AI agents for customer service in SMEs

Today

Emails, website forms and voicemails land in a shared inbox. Someone reads them one by one, looks up the customer and the order, then decides who should answer and what is urgent.

With AI

For each email or form received, the agent identifies the sender in your CRM or ERP, finds the order or contract concerned, files the request under your categories (quote, order, complaint, admin) and detects language and urgency according to your rules. It creates the ticket in Zendesk, Freshdesk or HubSpot with these details, assigns it to the right queue and notes what it has done. Ambiguous cases go to a queue for checking, and the team corrects assignments, which refines the rules.

Today

The same questions come back every week: lead times, return conditions, opening hours, documents required, order status. Each staff member looks up the information and rewrites the answer in their own way.

With AI

We switch on and set up the AI agent built into Zendesk, Freshdesk or HubSpot. For each open ticket, it searches your terms and conditions, approved answers and the customer’s history, checks the order status through the relevant connector, then leaves a draft reply in the customer’s language. The staff member reviews it, adds the specifics and sends it; any goodwill gesture remains their decision.

Today

Customers call for information already published on the website, or to find out where their order is. The phone line is taken up by simple questions while complex requests wait.

With AI

The agent introduces itself as AI, answers from your approved pages and documents while citing its source, and acts within a set scope: giving an order’s status after checking the order number and email, opening a complete ticket or booking a call-back in the team’s calendar. As soon as a question concerns a complaint, a contract or a goodwill gesture, it hands the conversation to a member of staff. Every exchange is logged, and the team reviews unresolved conversations.

Today

Google or TripAdvisor reviews and satisfaction survey answers pile up. Nobody has time to read them all or pick out the recurring topics.

With AI

Once a month, the manager exports the reviews and asks the assistant to group them by theme (welcome, waiting time, quality, value for money) and to suggest a courteous reply to each. The assistant does not act in any tool: the manager picks the replies, adjusts them, publishes them personally and decides what action to take.

Today

A complaint about a delivered part means gathering the order, batch number, previous exchanges and sometimes photos scattered across the ERP, Outlook and the network drive. Building the file takes time before any analysis starts.

With AI

We build an agent connected to your ERP, mailbox and network drive: when a complaint arrives, it finds the order and batch, gathers previous exchanges and photos, summarises the history and suggests an initial classification (defect, transport, use) based on your criteria. It opens the file in your tracking tool and prepares a draft acknowledgement. The quality or after-sales manager decides what happens next, including any replacement or goodwill gesture.

Today

A property management office receives fault reports, questions about service charge statements and termination requests every day, often incomplete. The assistant calls tenants back to get the exact address, the property and the urgency.

With AI

When a report arrives by email or form, the agent links it to the building and lease in your property management software, writes to the tenant, introducing itself as an AI agent, to obtain missing details, rates the urgency by your rules (leak, heating, lift) and prepares the work order for the contracted tradesperson. The property manager or caretaker approves before anything is sent to a tradesperson, and each step is recorded in the property file.

Today

At the garage, requests for servicing, tyre changes or repairs arrive by phone, email and web form. Each time the service adviser has to identify the vehicle, estimate the duration and find a free slot.

With AI

We build an agent connected to your garage software and workshop calendar: for each request, it finds the vehicle by plate or customer, estimates the duration from your standard times, checks bay and common-parts availability, then provisionally holds two or three slots and offers them to the customer, introducing itself as an agent. The service adviser confirms the appointment, and any repair estimate is still prepared by the workshop.

Today

An online shop receives return requests with reasons, photos and order numbers in varying states of completeness. Checking the return period, the item’s condition and preparing the label is done by hand.

With AI

We build an agent connected to the shop and the carrier: it finds the order, checks the deadline against your terms, asks the customer for a photo if needed, classifies the reason and generates the return label or prepares the exchange. It also prepares the refund in the shop, but the refund is only triggered once a staff member approves it, after the item has been received and checked.