A customer in Mymensingh receives a higher‑than‑usual electricity bill. Confused, she calls the helpline. Instead of waiting in a long queue, a voice bot greets her in Bangla: “আপনার বিল নিয়ে কোনো জিজ্ঞাসা থাকলে বলুন।” She explains the issue. The bot checks her billing history, identifies a possible meter reading error, and offers to file a correction request — all without involving a human agent.

This is Taritbandhu, a groundbreaking hybrid AI call center solution for Bangladesh's power sector. It's not just a chatbot; it's a three‑tier system that seamlessly blends conversational AI, deterministic data lookup, and human escalation.

The challenge: millions of customers, diverse queries

Bangladesh's electricity distribution companies serve over 4 crore (40 million) consumers. Call centers are overwhelmed, especially after billing cycles. Customers face long wait times, and agents spend most of their time answering repetitive queries:

• “আমার বিল কত?” (What's my bill amount?)
• “কবে নাগাদ বিল দিতে হবে?” (When is the due date?)
• “লেট ফি কত হবে?” (What's the late fee?)
• “নতুন সংযোগ নেব কীভাবে?” (How do I get a new connection?)

Taritbandhu was designed to handle these at scale, while ensuring that complex issues (like disputed bills or technical faults) reach the right human expert with full context.

The three‑tier architecture

Taritbandhu's strength lies in its layered design. Let's break it down.

Real‑world impact: before and after Taritbandhu

A pilot in a zone with 5 lakh customers showed dramatic improvements:

• Average wait time: Dropped from 4 minutes 30 seconds to 55 seconds.
• First‑call resolution: Increased from 62% to 84%.
• AI containment rate: 48% of calls fully resolved without human agent.
• Agent productivity: Each agent now handles 2.5x more complex cases per shift.

How deterministic matching changes the game

Many voice bots fail for utilities because they try to "understand" something that's purely numerical. Taritbandhu's logic layer knows the difference between:

• “আমার বিল কত?” → route to billing API → return exact amount.
• “বিল জমা দেওয়ার পর রসিদ কোথায় পাব?” → FAQ retrieval → explain payment confirmation methods.

This hybrid approach ensures accuracy for transactional queries while preserving the flexibility of AI for informational ones.

Built for Bangladesh's power sector

Taritbandhu wasn't imported; it was built ground‑up for our utilities. Features include:

• Multilingual support: Automatically adapts to Sylheti, Chittagonian, and standard Bangla.
• Load shedding info: Integrates with zone‑wise outage schedules.
• Payment confirmation: Connects to bKash, Nagad, and bank payment gateways.
• Voice biometrics: Optional authentication via voiceprint (coming 2026).

What customers and agents say

Scalability: beyond electricity

The three‑tier model isn't limited to power. Any utility or government service with a mix of transactional and informational queries can benefit:

• Water & gas utilities — bill lookup, leak reporting.
• Municipal services — tax inquiries, application status.
• Telecom — usage checks, plan changes.

Speaklar provides the building blocks: ASR, TTS, deterministic API connectors, and RAG for document retrieval.

Building your own hybrid AI call center

You don't need a massive IT team. With Speaklar, you can:

• Connect your existing databases (billing, CRM) via simple APIs.
• Upload FAQs and manuals — our system chunks and indexes them for retrieval.
• Define escalation rules (e.g., if customer says “এজেন্ট দিন” or repeats a question).

A prototype can be live in 2–3 weeks.

⚡ শক্তির সাথে এআই — গ্রাহক সেবায় নতুন মাত্রা

---