By default we work inside your environment, under your security controls, using AI you have already approved. Where you choose to share extracts, they are minimized, isolated, encrypted under a key you hold, never used to train AI, and destroyed with a certificate when we finish.
Request our trust packYour security team picks the tier and approves the exact fields before any work starts. The safest data is data we never hold.
We work on a device or virtual desktop you provide, signed in through your identity provider, using your approved AI service. We bring only our methods. Your data never leaves, and your controls apply to everything we do.
Customer names and personal data are replaced with codes before anything is shared. Pricing, inventory and operations analysis needs behaviour, not identities. You keep the key that maps codes back to names.
Extracts held in a dedicated Marso environment used by no other client, encrypted with a key you control and can revoke at any time, which makes the data unreadable.
Every client has its own cloud environment. No shared storage, tools or knowledge base between clients.
Encrypted in transit and at rest. Your key unlocks the data, and revoking it locks everyone out, including us.
Enterprise AI services under commercial terms that exclude training. We never train or fine-tune a model on client data.
Agents that touch client data cannot reach the internet. They read only that engagement's data and write only to its workspace.
Only the people named in your data schedule, through single sign-on with hardware security keys, on company-managed devices.
Every access and agent run is logged. All data is deleted within 30 days of close, with a signed certificate of destruction.
Your data never trains a model, by provider terms and by our contract. General-purpose models are used as they are, and the knowledge stays in the documents we hand you.
Our reusable assets are methods: code, prompts and templates with no client data in them. One client's data, findings or benchmarks are never used for another.
No work for a direct competitor at the same time without written consent from both. Cross-client benchmarks only with consent, anonymized and aggregated.
We put these commitments in writing, and we are happy to work from your paper.
Before any data moves, we hold a short call with your security lead to choose the tier and approve the data schedule. The trust pack answers the rest.
The three tiers, the architecture of a client environment, and how data flows through it.
Which models and services we use, under which terms, and how agents are controlled.
Standard commitments, the data schedule template, subprocessor register, policy summary and questionnaire answers.
We'll return the trust pack, and answer your own questionnaire if you have one.
Request the trust pack