Service · Virtual assistants & AI chatbots

Virtual assistants that talk to your systems

AI chatbots, internal copilots, automated customer support agents. Connected to your CRM, your catalog, Zendesk, your PrestaShop. They qualify, answer, execute, escalate intelligently. Our Herboristerie du Valmont case handles 70% of support requests without humans, with 27+ business intents.

27+ Herbo Valmont intents 70% without humans GDPR · EU

Three types of assistants we deploy

A virtual assistant is not just a chatbot in a corner of your site. It's an AI agent that understands natural language, accesses your data and systems, and performs concrete actions for your users or your teams.

Customer support assistants

On your site, on the front line. Order tracking, returns, product info, contact requests. Escalation to Zendesk (or other) when a human is required.

Sales and qualification assistants

On your landing page, like Synapse on ours. They gather prospect info, understand the project, generate a useful recap, feed the sales team.

Internal copilots for your teams

For your staff: internal doc search, writing assistance, ticket analysis, meeting note generation. Connected to your internal sources.

E-commerce assistants

On your store: recommendations, product comparisons, decision help, configurators. Connected to PrestaShop and your catalog data.

Concrete proof: Herboristerie du Valmont support

Our most accomplished virtual assistant use case is the Herboristerie du Valmont support chatbot, an online herbal shop with a rigorous selection of medicinal plants and teas. Here's what we built:

27+
Business intents
70%
Requests without humans
24/7
Availability

Business understanding

27+ intents specific to herbal commerce: order tracking, product return, plant advice, contraindications, batch traceability, organic certification, delivery.

Real data access

Live order status from PrestaShop, full product sheet, customer history, logistics info. The assistant answers with facts, not blabla.

Zendesk escalation with context

When a human is required, the assistant opens a Zendesk ticket with conversation summary, identified intent, info already gathered. No repetition.

Guardrails on sensitive topics

On anything related to health advice or medicinal use, systematic escalation to a qualified human. The assistant knows its limits.

Result: the Herbo team only handles complex or sensitive requests. The rest is processed instantly, around the clock.

How Synapse works (our own assistant)

Our own AI advisor, Synapse, illustrates what we can deploy for sales contact. His logic:

1. Open hook on the project

No 'give me your first name' upfront. Synapse starts with an open question: 'what brings you here?'. The prospect talks about their need, the assistant listens, rephrases, shows it understands. The conversation settles in naturally.

Technical detail: everything the prospect says is stored in the database from the very first line. No need to have the identity to start capturing the value of the conversation.

2. Identification at the right moment

Once the context is set, Synapse gathers the six identification fields conversationally and with a commercial justification:

  • First and last name (to personalize the recap)
  • Professional email (to send the recap)
  • Job title (to understand your role in the decision)
  • Company name and size (to adapt our approach to your scale)

The idea: every piece of info asked has a reason the prospect can see, not a disguised form.

3. Project deep-dive and dual email

In-depth Q&A on the project, then on the way out:

  • To the prospect: a clear email recap, a useful working document to share internally.
  • To the Synapse Up team: a pre-chewed analysis (likely use case, estimated complexity, technical points to watch) that lets us call back with a concrete proposal.

This pattern is exactly the one we deploy for clients who want to qualify their leads or structure their first contact.

Our method

Measure, fix, stabilize. Applied to an AI assistant, it looks like this.

01

Intent scoping

Which contact reasons? Which data needed? Which tone? We start by mapping the real need.

02

First deployment

Initial version with the main intents. Quick deployment, measurement of resolution rates.

03

Continuous enrichment

We add the intents that emerge from real usage. The assistant gets more capable every week.

Security, GDPR, transparency

An assistant talking to your customers handles sensitive data. Our commitments:

European hosting No third-party training on your conversations Explicit 'you are talking to an AI' notice Right of access and deletion Guardrails on sensitive topics Mandatory human escalation if in doubt Full logs and audit trail Sources cited in answers

Frequently asked questions

What's the difference between a classic chatbot and an AI assistant?

A classic chatbot follows pre-programmed decision trees: if the user clicks this button, then this answer. An AI assistant understands natural language, can adapt to unexpected phrasings, and reasons over context (your catalog, your orders, your documentation). It can also perform actions (create a ticket, modify an order) instead of just answering.

How many intents are needed for a useful assistant?

It depends on scope. For a consumer support service, about twenty well-built intents already cover 70% of requests. On the Herboristerie du Valmont project, our chatbot runs with over 27 business intents. The rule: start with the most frequent contact reasons (order tracking, return, product info), then add as you go.

What happens when the assistant doesn't know the answer?

Escalation to a human with full context. In our deployments, the assistant sends to Zendesk (or your ticketing tool) a conversation summary, identified intent, info already collected, and the escalation reason. Your team doesn't start from scratch.

Can we connect the assistant to our CRM or PrestaShop?

Yes. It's the heart of the value. An assistant that doesn't talk to your systems isn't worth much. We connect to PrestaShop for order status, to your CRM for the customer profile, to your catalog for product sheets, to Zendesk for escalation. The integration patterns are proven.

Are your assistants GDPR-compliant?

Yes. European hosting, no use of conversations to train third-party models, right of access and deletion, processing registry. Users are informed they are chatting with an AI, as the regulation requires.

What about hallucinations? How do you prevent invented answers?

Three guardrails: 1) RAG anchors answers in your documents and data, the assistant cites its sources. 2) Business rules filter answers on sensitive topics (payment, delivery). 3) Human escalation is mandatory when the confidence score is too low. The assistant is allowed to say 'I don't know, I'm transferring you'.

How does Synapse, your assistant on the Synapse Up site, work?

Synapse is our AI pre-qualification advisor. He starts by collecting your first name, last name, work email, job title, company and team size. Then he discusses your project with you. At the end, he sends you an email recap and forwards a pre-chewed analysis to our team. It's exactly the pattern we can deploy for you.

Which assistant are you missing today?

Describe your case to Synapse. Ten minutes to understand your need (support, qualification, internal copilot, e-commerce), and an email recap. Our team gets back to you with a concrete proposal.

Related services