PlentyONE & Claude: AI reply drafts in the Messenger

“Where is my order?”, “When will it ship?”, “My voucher doesn’t work”: in an online shop’s customer service, many questions repeat themselves. If you work with PlentyONE (formerly plentymarkets), they land in the Messenger – and are usually answered by hand, after checking the order, the shipping rules and old replies. For a food online shop, we therefore connected the PlentyONE Messenger to Claude, the AI by Anthropic. For every new customer message, a finished reply draft is now waiting – right in the conversation.

The principle: the AI drafts, a human decides

A small program on the server regularly fetches new and unanswered conversations through the PlentyONE REST API. It reads the thread, recognises messages from the contact form and collects the facts about the order. From this, Claude writes a reply draft and adds it to the conversation as an internal note. The customer doesn’t see any of it.

The team reads the note, copies the text into the reply, adjusts it if needed and sends it from PlentyONE as usual. Important: the AI never sends messages itself, creates no credit notes and changes no orders. It reads and writes notes – its PlentyONE user has no further rights.

What Claude delivers with every draft

• The reply text in the team’s tone – short, friendly and with the greeting the shop always uses
• A category such as shipping, voucher, complaint or product question
• An assessment of how reliable the reply is, with a reason
• Internal hints such as “payment not received yet” or “customer has asked twice already”
• Suggested next steps for the team and a flag when a human has to decide

What the AI knows – and where from

A good reply needs the right facts. That’s why Claude gets more than the customer’s message:

• Order data from PlentyONE: status, payment, tracking number and the expected shipping day – calculated from the shop’s shipping days and cut-off time
• The team’s knowledge: frequent questions, reply templates and an internal guide for goodwill decisions
• Tools: a product search in the shop, a check of voucher codes the customer mentions, and a search for points of sale nearby

If information is missing, Claude doesn’t make anything up – it flags the message for the team. That’s the difference from a simple chatbot: the AI works with real data and knows when it doesn’t know.

More than replies: analytics by question

The project quickly grew into a small dashboard. Besides the reply drafts, it offers:

• A spike alert: if several customers suddenly write about the same topic, such as a delivery delay, the team gets an email – often before anyone notices the problem.
• Analytics in plain language: “How many new customers did we have in October?” or “Which vouchers were redeemed most often?” Claude uses fixed, tested queries on a copy of the order data and only sees totals, never customer data.
• A weekly report with the key figures for management.
• A check of customer comments: if a note from the order, such as a requested delivery date, is missing in the PlentyONE order, the system raises an alert.

Privacy and security

Only what the reply needs is sent to Claude. For analytics, customers are pseudonymised so that the AI only works with figures. The connection uses Anthropic’s API; under its commercial terms, Anthropic does not use API content to train its models, and a data processing agreement is available. In PlentyONE, the AI uses its own API user with only the rights it needs. And because a human checks and sends every reply, responsibility stays where it belongs.

What it brings

• Less typing for standard questions and faster replies
• A consistent tone – even when someone on the team is on holiday
• All key order facts at a glance, without switching windows
• Problems spotted earlier, because spikes stand out
• Answers to business questions without digging through spreadsheets

Who it’s for

A connection like this pays off for shops that use PlentyONE and answer many similar customer requests every day. It can be adapted to your processes, shipping rules and tone and extended step by step – from reply drafts to analytics. We build such solutions in PHP and know the interfaces of PlentyONE, WooCommerce and Claude. If you’d like to see what this could look like for your shop, get in touch.