AI assistant for an online store
It answers about availability, delivery times and order status with numbers from your own system. An order that has gone wrong reaches a person before the customer writes again.
- Buildfrom $1,700
- Timeline2–4 weeks
- Supportfrom $570/mo
Who this suits: Stores where «is it in stock?» and «where is my order?» make up almost the whole correspondence, and each answer means somebody opening the admin panel.
What is in the first version
- item availability
- delivery times
- order status
- operator handover
The figures above are the package this sits on: AI agent or process automation. Anything past its edges the calculator adds before the work starts, not after.
How the path goes
- 01
First work out what is asked
Until the question is tied to a specific item or order, the assistant asks rather than answers.
- 02
The number comes at answer time
Stock and dates are read from your source right then, and if it is silent the assistant calls a person.
- 03
A problem order needs no complaint
An order stuck in one status, returned by a courier or cancelled lands on an operator’s list itself.
What decides whether this works or annoys people
Storefront stock is not warehouse stock
Answering yes from the storefront figure sells what is not there, so the source of truth is named before the build.
A delivery time is a promise
It comes from the carrier’s own terms, is never rounded down, and for a back order no time is given.
A third message wants a decision
A repeat, or any wording about money, a return or a complaint, goes to an operator with the whole history.
What to measure once it is live
- share of messages closed without an operator
- time to an answer at a sale peak
- cancellations after an availability confirmation
- repeat messages about a single order
Why this comes out faster
The shape of this one is known: the states, the edge cases and the things that usually go wrong have been decided before. Nothing here is a template, and the saving is not in your half of the work. It goes into your process, your content and the systems this has to talk to, which is the part nobody can have solved in advance.
A likely stack for this
Picked against the task when we scope it, not decided in advance. This is the shelf it usually comes off.
- Assistant and modelsPythonFastAPIOpenAIAnthropicpgvector
- Data and integrationsPostgreSQLRedisn8nDocker
- Channels and monitoringTelegram Bot APIClickHouseGrafana
Price it yourself, right here
Five steps, and you can see the number without leaving a contact. The estimate accounts for the kind of work, what you already have and what it will need inside.
These bills do not come from us
- AI model calls at the provider’s rate
- Hosting and the server, sized for peaks
- Paid access to carrier APIs
- SMS and push about order status
Asked before the first call
- Where does it get availability from?
- From your accounting system or the warehouse, at the moment of the answer. We do not ship a version that answers from yesterday’s export and keeps quiet about it.
- Can it place an order?
- It assembles the basket and takes the customer to your own checkout, but takes no money in a chat. Changing an existing order goes through a confirmation.
- What happens at a peak, when the volume multiplies?
- Handover thresholds are set before the season, not inside it. At a peak the first thing that changes is caution: on late days no delivery time beats the usual one.