Automating order, shipping, return and product questions with the knowledge base in e-commerce

In e-commerce the support load does not come from one place. "Where is my order?", "is this size available?", "how do I return this?", "do you offer cash on delivery?" — the same questions arrive every day, from every channel, dozens of times. Most teams answer them one by one on WhatsApp, Instagram DM and web chat, and during campaigns they simply cannot keep up. This article explains automating order, shipping, return and product questions with the knowledge base — that is, how you build the answer engine of your e-commerce operation on intusell. This is the order-and-product piece of our sector-specific "how to train your AI" series; for the e-commerce pillar of the series, see how to train your intusell AI.

Quick answer

Why a knowledge base

There are two reasons to move order and product questions to the AI: to reclaim the hours spent on the same questions, and to give consistent, correct information on every channel. In manual mode each agent answers from memory and phrasing of their own; the return condition varies from person to person, the shipping time is sometimes stated wrong. In connected mode there is a single source of truth, and the AI reads the same text no matter which channel the question comes from.

Think of the concrete difference like this:

This is where intusell's positioning as "not a chatbot, but a senior salesperson" becomes concrete: a senior advisor does not state the return condition off the cuff, they check the policy; they do not guess the size, they verify it.

Who is it for?

• Online stores and brands selling through WhatsApp, Instagram and their website.
• Support teams answering repeat questions — "where is my order", "is this size available", "how do I return" — dozens of times a day.
• Businesses whose message volume multiplies during campaigns and sales, where the team cannot keep up.
• Teams that want to give the same information consistently across multiple channels and avoid contradictions.

No technical knowledge is required; setup is done from the panel.

Two kinds of information: fixed text vs. customer-specific status

Separating two distinct kinds of information is the foundation of e-commerce answers. intusell reads them from different places:

Fixed text does not change from customer to customer: "Orders ship within 1-3 business days" is the same for everyone. You upload it to the Knowledge Base once, and the AI reads it from there every time. Customer-specific live status — "where exactly is my order" — is not static text; we handle that separately below. Keeping this distinction clear is what lets the AI give the right information from the right source.

1. Setting up the Knowledge Base (RAG)

The foundation of the answer engine is the Knowledge Base. You upload all of your store's fixed information here. intusell accepts several formats:

• PDF: Return policy, distance sales contract, shipping terms, product catalogs.
• Excel / CSV: Product list, size/color chart, feature matrix, a list of frequently asked question-answer pairs.
• Web URL: Your existing FAQ page, your shipping information page, your product descriptions — you give the link and the content is pulled in.
• Free text: Short rules you want to write quickly (such as "Cash on delivery is only available on orders under a set amount").

Each source you upload is chunked and made searchable with pgvector. When the AI answers a question, it automatically finds the most relevant chunks via match_knowledge_chunks and builds its answer only from those. From here, you will find a step-by-step guide to shaping the AI to answer correctly in how to train your intusell AI.

2. What information you should upload

In e-commerce, almost all repeat questions cluster under a few headings. Feed the Knowledge Base with these groups and the AI is ready for the questions:

An important distinction: policy and description text comes from the Knowledge Base; person-specific live status is handled separately. You do not mix the two, so the AI reads each piece of information from the correct source.

3. Unifying channels in a single inbox

Order and product questions come from every channel; the AI's value is in answering all of them in one place with one source of truth. intusell merges these channels into a single inbox:

• WhatsApp — connects via QR (no Meta approval required, ~1 minute) or via the official Cloud API.
• Instagram DM — connects with one-click Meta OAuth.
• Facebook Messenger, Telegram, web chat, email — all in the same inbox.

One caveat matters: Instagram comment replies are not yet live for everyone; this automation rolls out gradually, subject to Meta approval. Your first step is WhatsApp and Instagram DM, which are approved. The DM gate (instagram_dm_enabled) and the comment gate (instagram_comments_enabled) are managed separately; while you wait for comments, you operate at full capacity on DM and WhatsApp. You will find the field side of channel setup in detail in Instagram and WhatsApp automation.

4. The most critical rule: the AI does not invent information

This is the most important trust point of knowledge-based answers, and in e-commerce it touches customer satisfaction directly. The AI builds its answer only from the sources in the Knowledge Base. If a piece of information is not there — the stock status of a product you did not upload, a campaign condition you did not define — the AI does not guess. Instead, it:

1. Recognizes that the information is not in the sources.
2. Does not tell the customer a made-up figure, date or condition.
3. Honestly states that it does not know and hands the conversation off to your team.

So it does not say "it will probably arrive tomorrow" or "I think that size is in stock" — the kind of sentences that breed returns and complaints later. Saying it does not know is far more valuable than giving a wrong delivery date. This behavior works together with the AI's human handoff (lock) modes.

One point to clarify: intusell has a protection layer, but by default it runs in observation (shadow) mode — it detects and flags risky situations, it does not apply a hard block. What actually keeps the AI correct is the sector-specific system prompt that constrains it from inventing information in the first place; the protection layer adds scanning and flagging on top. So the accurate framing is not "we block the AI when it says something wrong" but "the AI is constrained so it will not invent, and the protection layer also scans and flags".

5. Customer-specific order and shipping status

The question "where exactly is my order?" cannot be answered with fixed text; it is person-specific. intusell handles it in two ways:

• Guidance: Based on the Knowledge Base, the AI explains where to find the order number, how to use the tracking link and the carrier's tracking page. Most "where is my order" questions are solved by guiding the customer to the right tracking step.
• Handoff: When a customer-specific live status lookup is needed or the topic becomes sensitive (damaged item, lost shipment, payment dispute), the AI hands the conversation to your team per your answer rules. This is a typical hybrid handoff scenario.

The key is that the AI does not invent an estimated delivery date. It gives policy information like "orders ship within 1-3 business days", but it does not promise what it cannot verify, such as "your package will be at your door tomorrow at 2 p.m."

6. Human handoff modes

In e-commerce not every conversation is the same; some the AI can carry easily, some call for a human touch. intusell manages this with three lock modes:

• ai_only — the AI answers on its own. Ideal for repeat order/product questions.
• hybrid — the AI answers, but in the situations you define (return disputes, damaged items, high-value orders) it hands the conversation to your team.
• human_only — the team carries the conversation entirely; you choose this mode during a sensitive complaint process.

One point to add: a customer sending a file, image or voice attachment is not, on its own, a reason for human handoff. The AI assesses the attachment; the handoff decision is made based on the content of the topic and your answer rules. You will find how these handoffs are managed in daily operations in the inbox in how to use intusell.

7. Re-engaging the quiet customer

In e-commerce many customers fill a cart but do not complete it, or ask a question and then disappear. intusell's CRM and proactive follow-up layer sends an opportunistic re-offer to suitable customers who are in follow-up. Let us correct a misconception: there is no managed "waiting list". The accurate description is that an opportunistic re-offer goes to suitable WhatsApp customers in follow-up, at the right moment. An interested customer who has gone quiet is re-engaged automatically, so every question can turn into a sales opportunity.

What it isn't

• It is not an e-commerce platform (store software). intusell answers order and product questions; taking payment, managing stock and hosting product pages is the job of your store platform. intusell is the sales and support layer that talks on top of it.
• It is not a shipment tracking system. It guides the customer to the right tracking step; it does not determine the carrier's location itself.
• It is not a bot that invents information. It answers only from the sources you upload to the Knowledge Base; it hands off what it does not know to your team.
• It is not an e-commerce-specific priced product. The knowledge base and multichannel answering come in every plan; the pricing model is based on messages and voice minutes and does not vary by sector. For plan details, see the pricing page.

Frequently asked questions

How do I teach intusell my order, shipping and return information?

There are two kinds. Fixed policy and product text — shipping times, return/exchange terms, size charts, payment methods — you upload to the Knowledge Base (RAG) as PDF, Excel, CSV, web URL or free text, and the AI searches it on every answer. Customer-specific live order/shipping status is not static text; this article explains how that is handled.

Will the AI invent a wrong shipping time or return condition?

No. The AI builds its answer only from the sources you upload to the Knowledge Base. If a piece of information is not in the sources, the AI does not guess a figure or condition; it says it does not know and hands the customer off to your team. This is the most critical trust message in e-commerce.

Which channels can I unify in a single inbox?

WhatsApp (via QR, no Meta approval required, or the official Cloud API), Instagram DM, Facebook Messenger, Telegram, web chat and email all merge into one inbox. Instagram comment replies roll out gradually, subject to Meta approval.

Can the AI tell a customer where their order is?

The AI first guides the customer: from the Knowledge Base it explains where to find the order number, how to use the tracking link and the carrier method. When a customer-specific live status lookup is needed, it either retrieves the information per your answer rules or hands the conversation to your team. The AI does not invent an estimated delivery date.

Will I stop answering the same questions over and over during busy campaigns?

Yes. The AI answers most repeat questions — "where is my order", "is this size available", "how do I return" — 24/7. Your team only looks at genuine exceptions and handed-off conversations.

Is it hard to keep the Knowledge Base up to date?

No. When a shipping time changes or you update your return policy, you edit the relevant source in the panel; the AI uses the updated text on its next answer. There is a single source of truth, so you do not update channel by channel.

Next step

After you build the answer engine, you go in two directions. To shape the AI with your products, tone and sales style, read how to train your intusell AI. To see how incoming questions and handoffs are managed day to day, move on to how to use intusell. You will find the channel-connection side in Instagram and WhatsApp automation, and the sector overview on the e-commerce solution page. For the same approach in other sectors, see the tour agency and clinic pillar articles. You can reach all articles via the blog.

If you would like us to set up your Knowledge Base together, Request a demo or write to hello@intusell.com; we will upload your shipping, return and product information in 20 minutes and test the first automatic answers together.