First-line customer support
Answers recurring questions by email, WhatsApp or Telegram, qualifies requests, escalates ambiguous cases to a human.
5 to 15 hours a week reclaimed on the support team, available 24/7.
TL;DR
What it is
Turnkey delivery of an autonomous conversational AI agent, scoped on your processes, EU-hosted, trained on your data.
For whom
French SMEs looking to automate L1 support, monitoring, admin assistance or internal search — with SSO, logging and GDPR by design.
Indicative price
From EUR 200 excl. VAT scoping + build on quote (typically EUR 6,000-25,000 excl. VAT). Ongoing support EUR 150 excl. VAT/month.
The typical starting point for an SMB
Too many consumer AI tools
ChatGPT on mobile, Claude in the browser, Copilot in Word — with no governance, no traceability, and a real risk of customer data leaks.
No business specialisation
General-purpose AIs don't know your catalog, your procedures, your recurring customers. Every prompt starts from scratch.
Nothing shared across the team
One employee's best prompt doesn't help the others. You waste the same time three times on the same problem.
What the MAG&Cie SMB AI Agent delivers
One agent, your organisation
An autonomous conversational assistant, aligned with your processes, your vocabulary, your reference documents, your catalog.
Sovereign hosting
AI model on European infrastructure (Anthropic Bedrock Paris, Mistral, OVHcloud), data stored in France, no third-party resale.
Security by design
Full logging, SSO, granular access control, scope separation. Aligned with France's ANSSI framework and GDPR article 32.
Integrates with your stack
Connects to your business tools via MCP (invoicing, CRM, calendar, drive), without rebuilding your IT. Every connector audited before activation.
Onboarding included
Your teams get a 1-hour session + a written usage guide. A Slack or email channel to ask questions for 30 days.
Scalable
The scope extends module by module (customer support, regulatory watch, admin assistant, meeting prep). You set the pace.
Six agents we deliver
Each agent is specialised on one job, with a measured time-saving criterion at 60 days. You pick where to start.
Prospect pre-qualification
Receives inbound leads, asks 3-5 structuring questions, scores the opportunity and hands over a call-ready sheet.
Sales focused on ripe prospects, appointment conversion rate doubled.
Appointment booking and confirmation
Takes meeting requests, offers slots, sends confirmations and reminders, reschedules cancellations.
3 to 6 hours a week reclaimed at the front desk, fewer no-shows.
Mail triage and prioritisation
Sorts the shared inbox by topic and urgency, drafts replies for standard cases, flags true urgencies.
Two hours a day less in the inbox, per person.
Meeting summaries and actions
Reads the meeting transcript, produces a structured summary, extracts decisions and actions with owner and due date.
45 minutes of note-taking avoided per meeting, teams aligned instantly.
Automated watch
Reads your defined sources each morning (competitors, press, regulatory), summarises the news, emails a one-page note.
2 to 3 hours a week reclaimed for leaders, no key news missed.
Reclaimed time flows back to what the machine cannot do: customer relationships, judgement calls, product decisions — the real high-value work.
Continuous improvement after go-live
An AI agent is not frozen. Every week it meets new edge cases, its answers can drift, your processes evolve. Recurring support keeps the agent up to date.
Monthly conversation review
We read a sample of real conversations, identify mis-handled cases, fix the instructions and the reference documents.
Knowledge base enrichment
New documents, new procedures, new catalog items added. The agent keeps up with your organisation.
Progressive scope extension
Once the first use case is stable, we add new connectors, new channels, new tasks. Always module by module.
Indicator tracking
Number of requests handled, response time, hand-off rate, satisfaction. A monthly report that documents ROI.
Recurring support
from €150 excl. VAT / month
€180 incl. VAT / month — monthly, cancel anytime
Indicative scope: 1 h review + 1 h adjustments per month, or a larger monthly package on quote depending on volume and modules. A personalised quote is always issued before we start.
4-phase method
- 1
Scoping (weeks 1-2)
Half-day workshop: map your processes, pick 1 to 3 priority use cases, set measurable success criteria and identify the available data.
- 2
Build (weeks 3-6)
Agent development: system prompts, MCP connectors, business tests with you. 30-minute weekly checkpoint on a staging environment.
- 3
Secure deployment (weeks 7-8)
Production release: EU hosting, SSO, logging, recovery procedures, operational documentation. Formal sign-off.
- 4
Training & follow-up (weeks 9-10)
1-hour team training, written usage guide, question channel open for 30 days. Review at 60 and 90 days to schedule next modules.
MAG&Cie guarantees
AI model hosted in the European Union (Anthropic Paris, Mistral, OVHcloud, Scaleway) — verifiable contract clauses.
Zero training on your data. Your data never leaves your technical perimeter.
Aligned with the French ANSSI (IT Hygiene Guide) and GDPR article 32 on technical measures.
Reversibility: at any time, you export the agent's configuration, the knowledge base, the full history.
FAQ
How much does it cost?
From €200 excl. VAT (one-shot scoping fee, indicative). Build and run are then quoted after scoping — pricing depends on the number of use cases, projected volume and IT integrations. A full quote, with a precise scope, build cost and monthly operating cost, is issued at the end of the scoping. No charge before you validate that quote.
How long from order to production?
Six to ten calendar weeks depending on complexity and your team's availability for validation points. Simple scopes with a single use case ship in six weeks.
What if our team changes their mind mid-project?
Any change to the scope is priced through an amendment. We favour a first module in production quickly, then successive extensions, over a single large project.
Does the agent learn from our data?
No. The AI model is never retrained on your data. It receives your documents and context as input on each query, via a mechanism called retrieval-augmented generation (RAG). Your data stays stored in France, under your control.
What GDPR obligations apply?
The agent processes personal data: we set up a processing register, a simplified impact assessment where needed, and document the responsibility split with the AI provider. Your DPO (internal or external) gets a full compliance dossier.
Can we start small, then extend?
Yes — that's actually the recommended method: one module in production delivers immediate value, plus a technical and operational base to add the next ones cheaply. Most SMBs we work with start with a single use case.
/ En résumé
Ready to scope your AI agent?
A half-day workshop is enough to validate whether your organisation is ready. We reply within 48 business hours.
Continue with
Pages that pair well with this one
Adjacent topics covered by MAG&Cie, hand-curated to stay on your problem.