The voice assistant that answers “Where is my parcel?” from the order record and logs return requests
For customer service teams: order status and delivery questions go to the assistant; requests it cannot resolve reach the support team with a ticket.
Common problems on the phone line
A wave of calls on campaign days
During campaigns and sales, delivery questions arrive all at once and the queue grows.
Always the same question: where is my order?
Delivery questions usually need a single piece of information; the team opens and reads the same record every time.
Return requests are scattered
Return and exchange requests discussed on the phone reach the support ticket with missing details.
How it works in this sector
Every call follows the same four steps; the assistant acts only with defined permissions.
- 01 / 04
Greets
Greets the customer and listens to the order or return request.
- 02 / 04
Understands and verifies
Verifies the customer and the order number.
- 03 / 04
Takes action
Checks order and shipping status in permitted records.
- 04 / 04
Confirms or hands off
Opens a support ticket if needed and shares its number; hands unresolved requests to the team.
What the assistant takes on
Understands order and after-sales requests and queries permitted records.
- Order and delivery status
- Return or exchange requests
- Support team handoff
- Delivery window information
- Support ticket creation
Example call
A fictional scenario: what the caller says, what the assistant understands and how it replies.
- Caller
- “My order has not arrived. Where is my parcel?”
- Detected request
- Delivery enquiry · Order information required
- Assistant response
- After verification and your order number, I can check shipping status and open a support request if needed.
Returns and refunds follow company policies and defined permissions.
Illustrative example · Not live system data
Related integrations
The assistant connects to your systems only with defined permissions; your PBX setup is reviewed during discovery.
CRM · ERP
API access only with defined permissions
Inbound and outbound calls
Answering, information and confirmation calls
Operator handoff
Calls that need a person, plus fallback routing
Call summary to the CRM
The call summary is written to the relevant record
Runs on your own server
- Your data stays in-house.
- A single GPU server is enough.
- Several calls at the same time, on the same system.
The architecture supports your compliance processes with restricted API access, configurable retention and operator fallback. Compliance depends on the deployment configuration, data processing practices and your organizational procedures.
The path of a call
- 01Phone lineNetgsm 0850 · SIP Trunk
- 02PBXAsterisk
- 03Your serverSTT → LLM → TTS
- 04Your systemsCRM · ERP
Frequently asked questions
- Does it approve returns and refunds itself?
- No. The assistant takes an initial request for returns and exchanges; processing follows your company's rules and defined permissions.
- How is the customer verified?
- Verification steps (for example the order number and contact details) are defined together with your company; the assistant queries only permitted records.
- Does voice data leave the organization?
- No. Speech-to-text, response generation and speech synthesis run on your own server; your data stays in-house. Calls arrive through your telephone carrier and PBX.
- What if the agent can't resolve a request?
- Calls that need a person are transferred to an operator, with fallback routing in place.
Request a demo
Your sector is preselected in the form. Briefly describe your call scenario, and we will schedule an online meeting at a time that suits you.
Let's plan your phone line together
In a discovery call we go through your call flows and the systems to connect.
* Footnote: Latency values were measured on real call recordings (from the moment the caller stops speaking until the first sound arrives, server side). Internet latency adds 0.05–0.3 s. Word error rate was measured on the public FLEURS-tr dataset (200 sentences, 8 kHz telephone audio). Language support is at model level. Live validation was done in Turkish; in other languages, voice and recognition quality may vary by language.