FR Contact Menu
Contact

Custom AI systems. From diagnostic to production.

Extra identifies the processes where AI has a measurable impact, from inventory management to customer service. A dedicated engineer builds the solutions inside the company’s existing systems, then supports the teams over time: operations, monitoring, improvements.

ContactThe diagnostic
Example: six company systems (ERP, CRM, tickets, documents, email, data) feed the solutions built after the diagnostic: a supplier matching agent and an order entry automation in production, a document search assistant in pilot. Their results go through human approval, a posting to the ERP, a sourced answer and the audit log.

/01 · Diagnostic

What the diagnostic can recommend.

The solution follows from the diagnostic. Never from a choice made in advance.

The diagnostic

  1. AI agents

    Cases handled end to end, inside existing systems.

    /0.1

  2. Automations

    Flows between systems. No more keying, no more rekeying.

    /0.2

  3. Assistants

    The company’s knowledge, sourced, one question away.

    /0.3

  4. Custom software

    In place of generic software, a tool built around the company. And maintained for it.

    • ERP
    • CRM
    • Inventory management
    • Integrated finance
    • Customer portal
    /0.4

  5. AI products

    AI at the core of applications and products.

    /0.5

  6. Process redesign

    Sometimes the best answer is a simpler process.

    /0.6

/02 · Method

Diagnostic. Build. Run.

One team, from the first interview to operations.

  1. /01

    Step 1: Diagnostic

    The ground, the data, the systems. Every case measured.

  2. /02

    Step 2: Build

    Built inside the company’s systems. Put into production step by step.

  3. /03

    Step 3: Run

    Operated, monitored, improved. Over time.

The method

/03 · Use cases

Every function. A system built for it.

  • /01

    Customer service

    Request triage, drafted answers with their sources, first line on chat and phone.

  • /02

    Finance

    Matching, collections, expenses and close, every figure tied to its document.

  • /03

    Legal and compliance

    Contract review against the legal playbook, deadlines, compliance questionnaires.

  • /04

    Operations

    Orders keyed into the ERP, supplier delays, stock, quality and schedules.

  • /05

    Sales

    Qualified leads, prepared meetings, an up-to-date CRM, answers to tenders.

  • /06

    Internal knowledge

    An assistant that answers the teams from internal documents, sources cited, permissions respected.

/04 · Example

Supplier invoice matching.

Example: an agent processes the batch of October 14, 312 supplier invoices. For invoice FAC-24-0869 from Métaux Arvel, it reads the invoice received by email, matches it against order 4500018273 in SAP and receipt 5000221 from the warehouse, and finds 132 units billed for 120 ordered and received, a €170.40 gap. The gap is above the €50 threshold: payment is held and accounts payable must decide. Every step is written to the log and the entry is prepared in SAP without being posted. Across the batch, 298 invoices are matched, 11 gaps are explained and 3 invoices wait for approval.

/05 · Security

On existing infrastructure, under internal control.

  • Client infrastructure

    The company’s cloud or servers, behind its own access controls.

  • Code and data ownership

    Code handed over in the company’s repository. No lock-in.

  • Model choice

    Claude, GPT, Mistral or open source, depending on the task and the constraints.

  • A person decides

    Every outward action (sending, publishing, paying) follows permissions set by the company.

Security

/06 · Team

A team that builds and runs AI in production.

Engineers and entrepreneurs. École polytechnique, HEC Paris, UC Berkeley.

  • Education

    École polytechnique, HEC Paris, UC Berkeley, Sciences Po, UCL

  • Founders

    Anton Jego, Rayan Souissi

  • Method

    An engineer embedded in the business teams

  • Control

    Permissions and a log of every action

  • Models

    Claude, GPT, Mistral, open source

  • Ownership

    Code and data owned by the client

The team

Contact