# Micro-business / TPE track — 30 days

## Target

A small team in which the same people cover operations, management, IT, and customer relationships.

## Intended outcome

One shared process improved by AI, with an owner, common rules, a manual procedure, and incident-response capability.

See the [complete synthetic case: from shared inbox to validated copilot](../../examples/en/tpe-customer-requests.md).

## Week 1 — Visibility

- name an AI lead and a process owner;
- inventory official and informal AI tools;
- publish a one-page usage rule covering prohibited data, verification, approvals, and reporting;
- map one frequent process and its exceptions;
- measure the baseline.

Deliverables: [AI system register](../../templates/ai-system-register.csv), [mandate](../../templates/mandate.md), and process map.

## Week 2 — Case and supplier

- compare three cases on value and risk;
- select a reversible case with human validation;
- classify risk and autonomy;
- assess the supplier, its subprocessors, data practices, costs, and exit options;
- define who may configure, use, approve, and inspect logs.

## Week 3 — Tests and shadow mode

- build 30 to 100 authorized real cases covering users and exceptions;
- preregister thresholds and severe errors;
- test missing data, malicious content, wrong recipients, and unavailability;
- run without real-world effect and measure human corrections.

## Week 4 — Copilot and decision

- train every role on the actual system;
- authorize a limited pilot with approval;
- monitor quality, time, cost, incidents, and workarounds;
- hold a gate review and record whether to continue, correct, or stop.

## Controls not to sacrifice

- separation between production and test;
- individual accounts, not shared credentials;
- read-only access by default;
- tested backup and manual process;
- supplier and change review;
- an accountable person reachable when something goes wrong.