It answers straight away, and knows when to hand over

The company sells online to other businesses and turns over more than twenty million. People often assume a B2B e-commerce needs less support than a consumer one, because there are fewer orders, customers know each other and relationships last for years. The opposite happens.

Someone buying for their company is in a hurry and orders large quantities. If an order is late, there is someone on the other side waiting for it. And if they get no answer they do not leave a bad review: they change supplier, often without saying so.

Requests arrived at any hour, nearly always about availability, delivery times and order status. Most could be solved in a minute with information already in the company’s systems. A small share, though, was urgent, and had to be recognised at once.

What to avoid was clear to everyone. Nobody wanted the kind of support where a customer spends ten minutes going round menus and automatic replies looking for a way to talk to someone, and never finds one.

• Immediate answers on catalogue, orders and availability
• Urgent requests go straight to a person, no runaround
• Customers' questions turn into direction on what to sell

At the front, answer straight away

Customers write the way they would write to a salesperson, and the assistant answers in natural language on catalogue, availability and orders. The answers come from the company's own data.

Knowing when to hand over

The most delicate part was teaching it to stop. From the content of a request and the tone it arrives in, the assistant can tell whether something is urgent or whether a customer is losing patience. At that point it stops insisting and hands the conversation to a person, who finds everything the customer asked already written down.

At the back, questions become decisions

Under the assistant works a second layer, the same pattern we use in very different projects. Conversations are collected and read as a whole, and concrete direction comes out for the people who sell: which products are searched for and not found, and where it is worth pushing.

How it is built in Microsoft 365

Anyone building the same double layer with Microsoft tools would use these pieces. An agent in Copilot Studio published on the website, connected to catalogue and orders through Power Platform connectors. Handover to a human handled through escalation to customer service. And conversations collected in Dataverse and read in Power BI.

Customers get an answer while they are looking for it, at any hour, and when they need a person they reach one straight away. For a B2B supplier that is the difference between a customer who reorders and one who tries a competitor.

At the back, customers' questions have stopped ending up in a log. They have become the list of products to push, and a concrete basis for growing revenue.