04 Chatbots & Voice active service

Across web, WhatsApp and your call center,
assistants that talk.

Natural in Turkish, on-brand, working 24/7. Voice-enabled with the Realtime API and ElevenLabs; intelligent with Claude and GPT; wired into your CRM, because it has to get real work done.

~5 weeksrollout
−41%call duration
7/24always on
<300msresponse latency
AI Assistant online
Could you book me an appointment for tomorrow at 14:00?
0:08
01 · Capabilities

Conversation in six dimensions.

Not an assistant but a team of assistants: it routes by topic, hands off to a human when needed, and writes the summary afterward.

Web & WhatsApp assistant

From the chat widget on your website to your WhatsApp Business account — one assistant, many channels.

  • Web chat (custom UI / iframe)
  • WhatsApp Business API
  • Instagram DM & Messenger
  • Telegram & Slack

Realtime Voice AI

Voice AI at phone-call quality. It can fully replace your IVR, with <300ms latency.

  • OpenAI Realtime API
  • ElevenLabs voice synthesis
  • Natural Turkish accent
  • Twilio / SIP integration

Brand voice calibration

The assistant speaks like your brand. Persona, tone, word choice, banned phrases.

  • Persona document
  • 50+ sample dialogues
  • Banned / red-line list
  • Tone testing + calibration

CRM integration

The assistant knows the customer and their history, opens tickets, books appointments — real actions, not fake ones.

  • Salesforce / HubSpot / Pipedrive
  • Zendesk / Intercom / Freshdesk
  • Custom API integration
  • Calendar (Calendly, Google)

Call summary & actions

When the call ends: an automatic summary, ticket and follow-up action — ready for the operator in one click.

  • Call transcript
  • AI summary + action list
  • Sentiment analysis
  • Automatic ticket / email

Human handoff

When things get complicated, it hands off to a human — the operator takes over with full context, and the customer doesn't have to repeat themselves.

  • Threshold-based escalation
  • Sentiment / risk detection
  • Operator panel (history + context)
  • Shift / working-hours rules
02 · Anatomy

Conversation in three stages.

From voice/text input to real action. Each layer has its own model and its own limits.

01

Input layer

Where is the user talking from? Voice, text, any channel — all accepted.

Phone (SIP/Twilio) WhatsApp Web chat widget Instagram DM Telegram Mobile app voice
02

AI brain layer

Voice → text → meaning → answer → voice. All in under 300ms.

STT (Deepgram, Whisper) LLM (Claude / GPT) TTS (ElevenLabs) RAG (company knowledge) Sentiment analysis Moderation filter
03

Output & action

An answer + real action. Tickets, appointments, transfers — everything leaves a trace in the CRM.

Voice/text reply Appointment booking Ticket creation Transfer to operator Call summary Follow-up email
03 · Industries

A talking assistant delivers in every industry.

Click your industry — use cases with real metrics.

24/7 customer support assistant

Answers questions from web and WhatsApp 24/7. Resolves most on its own; hands complex ones to an operator with full context. Within SLA, tracked.

Tier 1: 78% self-resolve Escalation with operator handoff SLA & KPI dashboard Multilingual (TR/EN/AR/DE)
%78self-resolve rate
−63%operator workload
4.6/5customer rating

Healthcare triage & appointment assistant

Turkish voice AI: listens to what the patient is looking for, finds the right specialty, books the appointment, sends reminders. Integrated with the EHR.

Specialty routing algorithm Calendar / EHR integration Reminders + cancellation management KVKK + KHGM compliant
−41%call duration
+38 minsaved per doctor/day
%92note approval rate

WhatsApp sales & support assistant

The customer searches for a product, the assistant recommends, completes the cart, starts the return. Orders + support in one window. Drives real sales.

Product recommendations + cross-sell Cart → payment link Return & exchange automation Re-engagement campaigns
2.4×WhatsApp conversion
+18%average basket
14minreturn processing time

Call center co-pilot

While the operator talks, AI assists on a side screen: customer history, suggested reply, next action. Training goes from weeks to minutes.

Live transcript + suggestions Customer 360 view Automatic summary + ticket QA + coaching panel
−32%AHT (avg. handle time)
+48%FCR (first-call resolution)
72%faster training

Legal assistant & first filter

Gathers prospective clients' questions at the first level, collects document uploads, routes by practice area. Qualified cases land on the lawyer's desk.

Case pre-assessment Document upload + analysis Practice-area matching Confidentiality + KVKK compliance
+185%qualified leads
−54%administrative workload
%100document confidentiality
04 · Stack

The conversation arsenal.

From realtime voice engines to messaging channels, from the brain to CRM connections.

Voice (STT / TTS / Realtime)
OpenAI Realtime APIElevenLabs DeepgramWhisper Twilio VoiceVapi Retell AI
Messaging Channels
WhatsApp Cloud APITelegram Bot Instagram Graph APIMessenger SlackMicrosoft Teams Web SDK (custom)
AI Brain
Claude Opus 4.6GPT-5 Gemini 2.5Custom RAG PineconeFunction calling Structured output
CRM & Integration
ZendeskIntercom FreshdeskSalesforce HubSpotCalendly Google Calendar
05 · Case

An example from the field.

An anonymized but real Voice project: the call center transformation of a multi-branch healthcare chain.

CASE / 04 Voice AI · Healthcare · 6 weeks
Problem

A private healthcare chain with 14 branches. 4,000+ calls a day, averaging 8 minutes. Patients spent hours listening to hold music, appointments were booked under the wrong specialty, and the no-show rate was high.

Our approach

OpenAI Realtime + ElevenLabs Turkish voice. Connected to the EHR: the patient calls, the assistant recognizes them, finds the right specialty, books an open slot and schedules reminders.

Result

Live in 6 weeks. Call time dropped from 8 minutes to 4:42, and no-shows fell from 18% to 7%. Operators now focus on complex cases.

−41% average call duration
−61% no-show rate
4.7/5 patient satisfaction
06 · Packages

Where should we start?

An X-ray of your current conversation channels, or a pilot assistant right away.

Audit 3 days

Conversation Audit

We scan your customer conversations and score the points that can be automated.

  • Call/chat transcript analysis
  • Automation opportunity map
  • ROI estimate
  • Proposed architecture (summary)
  • Persona & tone recommendation
Pilot 5 weeks
Most popular

Assistant Pilot

One channel (WhatsApp or phone), one use case. Live in 5 weeks.

  • Fixed scope + fixed price
  • Persona & voice calibration
  • CRM / Calendar integration
  • Handoff & escalation flows
  • 30-day hyper-care
Scale Monthly retainer

Conversation Team

Multiple assistants, multiple channels, continuous improvement + 24/7 monitoring.

  • Multi-channel orchestration
  • Weekly conversation analysis
  • Persona evolution & QA
  • 24/7 support with SLA
  • New channels/use cases added
07 · FAQ

Common questions about chatbots & voice.

How natural is the Turkish? Does it sound like a robot?

The Turkish voices from ElevenLabs and OpenAI Realtime got very close to human level in 2024-2025. We can also build a custom voice clone for your brand (a 5-minute sample is enough). Intonation, pauses, even breathing sounds come across as natural. Most users think they're talking to a person.

How does it connect to our existing phone system?

Three ways: (1) Set up in parallel with your current IVR, with AI handling specific numbers. (2) A full move over Twilio/SIP. (3) Native integration with central systems like Genesys and Avaya. The transition is gradual, with no risk.

How does it hand off to a human when it hits a complex case?

Three triggers: (1) when the uncertainty score crosses a threshold, (2) when the user says "I want to talk to a person", (3) when specific keywords/emotions are detected (anger, complaints, legal). The operator takes over with the full call transcript + the AI's suggestion, and the customer doesn't have to repeat themselves.

Data security? KVKK compliance?

Call recording is optional (default: not stored). When stored, recordings stay in the EU region, encrypted, under a retention policy. PII (ID, phone, card) is masked in the conversation flow. KVKK, GDPR and KHGM compliance checks are standard. A separate protocol applies for healthcare.

Are there traffic / usage limits? What does it cost?

No limits — you pay for what you use. Typical cost: WhatsApp assistant ~₺0.4-1.2/conversation, voice assistant ~₺2-6/minute. A monthly budget cap is set, with an alert at 80%. A detailed cost dashboard is always visible.

What if the assistant gives a wrong answer?

Three layers of protection: (1) RAG, so it answers only from company documents; (2) when uncertain, "let me check and get back to you" + escalation to a human; (3) QA + evals after every conversation for continuous improvement. The hallucination rate is kept below 1%.

Contact

Let's start with a demo call.

Talk to your assistant within 5 minutes. A Turkish voice sample, your brand voice, real usage. We reply in under 24 hours.

We reply in under 24 hours.