Sample Company, 500 people
Each Mapped Margin client gets this dashboard for its own AI agents. This copy runs on sample data. Sample Company is invented, and so is each name, number and record on the page. It runs the eight agents from the sample Blueprint on our site and the nine use cases, more departments than one company of 500 would hold.
Your changes stay in this page, and a reload resets it to the sample data.
New sample cases arrive every three seconds while the clock runs.
Agents
Each agent reports to one owner, a role on your team, and that owner gets its exceptions. Below its confidence threshold the agent stops and hands the case to the owner. Pause an agent and your team runs the manual procedure in its runbook until someone resumes it.
| Agent | Status | Owner | Works in | Cases today | Exceptions today | Threshold | Last action | Pause or resume |
|---|
Moving a threshold reruns today's sample cases at the new setting, so the exception count shows what that setting would have sent to a person. In production a new threshold applies from the next case.
- Running: the agent works live cases and hands the unclear ones to its owner.
- In parallel run: your team still works each case, and we compare the agent's result with theirs before it takes live work. Its exceptions are counted, not queued.
- Paused: the agent takes no new case, and your team works from the runbook.
Exceptions
The agents handed these cases to a person. Each shows the reason, the agent's confidence and the record it points at. Approve lets the agent carry out the action it proposed. Send back returns the case to the agent to run again, and Reassign hands it to another role.
Log
The log holds each document an agent read and each record it wrote. The cases it handed to a person sit in the same log, and so does each control action taken on this page.
| Time | Agent | Action | Record | Detail |
|---|
Measurements
We measure each live agent against its Blueprint forecast at 30 and 90 days after go-live. Before go-live each agent runs in parallel with your team, and we measure its error rate against theirs. If a result lands below the forecast, we tell you the week we find it.
Savings at a measurement are the measured hours at the cost per hour in the agent's Blueprint row. A paused agent drops out of the measured run rate until it resumes. The red flag marks a measurement more than 5 percent under its forecast.
Hours returned a month, forecast against measured
Days from trigger to done, before and after
Error rate in parallel run, your team against the agent
Live agents against their forecast
| Agent | Live since | Hours forecast | At 30 days | At 90 days | Savings forecast | Savings measured | Days before | Days after, forecast | Days after, measured | Next measurement |
|---|
Agents in parallel run
| Agent | Since | Cases compared | Your team's error rate | The agent's error rate | Fewer errors |
|---|
Access
Each agent signs in under its own service account, with the narrowest permissions its task needs. Revoke pauses the agent and marks its account revoked. The agent stays paused until your IT team grants the account again.
Service accounts
| Agent | Service account | Permissions | Granted by | Granted on | Status | Revoke |
|---|
Your users and roles
| User | Role | Access | Owns |
|---|
- Admin: all controls, access changes and the export.
- Owner: pause, resume, threshold and exceptions for the agents the role owns.
- Reviewer: works the exceptions assigned to the role.
- Viewer: reads the log and the measurements.
- Vendor support: reads the log and the measurements, and can pause an agent. Each pause it makes notifies the owner.
Handoff
You are not locked in. If you want another firm or your own team to take over, we hand over the files and complete control within 24 hours. The agents keep running through the handoff.
The accounts each agent signs in with transfer to a person you name.
What the export contains
Agent code and prompts
The repository for each agent, with its prompts and settings.
Runbooks
The manual procedure for each agent, the one your team runs during a pause.
Logs
Each read, write, handoff and control action, with the audit trail.
Account handover
Each service account with its permissions and who granted them.
In this demo the button builds a manifest of the sample export in your browser and downloads it as a JSON file. The page sends nothing anywhere.