Every online seller in Bangladesh has heard that "AI will change everything". But when you run a Facebook page with 30 orders a day, the real question is smaller and more useful: what can AI for eCommerce actually do in my shop this week, what will it cost, and where can it go wrong?
This guide answers exactly that. You will see eight practical ways a small online shop can use AI, the jobs you should never hand to it, a worked cost example in taka, and a simple 30-day plan. No hype, no invented numbers — just what works when you sell with cash on delivery, couriers and a busy inbox.
What "AI for eCommerce" really means for a small shop
For a small seller, AI is not a robot that runs the business. It is a very fast assistant that is good at reading and writing:
- It can read a customer's message, a product photo, a price list or a day's sales numbers.
- It can write a reply, a product description, an SEO title, a Facebook caption or a short summary.
That is it. It does not know your stock, your prices or your courier rules unless you tell it. It can sound confident even when it is wrong. So the whole skill of using AI in eCommerce is simple: give it your real facts, give it a narrow job, and keep a person in charge of money and promises.
When you get this right, the time saved is real. Many sellers find that the work that eats their day — typing the same answer 50 times, writing product pages, checking numbers at night — is exactly the kind of reading and writing AI does well.
8 practical ways to use AI in your online shop
1. Write product pages from a photo
Writing a good product page takes time: a name, a short description, a long description, the fabric or material, sizes, tags and an SEO title. For a shop that adds 20 new items a week, that is hours.
An AI model that can see images can look at your product photo, take the price you type, and write a first draft of all of it. Your job becomes checking and correcting, not writing from a blank page.
Tip: always type the price and the key facts (fabric, size range) yourself. The AI should describe what it sees, never guess what it cannot see.
2. Answer common customer questions in Bangla and English
Most messages to an online shop are the same few questions: "price koto?", "delivery charge koto?", "Dhakar baire kotodin lagbe?", "COD ache?", "size chart din". AI can answer these in natural Bangla, Banglish or English, any time of day — if it answers only from facts you gave it.
The good pattern is:
- You write your facts once (delivery charges, delivery times, return rule, payment methods).
- The AI answers from those facts only.
- Anything else — a complaint, a special request, a photo of a damaged item — goes to a person.
If you want to go deeper on reply habits, read our guide to customer service for online shops.
3. Draft replies for your team instead of replacing them
You do not have to let AI answer customers by itself on day one. A safer first step: the AI writes a suggested answer and your seller reads, edits and sends it. Your team becomes faster, the tone stays human, and you learn where the AI is good and where it is weak before you trust it with more.
4. Write SEO titles and meta descriptions
If you have your own website, every product and category needs an SEO title (about 50–60 characters) and a meta description (about 140–160 characters). Few small shops ever write them, so their pages show up in Google with messy text — or not at all.
AI is very good at this short, structured writing. Give it the product name, price and category, ask for a title and description under those lengths, and check that it did not invent offers or prices.
5. Write Facebook captions and ad text variations
For a new product, ask AI for five caption ideas in different styles: one short, one with a question, one with the offer first, one in Bangla, one in Banglish. Then test the best two. AI saves you from the blank page; you still choose what fits your brand.
Never let it invent a discount, a "limited stock" claim or a delivery promise you cannot keep. Customers remember broken promises, and Facebook can reject ads that make claims you cannot back up.
6. Get a daily business summary in plain words
At the end of a busy day, most owners do not want a spreadsheet. They want to know: How did today go? What is stuck? Who is behind? What should I do tomorrow morning?
AI can read your sales, delivery and money numbers and turn them into a short summary in plain Bangla or English, with the three most important actions. This is one of the most valuable uses of AI for an owner — but only when it reads your real numbers. An AI that has not seen your data can only guess.
7. Turn your own documents into answers
You probably already have a price list in Excel, a return policy in Word, or a size chart as a PDF. Modern AI can read these files and learn the facts in them, so you do not have to type everything again. When you change the price list, update the file so the AI does not keep old prices.
8. Find patterns in returns and complaints
Paste a week of return reasons or complaint messages into an AI chat and ask: "Group these by reason and tell me the top three." You may see that most returns come from one size, one courier area or one product photo that looks different from the real item. That is the kind of insight that cuts returns — see our guide on how to reduce courier returns for what to do next.
What AI should never do in your shop
AI makes mistakes in a particular way: it sounds sure even when it is wrong. In eCommerce, a confident wrong answer costs real money. Keep these jobs with people:
- Setting or changing prices. An AI that "rounds" ৳1,450 to ৳1,400 in a reply has just given away ৳50 per order.
- Promising delivery dates. Couriers have delays; a promise you break becomes a return.
- Approving refunds, exchanges or money entries.
- Handling angry customers. A complaint needs a person who can apologise, decide and fix.
- Anything it was not told. If the answer is not in your facts, the right reply is "let me check with the team", not a guess.
A good AI setup for a shop is built around this: it answers only from your facts, says plainly when it does not know, and hands the chat to a person when the customer is upset, sends a photo of a problem, or asks for a human.
How much does AI cost? A worked example in taka
AI companies charge by "tokens" — small pieces of text, roughly three to four English characters each (Bangla text uses more tokens for the same words). You pay a small price per million tokens the AI reads and a higher price per million tokens it writes.
Here is a worked example with clearly stated assumptions, not a quote:
- A fast, low-cost model priced at US$0.75 per million tokens read and US$3.75 per million tokens written.
- Each customer answer reads about 3,000 tokens (your shop facts plus the chat) and writes about 200 tokens.
- A dollar costs ৳125.
| Item | Calculation | Cost |
|---|---|---|
| Reading per answer | 3,000 × $0.75 ÷ 1,000,000 | $0.00225 |
| Writing per answer | 200 × $3.75 ÷ 1,000,000 | $0.00075 |
| Per answer | $0.003 × ৳125 | about ৳0.38 |
| 100 answers a day | ৳0.38 × 100 | about ৳38 a day |
| 30 days | ৳38 × 30 | about ৳1,140 a month |
Now compare that with the time it replaces. If those 100 answers would take a seller 2 minutes each, that is more than 3 hours of work a day. Your numbers will be different — longer facts, photos and bigger models cost more — but the method is the same: estimate tokens per job × price × jobs per month, and check the real bill after the first month.
A 30-day plan to start using AI for eCommerce
Do not switch on everything at once. Start where the risk is lowest and the time saved is highest.
Week 1 — Write your facts. One page: delivery charges inside and outside Dhaka, delivery times, payment methods (COD, bKash, Nagad), return and exchange rule, office hours, size chart. This page is useful even without AI — your sellers need it too.
Week 2 — Use AI for writing only. Product pages from photos, SEO titles, captions. Nothing goes to customers without a person checking it. Note how many corrections you make.
Week 3 — Suggested replies. Let the AI draft answers in your inbox; sellers edit and send. Each time a seller corrects the same mistake, add the right fact to your facts page.
Week 4 — Automatic answers for simple questions, at quiet times. For example, only outside office hours or when no seller has answered for a few minutes. Keep the hand-over rules strict. Check a sample of AI chats every day.
At the end of the month, look at three numbers: hours saved, the AI bill in taka, and how many chats needed a person anyway.
eCommerce Autopilot has this built in. Every store gets Autopilot AI with no AI key to set up: 1-Click AI Products writes product pages from your photos and price, Teach your AI learns your shop by chat, voice or your own files, the website chat (and the Facebook, Instagram and WhatsApp inbox on Premium) can answer from what you taught it, and the store owner's AI CEO Assistant writes a daily brief from your real numbers. You can start free and see it in your own shop.
Your AI for eCommerce checklist
Use this before you let AI near customers:
- I wrote one page of shop facts (charges, times, payment, returns, hours).
- The AI answers only from those facts and says "let me check" when it does not know.
- Prices, discounts, refunds and delivery promises stay with a person.
- Angry customers, photos of problems and "I want a human" go to a person at once.
- I check a few AI answers every day for the first month.
- I know what each AI job costs and I look at the bill monthly.
- When my prices or rules change, I update the facts the same day.
Common mistakes sellers make with AI
Letting it guess. An AI with no facts will still answer. That answer may be wrong about the one thing the customer cares about: the price.
Copy-pasting without reading. AI product descriptions sometimes add details that are not true ("100% cotton" when it is a blend). One wrong detail creates returns.
Using it to sound like everyone else. If every caption reads like a generic template, customers stop noticing. Use AI for the first draft, then add your shop's voice.
Forgetting the cost. Small per-answer costs add up. Know your monthly number before you switch on automatic answers for every chat.
Expecting it to fix a weak business. AI makes a good process faster. It cannot fix bad products, slow delivery or unclear prices. If you are still setting up the basics, start with what to automate first.
The one thing to do today
Open a blank page and write your shop facts: delivery charges, delivery times, payment methods, return rule and office hours. That single page is the foundation of every good use of AI for eCommerce — and it will make your human team faster today, even before any AI reads it.