When a customer sends a photo instead of a product name

Updated 2026-07-26 · 7 min read

A screenshot lands in your inbox. No product name, no size, sometimes a competitor's watermark still sitting in the corner. Under it, one line: "eta ache?"

Most sellers treat that as a lazy message. It is the opposite: the closest thing to a placed order that arrives on its own.

A photo is a decision, not a question

Someone who types "dam koto?" is still shopping. Someone who sends a photo has stopped shopping. They have already picked the exact thing they want and are now only looking for a shop that has it. Nothing is left to persuade them of — just availability, price, and delivery.

That is why a photo is the most valuable message in your inbox, and why losing one costs more than losing a "dam koto?". Reply speed decides most Facebook orders, but on a photo message what you say matters as much as how fast you say it.

The photo usually comes from one of four places:

• A competitor's page. They are checking whether you have it cheaper. Price and stock, immediately.
• A friend's story or a reshare. They have no idea of the brand or the model name. They cannot type it even if they want to.
• Your own old post. Possibly four months old. Possibly repriced, possibly gone.
• A Reel. Not an image at all — it arrives as a shared link.

Three situations, three replies

Every photo message is one of three cases. Most sellers only have a script for the first.

Whichever case you are in, the reply needs four things before the buyer can act:

1. A name. They cannot say yes to a thing with no name.
2. That product's own price. If the photo is the digital model, quote the digital price — not the cheaper analog one because it was easier to find.
3. The delivery charge, inside Dhaka and outside, since you do not know which applies.
4. One question at the end. Size, colour, quantity. A reply that ends in a full stop ends the conversation.

"Eta nai" is where the money leaves

Two words, completely accurate, and the single most expensive thing you can type. The buyer did not ask for permission to leave — you gave them the exit.

The person messaging you is not attached to one line in your stock list. They want a black cotton kurti with that neckline, at roughly that price, delivered this week. Your exact item is one way to get that; two others in your catalogue are also ways to get it.

"Eta nai" is a true sentence that ends a sale. "Eta nei, tobe ei duita ache — dekhben?" is a true sentence that continues one.

The rule is simple: never send a no without two alternatives attached, both with photos and both with prices. It costs thirty extra seconds of typing.

When the photo is your own old post

The buyer screenshots a post from four months ago and asks for it at the price printed in the image.

Do not argue with the screenshot and do not say "purono post". Say what is true now, in one line, and move forward: the price today, or the closest thing currently in stock. If you honour the old price, say that you are honouring it — that is a favour, and unstated favours earn nothing.

Customers send Reels too, and share links change every time

Plenty of buyers do not screenshot at all. They hit share on a Reel and send it.

There is a technical trap here. A Facebook share link carries a per-share token — facebook.com/share/r/<token> — and that token is different every time the same reel is shared. So a link you saved yesterday will never be the same string a customer sends tomorrow. Matching them literally does not work.

Tanet Chat normalises both sides — the customer's link and your saved trigger — down to a stable video id, resolving share tokens where it can. When a customer's reel maps to one of your FAQs, that FAQ's answer goes out directly with its images and links attached, without the AI in the loop at all.

Two honest caveats. Direct facebook.com/reel/<id> and /videos/<id> links need no lookup and always work. A /share/ token needs one fetch to resolve, and if Facebook throttles that fetch, the message quietly falls through to the normal AI reply instead of matching. So when you save a reel as a FAQ trigger, use the direct reel URL if you can get it.

How Tanet Chat matches a photo to your catalogue

There are two layers, and only the first is on for everyone. Both sit in front of the ordinary reply machinery — price cards, variant photos and order capture work the same whether the customer typed a product name or sent a picture.

Layer one, on by default. The photo is downloaded once and handed to a vision-capable model together with your catalogue. The instructions it works under are deliberately conservative:

• An unmistakable match to exactly one product: name it, and give that product's price and details — explicitly not a cheaper look-alike's.
• Ambiguous, because two of your products look nearly identical in a photo: show all the candidates with their prices and ask which one. It is told to default to this whenever it is not certain, rather than guess.
• Not in the catalogue: say so plainly and point back to what is available.
• Not a product at all — a meme, a selfie: acknowledge it briefly and return to selling.

Once your catalogue passes about 50 products, the agent stops sending the whole list and sends the slice that matches the customer's words — and "eta ache?" matches nothing in particular, so large catalogues depend on products being named the way customers say them. A photo sent with no text at all skips that narrowing and goes out against everything you sell.

If no model that can see images is available when the photo lands, it does not fail silently. It drops the image and asks the customer to describe the product or type its name.

Layer two, off by default. Reference matching compares the customer's photo against your real product photos, image to image, before the reply is written. When more than four of your products have photos, a first pass — the customer's picture plus your product names — shortlists four; those four reference photos are then compared against the customer's picture in a single call. It only counts as a match at 70% confidence or higher, and if the first model is not sure, it retries on a different one before giving up. On a confident match the reply leads with that product and does not ask the buyer to confirm which item they meant.

There is no switch for this in the dashboard. It is a flag on your account (vision_config.reference_matching), off until someone at Tanet turns it on, so you have to ask. Ask if your catalogue has look-alikes — the same shirt in six colours, analog and digital versions of one watch. If every product you sell looks different from the rest, layer one is enough.

One hard limit on this layer: only active products with at least one photo can be matched picture-to-picture. A product you typed in without an image still sits in the catalogue as text, and layer one can name it from its description — but nothing can match it photo to photo.

What to do this week

• Photograph every product you actually sell. One clean shot, plain background, in the catalogue. This is the single change that makes photo messages answerable. The free plan holds 50 products and one page; Starter at ৳599/month removes the product cap.
• Write your three replies once — have it, have something close, do not have it — and keep them where you can paste from.
• Take your three best-performing Reels and save each one as a FAQ, with the reel link as the trigger phrase and the answer, price and photo attached.
• Delete "eta nai" from your vocabulary. Replace it with a no plus two alternatives, every time.

Access is invite-only for now — apply for a page and the setup happens with you, not in a signup form.