Why voice assistants still don't work for Turkish SMEs

Five concrete technical blockers

Existing solutions cannot reliably serve Turkish SME calls on these five points.

1. 01

   Turkish proper-name recognition error

   Turkish place, brand and personal names are systematically garbled by ASR. Out-of-vocabulary issues, agglutinative morphology, and narrow phone-band audio combine. A single misheard name sends the job to the wrong address.

   blocker · proper-names
2. 02

   Free-form location is unresolvable

   Even though location is decisive in 59% of calls, callers give it in everyday phrasing rather than official administrative-unit names. Existing solutions do not resolve the utterance into an administrative-unit identifier.

   blocker · location
3. 03

   Generative models fabricate information

   Large language models tend to fill empty fields with plausible values and report un-completed work as completed. In a call-centre context, that means a customer waiting for a service that will not arrive.

   blocker · hallucination
4. 04

   Latency

   Silence while a backend query runs causes the caller to hang up. On cloud paths, the network round-trip adds to that silence.

   blocker · latency
5. 05

   Cost and dependence

   Cloud solutions charge the SME twice: call-forwarding fees to the operator and per-minute usage fees to the AI service. When the internet drops, service stops entirely. Voice data leaves the premises.

   blocker · cost