AI for agencies, with controls

• It has no access to your numbers. So it answers in general terms.
• It cannot tell who is asking. A junior and an owner get the same answer.
• It leaves no trace. Nobody can check where an answer came from.

The AI in Agency Root is a layer over structured business data: people, clients, costs, meetings and documents. Each material output follows the same four steps:

1. 1
   Source. Where the information came from, with a link to the record when you are allowed to open it.
2. 2
   Interpretation. What the data shows, with facts kept separate from inferences.
3. 3
   Recommendation. What the AI suggests doing.
4. 4
   Review. A person checks and approves before anything leaves the agency.

When the evidence is thin, the AI is designed to say so instead of guessing.

The Agency Root plan includes a basic AI assistant that can:

• summarize information and explain dashboard metrics;
• write the narrative for internal reports;
• draft announcements and SOPs;
• answer simple questions over the data you are permitted to see.

Basic AI assists. It does not execute commercial actions, generate contracts or run agent workflows.

In Agency Scale, the CEO AI assistant answers business questions in plain language, within the user's permissions. It is read-only by default.

Leadership reports can be scheduled: a daily morning brief, a weekly report and a monthly business review. Alerts cover finance, clients, people, attendance, operations and the calendar.

An AI agent is an AI capability with one defined job, its own permissions and its own outputs.

An admin sets how much each agent may do:

1. 1
   Suggest only. The agent analyzes and recommends. Nothing is sent or changed.
2. 2
   Prepare and wait for approval. The agent prepares the document or message, then waits for an authorized person. This is the default for quotations, emails, contracts and minutes.
3. 3
   Controlled automation. The agent acts inside rules, recipients and thresholds that an admin approved in advance. An admin has to opt in.

Autonomy can be set per agent, role, client, deal type and amount. Permission to use an agent is not permission to execute everything it can prepare.

• Tenant and role permissions are checked before the AI retrieves anything. It cannot return data the user could not open in the product itself.
• Human approval is the default for external communication, commercial commitments, price or discount decisions and legal documents.
• The AI does not override a price, exceed a discount limit, send an email or contract without authorization, or change source financial data.
• Outputs are versioned, and AI activity is recorded in the audit history.

• Agency Root uses AI and speech providers only under terms that prohibit training on customer data.
• Each request sends the minimum context needed, not the whole company dataset.
• Sensitive records such as IDs and salaries are retrieved only when the user is authorized and the question requires them.
• An agency can turn AI off for the whole company.

More detail is on the security page: AI privacy and provider controls.

Agency Scale customers can opt in to Labs, an area for beta agents and experimental features. Each one states its purpose, known limitations, data use and how to opt out.