Solutions / 
Industries

AI customer service for retail: a concierge connected to your shop and order systems

From the first product question to returns and the next purchase: what the concierge can answer, what it can change, and what stays with your team.

Where is my order, can I still change it, how do I send it back, which model fits me? The answers sit in your shop, product, order and carrier systems. An AI concierge that can read those systems answers with the customer's own order in front of it, in the channel and language the customer chose. Where you allow it, it also makes the change, and it hands a case to your team when judgment matters.

Below, stage by stage: what customers need, what we've learned from implementations about the data and decisions behind each stage, and how Typewise handles it.

  1. Discover and chooseAdvice before buying"Which of these fits me?"
  2. BuyQuestions at checkout"Can I still change the size?"
  3. Wait for deliveryWhere is my order?"When will my parcel arrive?"
  4. Return or repairAfter something goes wrong"It doesn't fit. How do I send it back?"
  5. Come backThe relationship after the sale"Is there a spare part for what I bought?"
The retail journey a concierge can accompany

How the concierge works with your shop, ERP and order systems

Typewise doesn't replace your shop, ERP or order management. It connects to them and keeps three things apart.

Information. Lookups read from connected systems: orders, stock, products, tracking. Typewise has a Magento integration for orders, products and customer details, a parcel tracking integration, and connects other shop, ERP or order systems through an MCP server that exposes their data. See the integrations for what is available.

Actions. Actions change something in a connected system, such as cancelling an order, if that system offers the operation. You configure each action separately, and approval by a person is on by default. An input can be limited to verified values, such as the ID of a signed-in customer, so the concierge only changes that customer's own order.

Handoff. When a case needs judgment, the concierge hands it to your team with the conversation history and a summary, in Typewise, Zendesk, Salesforce or by email.

CustomerChat, email, WhatsApp or web form
ConciergeThe Typewise AI Agent, configured by your team
  • Reads orders, products, stock and tracking through lookups
  • Acts where a system offers the operation and you allow it, with approval on by default
  • Hands over to a person with the history and a summary
  • Shop or commerce platformfor example MagentoReadsActs
  • Product catalogoptions, availability, linksReads
  • ERP or order managementorders, stock, invoicesReadsActs
  • Carrier trackingdelivery statusReads
  • Returns portal or labelscan belong to a partnerReadsActs
  • Your teamin Typewise, Zendesk, Salesforce or emailHands over
What the concierge reads, what it may change and when it hands over. Which systems you connect, and which operations they offer, depends on your setup.

Your admins set this up: which Specialist handles which topic, which lookups and actions it may use, and when it hands over. Nova, the AI Operator in Typewise, can draft that configuration with them. Its changes stay drafts until someone publishes them, and Nova never writes to customers. The concierge your customers talk to is the AI Agent.

What customers need at each stage, and how the concierge handles it

1. Can an AI concierge give product advice in an online shop?

Yes, as long as it answers from product data that is complete for the shopper's market. A shopper comparing two models wants to know which one fits them, in their own language, without waiting for an email reply.

From implementations. The answer is only as good as the data. If your specifications list combinations that can't actually be built, or links point to items no longer sold in that country's shop, the concierge can repeat the mistake. Check the product data for each market before you ask the AI to advise. Our article on data readiness goes through the checks.

In Typewise. The chat on your site can pass the product or cart the shopper is looking at into the conversation, and the concierge answers from the knowledge and catalog lookups you connect. Advice that needs a specialist, such as a custom configuration or a large order, goes to a person who sees the conversation so far.

Customer example. Lehner connected its Magento webshop and inventory, so its AI agents give specific, advisory answers instead of generic ones. Read the Lehner story.

2. Can the AI change or cancel an order?

Where your shop or order system offers the operation and you configure it as an action, yes. Looking an order up and changing it are separate decisions: a lookup only reads, an action changes the order, and you decide per action whether a person approves it first.

From implementations. A practical first step is a prepared change. The AI collects the details and prepares the change, a colleague confirms it in the order system and gives the reason, and the AI writes the answer, including that reason. As the team comes to trust the prepared changes, let the AI make the ones your policy allows on its own.

In Typewise. Each action has its own description, inputs and approval setting, and you can check it in test conversations before it goes live. Our article on AI agent permissions covers how to decide which actions to allow.

Customer example. At Lehner, the AI agent qualifies a cancellation request, gathers the information and prepares the cancellation. A person reviews and approves it, and the AI agent then updates the customer. Lehner went live in five weeks, and its AI agents now resolve 95% of chat requests. Read the Lehner story.

3. Can the AI answer "Where is my order?"

Yes, when it can read the order status and the tracking number. Delivery questions need a quick answer, and when people answer them by hand they occupy the same staff and phone lines as shoppers who are still deciding.

From implementations. Make sure the concierge can find the tracking number. If it sits in a free-text note or in a carrier portal nobody connected, the AI can only explain how delivery works.

In Typewise. The concierge looks the order up in your connected shop or order system and checks the shipment with the parcel tracking integration, which takes a tracking number and returns the delivery status. A parcel that is stuck or lost becomes a case for your team, with everything the concierge already looked up in the conversation. Carriers and logistics providers have their own page: AI customer service for logistics.

Also possible. Telling customers about a delay before they ask. Your team can start an email conversation from Typewise, and the reply comes back to the same ticket. Sending these messages automatically when your order or carrier system flags a delay is an integration to scope with us, and on WhatsApp a message outside the 24-hour window needs an approved template.

4. Can an AI agent handle returns?

It can check whether a return applies and guide the customer through it when it can read order and return data. Without that data, it can only explain your returns policy.

From implementations. The catch is ownership. Return labels and portals can belong to a brand, a marketplace or a logistics partner, and getting access can mean an interface project, a contract change and a privacy review. Start that conversation early. Until then, point customers to the self-service portal where one exists, and send the cases a portal can't handle, such as mixed orders or items shipped by a store, straight to a person.

In Typewise. The concierge checks eligibility through a lookup, can attach a returns form you upload as a public file, and sends a return label only where an integration produces one. Goodwill decisions and claims that need evidence stay with a person, who sees exactly what the customer was told.

Customer example. At Beurer, the AI agents check whether a return or refund applies and make sure the address and payment details are complete before a person decides. With Salesforce Service Cloud connected for reading and writing, this approach cut handling time for the tickets that still need a person from 12 to 6 minutes. Read the Beurer story.

5. How should the AI handle warranty and spare-part questions?

Carefully, because some questions that look simple aren't. A request for a replacement cable can be a warranty case, and some product questions are really safety questions.

From implementations. Instead of handing every such topic to a person, let the concierge first ask the questions a colleague would ask. Then it passes on the case with a summary that says more than a one-line note.

In Typewise. Set this up as a Specialist for the topic, with instructions for the questions to ask and handoff conditions for when a person takes over. If the customer is identified as the same customer record, for example signed in on chat or by phone number on WhatsApp, the concierge can find their earlier conversations. An anonymous chat visitor has no history to find. Questions that depend on a personal relationship, such as a regular client of a store expert, stay with the people who know them.

Also possible. Advice that takes earlier purchases into account, such as the spare part or accessory that fits what the customer bought. It needs the customer to be identified and their order history connected as a lookup.

Which systems to connect, and what to check first

Each system answers a different part of the journey, and each can have a different owner.

SystemWhat the concierge readsActions you might allowCheck first
Shop or commerce platformOrders, products and customer details (for example through the Magento integration)Order changes, where the platform offers them and you connect themWhich operations does it expose, and who maintains the connection?
Product catalogOptions, availability and links for each marketNone needed for adviceCan every option the data lists actually be bought in that market?
ERP or order managementOrder status, stock and invoicesThe changes your policy allows, with approval where you want itWhich changes may the AI make, and which need a colleague's confirmation?
Carrier trackingDelivery status from the tracking numberNoneIs the tracking number in a field the AI can read?
Returns portal or label serviceEligibility and return statusCreating a label, where an integration produces oneWho owns the portal, and what contract or privacy step does access need?

When a person takes over, the conversation continues

A concierge is only as good as its handover. When a request needs a person, the customer shouldn't have to start again, and your colleague shouldn't have to reconstruct the case.

The handover carries the conversation history and a summary. It can stay in Typewise or go to the helpdesk your team already uses, such as Zendesk or Salesforce, or arrive by email. If the reply has to wait for a colleague, the customer keeps one conversation rather than starting a new one. Our article on handoff ownership covers who owns the case after it.

Keep one knowledge base and one tone of voice for every channel. Teams that run a separate knowledge base for the chatbot, the agents and the voice bot end up maintaining three. And articles written for experienced agents can be hard for an AI to read correctly. Your team decides where the line sits: which topics people handle, which actions need approval, and which answers the concierge gives on its own. That line can move as trust grows.

How to start

Start at the stage where your data already gives a correct answer, and set up your brand voice and knowledge once so every later stage inherits them. Before go-live, run test conversations in Typewise that check which Specialist answers, which lookups run and whether the case is handed over.

Go live supervised: the concierge drafts, a colleague approves, and new conversations go to it while existing threads stay in your current helpdesk. Add the next stage when its data and decisions are ready: returns once you have access to labels or the portal, product advice once the product data covers what customers ask in each market. Avoid going live in your busiest weeks, when nobody has time to read the first conversations. More on the first weeks in how support teams go live with AI.

Common questions

Which retail and e-commerce companies use Typewise?

Typewise has published customer stories with Lehner, Brack.ch and Beurer. Each story describes its own setup, and the results in it are the ones the customer reported.

Does Typewise integrate with Magento, Shopify or our ERP?

Typewise has a Magento integration for looking up orders, products and customer details, and a guide for embedding the chat widget in a Shopify store with signed customer details. Other shop, ERP and order systems connect through an MCP server that exposes their lookups and actions, or through the other integrations Typewise supports. What the concierge can do then depends on the operations your system offers and the ones you allow.

Is this a shopping assistant or customer service?

Both, seen from the customer's side. If marketing owns the shopping assistant and service owns the inbox, that can work. What matters is that both draw on the same product knowledge, speak in the same voice and, where the customer is identified, know their history, so the brand feels like one company.

Can it change a delivery address after the order has shipped?

Not on its own. Once a parcel is dispatched, the carrier controls the address, so that change stays with your team or the carrier. Before dispatch, the concierge can change the address where your order system offers the action and you allow it, with your team's approval if you want it.

How does this fit a small service team?

Plan the handover around the coverage you have. Set the concierge up to tell customers outside staffed hours when someone will reply, and don't make answers wait on an approval from someone who isn't there.

Who configures the concierge, and what does Nova do?

Your admins configure it: the knowledge it answers from, the Specialists for each topic, the lookups and actions it may use, and which actions need approval. Nova, the AI Operator in Typewise, can draft that configuration. Its changes stay drafts until someone publishes them, and Nova never replies to customers.

Customer stories

Published customer stories.

Typewise has published customer stories with Lehner, Brack.ch and Beurer, from online retail and consumer brands. Each story describes its own setup and results.

See all customer stories →

See the concierge on your own products and orders.

Book a demo and we'll walk through your customer journey, the systems behind each stage and where your team wants to stay involved.