Business automation in Bangladesh: 3 real projects and the numbers behind them

"Automation" gets sold in Bangladesh as a magic word. Business owners hear it, nod, and still don't know what will change on Monday morning.

So here are three projects I worked on before founding Arwahyn, with the problem, the build and the result. All three are Bangladeshi businesses with ordinary teams and ordinary budgets.

1. Boibari.com: reply rate from 18% to 71% in 14 days

The business: an online bookstore and publisher in Dhaka.

What was broken. Customers reached out by phone, WhatsApp and the website chat. Each channel lived in a different place. Some customers got two answers from two different people. Many got none. The contact list had the same customer saved four or five times, with different spellings.

Only 18 out of every 100 people who reached out got a useful reply in time.

What we built.

• Cleaned the contact database first and moved it into one CRM. Automating on top of duplicate data only makes the mess faster.
• Rebuilt the response flow: who gets which message, how fast, in what order, and when to follow up.
• Put phone, WhatsApp and web chat into one routed inbox so each conversation has exactly one owner.
• Added AI agents to answer common questions (stock, price, delivery time) across voice, WhatsApp and text.
• Built a live dashboard showing lead volume, reply times and whether any part of the system had failed.

The result. Reply rate went from 18% to 71% in 14 days. Double answers stopped. The team stopped asking "did anyone reply to this person?" in the group chat.

The lesson for you: before you buy a chatbot, count how many conversations get no answer. That number is your real opportunity.

2. Prodhan.com: 38% more abandoned carts recovered

The business: an e-commerce store selling online across Bangladesh.

What was broken. People added items to the cart and left, and nobody knew who they were. Every order that did come in was typed again into the CRM, the courier portal and the finance sheet. Three people did the same data entry.

What we built.

• Event-driven messages on WhatsApp and email: order confirmed, shipped, out for delivery, abandoned cart, and a follow-up after delivery.
• One automation layer connecting WooCommerce, the CRM, the courier API and finance, so an order is entered once.
• AI answers for product questions (size, material, delivery area), with a person reviewing the edge cases.
• A holdout group: a slice of abandoned carts got no message, so we could measure what the automation actually added.

The result. Cart recovery went up 38% on the measured flows. Manual re-entry of orders dropped to zero.

The lesson for you: measure against a holdout. Without it, every automation "works", and you never learn which message earns its cost.

3. Biddabari: admissions and support answered 24/7

The business: an EdTech company, part of a group of three sister companies.

What was broken. Students and parents asked about admission, fees and payments at night and on weekends. Leads came from Facebook ads, the website and Messenger, and landed in different places. Nobody owned them and nobody had a deadline.

What we built.

• An AI first-response layer on biddabari.com and WhatsApp that answers admission, payment and support questions, creates the CRM record, tags where the lead came from, and passes real cases to a person.
• Lead scoring and routing: every lead gets an owner and a follow-up deadline, and the manager is alerted when one is missed.
• Data rules (no duplicates, required fields, source tags) so the reports could be trusted at review time.
• Over 60 production workflows, each with retries, error handling and an alert that names the exact record that failed.
• Written SOPs and training so the non-technical team could run the system themselves.

The result. Every enquiry gets a first answer in seconds, day or night, and only the cases that need a person reach the team. The same group later ran orders for all three companies through one shared automation layer. Read that case.

The lesson for you: plan the hand-off from AI to a human before launch. The bot is only as good as the moment it knows to step aside.

What the three projects have in common

None of them started with a tool. Each one started with a count: unanswered messages, abandoned carts, leads without an owner. Then we fixed the data, built the smallest flow that moved that number, and put an alert on it so failures showed up the same hour.

That's the method we use at Arwahyn for every client. The full case studies, with stacks and step-by-step builds, are on our work page.

What would this look like for your business?

Pick the one number you'd most like to change. Reply rate, COD returns, cart recovery, leads followed up. Then book a free 30-minute call and we'll map the smallest automation that moves it, with a fixed price.