Bangladesh-based Ascend AI is developing FastCom AI, a platform that connects customer conversations with businesses' internal systems so AI agents can handle enquiries and complete tasks rather than simply answering questions. The system is already handling around 95% of customer conversations in some early F-commerce deployments, according to the company.
For customer support teams, the shift matters. Earlier rule-based chatbots relied on fixed question-and-answer scripts and often failed when a conversation required information from internal systems or an action beyond a canned response. FastCom AI is part of a newer category of technology - commonly described as agentic AI - designed to identify what a customer wants, determine the next step, and carry out a task within limits set by the business.
From answering questions to completing tasks
The distinction is clearest in routine order enquiries. A chatbot without access to a company's order system may be unable to provide an update. An AI agent connected to that system can look up the order and return its status. Depending on configuration, the agent could also register a complaint, initiate a return request, book an appointment, update a customer record, or transfer the conversation to a human representative.
FastCom connects customers with a company's existing systems, bringing together conversations from Facebook Messenger, WhatsApp, Instagram, web chat and email. The AI can integrate with customer relationship management (CRM) software, ticketing platforms, order and inventory systems, enterprise resource planning (ERP) software, and company knowledge bases. This allows the agent to retrieve information from approved databases and business systems when responding to customers, rather than relying solely on prepared answers.
For an e-commerce business, that could mean checking product availability, confirming an order or providing a delivery update. For an electronics company, it could involve answering warranty questions, registering a service request or checking repair status. An automobile company could use the system for model and price enquiries, test-drive bookings and after-sales service questions. Healthcare providers could use it for appointment bookings, doctor availability and report-status enquiries.
Built for how customers communicate in Bangladesh
FastCom is also being designed around local communication patterns. The platform supports Bangla, Banglish and English, allowing customers to move between languages within a single conversation. A customer might ask about a product in Bangla, type its model name in English, send a follow-up in Banglish and then ask about delivery. Supporting these exchanges matters where customer interactions rarely follow a fixed script.
Some of FastCom's earliest deployments have been with F-commerce businesses, where much of the sales process takes place through Facebook and other messaging platforms. The company cited one seller as an example. Before deployment, the business had around 20 people involved in customer management. After rollout, two representatives handled exceptions and escalations while the AI managed most routine interactions. FastCom said customer-management costs fell by more than half and revenue increased by around 20% after deployment, though revenue changes can reflect several factors beyond customer-service automation.
The potential benefits extend beyond staffing costs. An AI system can respond outside normal working hours, so a customer who sends a late-night message can receive an answer without waiting until the following morning. Routine enquiries can be handled separately from complaints and unusual cases requiring human attention.
Response speed and the limits of automation
FastCom is now being introduced to larger corporate businesses in Bangladesh. A business does not need to receive tens of thousands of messages a day to have a response-time problem. FastCom says some businesses it previously worked with took an average of around two and a half hours to respond to an enquiry. In some deployments, that has fallen to roughly 25 seconds with AI handling the initial interaction.
"Companies are not looking at this only as a cost-saving tool," said Yaseen Nur, founder and CEO of Ascend AI. "For some of them, the bigger issue is that a customer should not have to wait two or three hours for an answer the company already has."
Human intervention remains essential. One frustration with earlier chatbots was that customers could remain stuck in an automated conversation after the system had reached the limits of its ability to help. FastCom includes human handover in its operating model. If the AI is not confident it can handle an interaction, the conversation can be escalated to a representative with its previous context attached, so the representative can continue the exchange without asking the customer to repeat everything.
Businesses can also determine the extent of the AI's authority. For sensitive uses, responses can be restricted to approved company information, and businesses can log actions for review and limit access to the systems and data required for a particular task. These controls become more consequential as the technology gains the ability to act. An incorrect answer can cause problems, but an unauthorised change to a customer record, an incorrect service request or an unapproved promise can have wider operational consequences. FastCom says its enterprise architecture includes confidence-based escalation, approved-answer controls and audit trails for AI actions, along with encryption, role-based access, and configurable data-retention controls.
Why this matters for customer support professionals
Agentic AI changes the division of labour in customer service teams. Routine, high-volume enquiries - order status, product availability, appointment bookings - shift to automation, while human agents concentrate on exceptions, escalations and cases requiring judgement. For support professionals, the practical skill becomes configuring and supervising these systems: deciding which actions the AI can perform automatically, which require approval, and when a conversation must transfer to a person. The technology is still in early stages in Bangladesh, and FastCom describes a structured deployment process where the platform connects to relevant business information and systems, then tests and adjusts against real customer traffic before wider rollout. Support teams that learn to manage AI for Customer Support - including escalation rules, approved-answer controls and audit trails - will be positioned to oversee these systems rather than compete with them for routine ticket volume.
Your membership also unlocks:
