Skip to content
Paper-cut illustration of hillsides crossed by a path, in low sunlight

AI agents working together in your company

Your teams become more productive and creative, without changing their habits.

Where Marsanne comes in

Marsanne works on your core business.

Generic requests Upstream, off-the-shelf software handles them.

  • Calendar
  • Emails
  • Expense reports
  • Meeting minutes
  • Translation

Marsanne Studio

Your core business

  • Demand forecasting
  • Quotes
  • Data source unification
  • Orders
  • Follow-ups

The next step

  1. New offers
  2. New services
  3. New territories

Principles

What guides each of the studio's projects.

  1. Intent

    Knowing exactly what we are building, and for whom.

  2. Simplicity

    Looking first at what can be removed.

  3. Care

    The kind that shows in the details no one else thought of.

  4. Passion for users

    Designing for the people who will use it every day.

  5. Obsession with the small things

    They almost always matter more than they seem to.

Our convictions

The challenge for an SME: orchestrating agents across its own workflows.

  1. 01

    The gain comes from orchestration.

    Models are becoming increasingly comparable. The result depends on how the agents are chained together across your workflows.

  2. 02

    Your teams keep their habits.

    The agents deliver their work to the email, spreadsheets and software you already have in place.

  3. 03

    You choose your models.

    This includes 100% European, sovereign models. You are not dependent on any single provider.

What the agents do

Agents that deliver finished work.

Illustrated portrait of Hélène, the studio's AI finance agent.

Hélène finance agent

Sales administration

Enter orders received by email

Today
Each purchase order is re-keyed by hand, with errors in references and quantities.
What the agents do
An agent reads each order as it arrives and prepares it in your software.
What is checked
A second agent checks prices and quantities against the customer's price list.

What is measured

  • Orders handled per person
  • Data entry errors
Drawn portrait of Armand, the studio's AI industry agent.

Armand industry agent

Sales

Price non-catalogue requests

Today
A tailored quote takes several days. Some requests are declined for lack of time.
What the agents do
An agent starts from the closest past quotes, applies your prices and produces a quote for approval.
What is checked
A second agent checks the margin and the lead time quoted.

What is measured

  • Quote turnaround time
  • Conversion rate
Illustrated portrait of Inès, the studio's AI logistics agent.

Inès logistics agent

Supply chain

Forecast demand for each item

Today
The forecast lives in a spreadsheet updated once a month. Stockouts and overstocks are discovered too late.
What the agents do
Each week, an agent cross-references sales, open orders, promotions and seasonality, and proposes a forecast for each item.
What is checked
A second agent compares each forecast with actual sales. The planner approves before any supplier order is placed.

What is measured

  • Stockout rate
  • Days of stock
Illustrated portrait of Camille, the studio's retail AI agent.

Camille retail agent

Customer service

Reply to customers based on their file

Today
Every request means reopening the contract, order and history. Simple questions have to wait.
What the agents do
An agent prepares the reply in your email and cites the documents in the file.
What is checked
Every statement links to a document. No goodwill gesture is offered without approval.

What is measured

  • First response time
  • Requests resolved at first contact
Illustrated portrait of Camille, the studio's retail AI agent.

Camille retail agent

Customer retention

Follow up with each customer at the right time

Today
A customer who orders less only shows up in the year-end review.
What the agents do
An agent spots customers who depart from their usual pattern and prepares a follow-up for the salesperson.
What is checked
Each suggestion cites the orders it is based on.

What is measured

  • Repurchase rate
  • Order frequency

The portraits are illustrations. None of these agents is a person.

The question that always comes up

What if the AI gets it wrong?

One agent produces. Another checks. An auditable record remains.

A trail remains

Documented, dated, auditable.

  1. Prices checked

    against the supplier price list, by a second agent · 14:05

  2. Margin and lead time checked

    against management's rules, by a third agent · 14:06

  3. Trail retained

    Documented, dated, auditable · 14:06

What the studio has already built

Products designed, put into service and maintained by the studio.

Each time to speed up the client company's operations in its core business.

View all work →
  • Carnets

    Travel journals for photographers, with professional photography at the heart of the story.

    • Meta data export
    • Upload and processing pipeline
    • Performance with large volumes of media
    • +5 more
    View live project →
  • GCSE Beast

    A revision app for English students preparing for GCSE Physics, with 770 exam-style questions, an AI coach that explains and draws, and progress you track like a sports training programme.

    • Multilingual AI teaching coach
    • Gamification of learning
    • Mobile experience
    View live project →
  • Deboîte

    For a company that installs AI for SMEs on a machine set up on their premises: the launch website, with use cases by function and a pricing simulator.

    • Making a technical offering accessible
    • Sizing and pricing simulator
    View live project →
  • Carnets Maps

    The same foundation, adapted for agencies and travel planners, with itineraries and maps shared with the client.

    • Geomapping of unstructured data
    • Route calculation
    • Aggregation of heterogeneous data
    • +1 more
    View live project →
  • Screenshot of the Sauna Garden 3D configurator, showing a garden sauna in its setting.

    Sauna Garden 3D

    A 3D configurator for a timber construction specialist, allowing customers to visualise a garden sauna before it is offered for sale as a kit.

    • 3D modelling
    • Client-side model configuration
    • Bill of materials
    View live project →
  • Screenshot of the Nagomi Travels website in comment mode: numbered markers placed on the page, with the client's feedback panel open on the right.

    Nagomi Travels

    Design of a website and back office for a travel planner, to generate highly personalised itineraries and share them easily with clients.

    • Premium/luxury user experience
    • Customer feedback loop
    View live project →

Use cases

The studio's latest articles.

See all insights →

AI watch

What the press reports about AI in business.

Every day, the studio's agents read the press on AI in business. Every week, they distil the key points for an SME, with sources.

Frequently asked questions

What executives ask before getting started.

Does my data have to go through your servers?

No. The first go-live works on an export chosen with you, with a few users, and your data stays with you. Connecting to your ERP, email or Drive is decided afterwards, with your agreement.

Will our teams have to change the way they work?

No. The agents deliver their work to the tools your teams already use: email, spreadsheets and the management software. The solution is designed with the people who will use it, and fine-tuned in the field.

Are we dependent on an AI model, or on Marsanne?

No. Each agent runs on the model of your choice, including a 100% European, sovereign model, and can switch models without the solution being rebuilt. A full transfer of the technology is possible at the end of the engagement if you wish to remain autonomous.

How long before a first concrete result?

The executive diagnostic takes a few days. The first go-live generally happens within a few weeks, on a specific process rather than across the whole business.

At what point does our IT department get involved?

The first discussions take place with management and the people who will use the solution. While it runs on an export with a few users, it makes few demands on your IT. Your IT department or your service provider becomes involved for the integration, with measured results to support the decision.

How is this different from a generic AI tool such as ChatGPT or Copilot?

These tools cover cross-functional tasks well: notes, meeting minutes, reports. Marsanne orchestrates agents on the processes that are your business, with your data and your rules.

Portrait of Vincent Piche, founder of Marsanne Studio

The studio

Vincent Piche

Founder of Marsanne Studio

More than twenty years in international companies: watchmaking, jewellery, retail, industry and finance. Polytechnique, HEC. He leads the studio and its team of agents.

The studio's agents → LinkedIn →

Getting started

The executive diagnostic: where your company is losing days today.