FDE
Deployment and control
We do not ship a box and we do not ask you to "deploy it yourself". A BusinessOS engineer goes inside your process, takes it apart with its exceptions and assembles a digital employee on your data. Control is built into the model, not bolted on.
Why forward-deployed engineering lowers the risk
The process is taken apart for real
At the first meeting staff rarely describe how a process actually runs. The engineer sees exceptions on real examples and builds them into the rules — not into a slide.
The promise is tested on data
We do not promise an "optimal plan" or "replacing a department". We promise a concrete result on your data by a checkpoint — and show it.
The core stays with you
After the pilot the company owns an asset: object map, rules, memory, metrics. The second digital employee does not start from zero.
Four steps
- Review1 weekOne process, one owner, input and output examples, success criterion, data and access map.
- Pilot4–8 weeksThe digital employee works on your data under supervision. Checkpoints, expert comparison, training of the pilot group.
- Regular operationsubscriptionAgreed scope, support, metrics in the outcome log. Autonomy expands by action class.
- Scaleby resultNext roles on the same core — cheaper and faster. Expansion is decided on measured effect.
Governed autonomy
A leader's biggest fear: "AI will start making risky decisions on its own". In BusinessOS this is impossible by design: autonomy is set explicitly, as a ladder, separately for each class of action.
Never become autonomous: money, contracts, legal obligations, hiring, material commercial promises. This is not a setting — it is a limit built into the system.
Every action is stored with its reason, context and owner. If the agent errs, it is not "AI did something" but a concrete review: where the rule broke and how to fix it.
Where and on what it runs
In your perimeter
A Linux VM, Docker distribution, user access over HTTPS. Source code is not handed over, data does not leave the perimeter. Backup for the pilot period — a secured BusinessOS stand.
Any language model
Claude, GPT, DeepSeek via API on anonymised fragments — or an open model inside the perimeter. No vendor lock-in. Company data is never used for training.
Connections
1C (read and, by agreement, document creation), email, messengers, file storage, MSSQL and data warehouses. Each connection is configured and accepted separately.
Permissions and audit
Role- and workspace-based access, external sending only to approved recipients, a log of every action with reason and owner, versioned rules.
What we fix before a pilot starts
On one page: which process is tested; which data and access are needed; what our team does and what the client does; what result we commit to show; how quality is checked; which actions in working systems are prohibited or require approval; scope and period; deadline after materials; cost; acceptance date and participants; data handling after completion; the next commercial step on success.
How pricing works
Launch
Process review, digital employee configuration, connections, verification on client data, pilot group training.
Regular operation
Subscription for the agreed function within scope: execution, support, metrics in the outcome log.
For results
Where the effect is measurable, payment is tied to it: reduced manual time, processed items, closed requests.
External service and language model costs are stated separately. Price is not built by comparison with an employee's full salary when the digital employee takes one of their functions.
A review of one of your processes. 20 minutes, no slides
You show how the work is done today: what comes in, what should come out, where time is lost. We show which digital employee fits, what it will do on its own, where it stops and asks a person, and how we will count the effect in money.