Bangladeshi businesses begin testing AI agents that do more than answer questions

For years, businesses have used chatbots to handle common customer questions. For many customers, however, the experience has remained frustrating.

The reason is relatively simple.

Most earlier chatbots were built around a fixed set of questions and answers, which often meant they struggled as soon as a conversation required information from internal systems or something needed to be done.

A newer category of technology, commonly described as agentic AI, is designed to go a step further.

Instead of only generating a reply, an AI agent can try to understand what a customer wants, determine what needs to happen next and, within limits set by the business, carry out that task.

The difference is easiest to see with a routine request.

If a customer asks where an order is, a conventional chatbot may not have access to the information needed to answer. An AI agent connected to the company's order system can look up the order and return its status.

Depending on how the system is configured, the same agent could also create a complaint, start a return request, book an appointment, update a customer record or pass the conversation to a human representative.

This shift from answering questions to completing tasks is now beginning to appear in Bangladesh.

From that point onwards, the distinction between a chatbot and an AI agent becomes less about how naturally the software can speak and more about what it is able to do.

From customer message to completed task

FastCom AI is designed as a service layer between customers and a company's existing systems.

A business can connect customer conversations coming from platforms such as Facebook Messenger, WhatsApp, Instagram, web chat and email, while the AI can also be integrated with internal systems including CRM software, ticketing platforms, order and inventory systems, ERP software and company knowledge bases.

This means the AI does not necessarily have to respond from a static list of answers.

If the required information exists in an approved company database or business system, the agent can retrieve that information and use it to respond.

For an e-commerce business, this could mean checking whether a product is in stock, confirming an order or providing a delivery update.

For an electronics company, it could mean answering warranty questions, registering a service request or checking repair status.

An automobile company could use a similar system for model and price enquiries, test-drive bookings and after-sales service questions. Healthcare providers could use it for appointment bookings, doctor availability or report-status enquiries.

FastCom is also being designed around the way customers communicate in Bangladesh. The platform supports Bangla, Banglish and English, allowing conversations to move between languages rather than requiring customers to follow a rigid format.

This is particularly relevant for businesses where customer interactions rarely follow a clean script.

A customer may ask about a product in Bangla, type the model name in English, send a Banglish follow-up and then ask about delivery, all within the same conversation.

Early results in F-commerce

Some of FastCom's earliest deployments in Bangladesh have been with F-commerce businesses, where a significant part of the sales process takes place inside Facebook and other messaging platforms.

According to FastCom, its AI system is currently handling around 95% of customer conversations and the order flows associated with them across some of these deployments. Cases requiring human judgement are passed to staff.

One seller provides an example of how the economics of customer service can change.

Before FastCom was introduced, the business had around 20 people involved in customer management. After the deployment, FastCom says the operation was reduced to two representatives who primarily handle exceptions and escalations, while the AI manages most routine interactions.

The company says customer-management costs for the business fell by more than half. Revenue also increased by around 20% in the period following the rollout, although changes in revenue can be influenced by several factors beyond customer-service automation.

For the seller, the benefits are not limited to staff costs.

An AI system can continue responding outside normal working hours. A potential customer messaging late at night does not necessarily have to wait until the following morning, while routine questions no longer remain in the same queue as complaints or unusual cases that genuinely require a person.

The technology also allows the business to maintain a more consistent response process regardless of when the customer sends the message.

For larger companies, the problem is often speed

FastCom is now also being introduced to larger corporate businesses in Bangladesh.

Their reasons for experimenting with the technology are not always the same as those of high-volume F-commerce sellers.

Many corporate businesses may not receive tens of thousands of customer messages every day. But they can still have a response-time problem.

FastCom says some businesses it is working with previously took an average of around two and a half hours to respond to a customer enquiry. With AI handling the initial interaction, that has fallen to roughly 25 seconds in some deployments.

"Companies are not looking at this only as a cost-saving tool," said Yaseen Nur, founder and CEO of Ascend AI, the company behind FastCom 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."

For these businesses, AI can therefore become part of the brand experience rather than simply a tool for reducing headcount.

A customer contacting a large company may reasonably expect a faster response than one received hours later, particularly when the question involves information already available inside the organisation.

What happens when AI cannot solve the problem?

One of the weaknesses of earlier chatbots was the tendency to keep a customer trapped inside an automated conversation even when the system could no longer help.

AI agents still need a way out.

FastCom includes human handover as part of its operating model. If the AI is not confident about how to handle an interaction, the conversation can be escalated to a human representative with its previous context attached.

The human agent therefore does not necessarily have to ask the customer to repeat the conversation from the beginning.

Businesses can also determine how much freedom the AI receives.

For more sensitive use cases, responses can be restricted to approved company information. Actions can be logged and reviewed afterwards, while access can be controlled based on the systems and data the AI needs.

These safeguards become more important as the technology becomes more capable.

A chatbot giving an incorrect answer is one problem. An AI agent making an unauthorised change inside a CRM, creating the wrong service request or making a promise the company did not approve can have more serious consequences.

FastCom's enterprise architecture therefore includes features such as confidence-based escalation, approved-answer controls and audit trails for AI actions. The company also lists encryption, role-based access and configurable data-retention controls among its enterprise security features.

Still an early market

Despite the rapid development of the technology, agentic AI in Bangladesh remains at an early stage.

For businesses, deploying these systems is not simply a matter of turning on a chatbot.

The AI needs access to accurate company information, clear policies and appropriate integrations. Businesses also need to decide which actions the agent can perform automatically, which require approval and when the conversation must be transferred to a person.

FastCom's enterprise model reflects this through a structured deployment process in which the system is first connected to the relevant business information and systems, then tested and tuned against real customer traffic before wider rollout.

This may ultimately determine whether AI agents become a significant part of customer service in Bangladesh.

The earlier generation of automation focused on answering questions faster. The newer generation is attempting something more ambitious: allowing software to participate in the work that happens after the question is asked.

For businesses, that creates the possibility of faster responses, round-the-clock customer service and lower operating costs.

It also creates a much higher requirement for accuracy, security and control.

The technology will therefore be judged not only by how naturally an AI can talk to a customer, but by whether businesses can trust it to know when to answer, when to act and when to hand the conversation back to a person.