation can produce a booking, qualify a lead, update a CRM record, translate two speakers, detect signs of fraud, or trigger a payment workflow. So, with AI, a call is no longer only a connection between two people, but it becomes part of a business process.
Orange has caught that shift. Its new Voice API gives developers direct access to calling through Orange’s telecom infrastructure.
The targeted launch window is September 2026, subject to final technical, operational and commercial validation. Orange is also holding a developer masterclass on September 29, which makes clear that the product is still moving from technical availability toward a wider commercial launch.
The important part is not that Orange has launched another API. It is that the company is beginning to turn the phone call into software.
Orange is opening the call to developers
The new Orange Voice API allows an application to make outbound calls, receive calls on an allocated number, connect participants, create conferences, play audio, generate speech, and collect either spoken or keypad input. A developer can also change what happens while the call is in progress.
Orange does this through Orange Markup Language, or OML, a JSON-based way of defining the call flow. The application can start with one scenario, listen to what the caller says, and then replace that scenario with another one. A call could begin as a simple receptionist, move into identity verification, connect to an external agent, and then transfer to a person without ending the session.
Orange sends call events, action results, media files, and transcription updates back to the application through callbacks. The customer can feed that information into a CRM, booking platform, fraud engine, analytics service or external AI model. Orange manages the number, routing, media, call control and orchestration. The customer controls the business rules, data processing and final user experience.
This division is one of the strongest parts of the offer. Orange is not forcing an enterprise to use one Orange model or one software stack. A company could connect Gemini, OpenAI, ElevenLabs, a European model or its own private system. Models will change quickly. The connection to the regulated telephone network, the phone number, the local operating licenses and the network information around the call are much harder to replace.
The customer also remains responsible for long-term data storage. Orange may make temporary audio and media available under the limits of the contract, but the enterprise decides where recordings and transcriptions are kept and when they are deleted. That is useful for banks, hospitals, governments, and companies that need clear control over where customer conversations are processed. Authentication can use OAuth 2.0 or an API key depending on the deployment.
There are some missing aspects that I believe should be added to the roadmap. For example, Orange Voice API does not include the AI analysis itself. It provides telephony, supported transcription, media, and call orchestration. Summaries, recommendations, and business actions must come from the customer or another service. SMS is also a separate Orange API.
The API cannot silently record every normal mobile call made between two Orange subscribers. It only controls calls routed through a Voice API-managed number, scenario, or conference. This is important because the product is not yet a general intelligence layer inside every Orange call. It is a programmable telephony service for specific business applications.
Even within that boundary, the opportunity is large. Orange says voice accounts for 70% of customer service interactions, while only 25% of those calls are used to create valuable analytics. The figures will vary by company, but the gap is big. A huge amount of customer intent, frustration, and commercial information is still lost when a call ends.
A hotel could use Orange Voice API to answer calls after midnight, check availability, complete a reservation, and send confirmation by SMS. A clinic could schedule appointments and route urgent cases to the right person. A recruiter could carry out an initial screening and write the result into the applicant system. A field-service company could receive a spoken repair report and automatically update the job record.
Complete voice agents will attract most of the attention, but simpler workflows may generate revenue faster. Orange can help an enterprise transcribe a call, identify the customer’s request, create a short summary, and write the next action into the CRM. The system can create a task, send an alert, or trigger a follow-up message. That removes manual work without forcing the company to replace its contact center or trust an autonomous agent with the entire conversation.
The offer becomes more interesting when voice is combined with Orange’s other APIs. Branded Calling can replace an unknown number with an operator-certified business identity. Number Verification can confirm that the mobile number being used belongs to the device. SIM Swap can show whether the SIM was recently changed, which may increase the risk of fraud. Orange describes Branded Calling and Number Verification as a two-way trust loop: the customer can see who is calling, and the business can verify the customer’s device.