The brief

The brief is a short document about your agent: what it is for, why people call, how a call goes, what counts as a good one and what it is not meant to do. SignalMD drafts it after you connect, from your agent’s configuration and the calls you choose to import. You correct it, and you approve it. Nothing is analysed before you approve.

​

Why it matters

Your configuration says what the agent does. It cannot say what you want. A transfer to a person is the goal for one business and the failure for another. A caller who asks about billing is a missed call for one agent and correctly turned away by another. The brief is where you settle those questions once. It decides:

• What counts as a defect and what is designed behaviour. Something the brief says the agent is not meant to handle is not counted against it, and an ending you mark as good is not a failure.
• What a good call is. SignalMD reads every call against the purpose and the success line, so “got what they came for” means what you mean by it.
• Which parts matter most. Issues located at a part you mark High move up the board, and issues at a part marked Low move down. See How the board is ordered.
• What your tests check. A test built from a call can point at a line of the brief, such as an intent’s Done when. Edit the line and every test that points at it changes with it. See Test a fix.

​

How SignalMD drafts it

SignalMD reads the agent’s configuration from Retell (its steps, prompts, tools and knowledge bases), imports the calls you choose (how far back and at most how many, 30 days and 500 calls unless you change them), and then writes the document from both: the configuration says what the agent can do, and the calls, read in full, show why people ring and how the agent answers them. The same panel sets what percent of new calls to keep after setup and an optional cap on them per day, week or month. If you skip the import, it writes the document from the configuration alone. It asks about whatever the configuration and the calls could not settle. The draft is a hypothesis about your business, so read it as one. Beside the brief, What recent calls show counts the calls it was drafted from, as the platform recorded them: how many and on which dates, how they ended, the last part each one reached, how long they ran and any tool failures. Nothing in it is judged, and it does not change the brief. If the window had no calls, the brief is drafted from the configuration alone and the page says so. The tag on the panel tells you where the text came from:

​

What is in it

You can edit any section in place.

​

What to review before you approve

You do not need to polish every section. These carry the most weight:

1. A call worked when. If this line is vague, every success rate is vague. Make it something you could check on one call: “The caller leaves with a confirmed appointment, or a booked callback if the clinic cannot take them.”
2. How a call can end. A transfer stops counting as a failed call only when every transfer ending is marked Good ending. One Bad ending, Depends or Say which and transfers stay failures.
3. Not this agent’s job. List what the agent should turn away, so turning it away is not read as a miss.
4. The parts. Set Matters to High for the parts where a failure costs you most.

​

Answer the questions

Questions appear in the Calibrate panel beside the brief, one at a time, with the reason each one matters and often answers you can select. Your answer is written into the right section of the brief, and that section is highlighted. The footer counts them, such as “2 of 5 questions settled”. Questions you have not answered are listed under Needs your confirmation. If the assistant is off for your workspace, the panel says so and you edit the brief directly. Each message you send in the Calibrate panel uses credits from your plan’s balance. See Pricing.

​

Approve it

Select Approve and analyze when calls are waiting to be analysed, or Approve when there are no calls. SignalMD needs a purpose and a line on what makes a call successful; without them it says “The brief needs a purpose and a line on what makes a call successful.” Above the button, the page says what analysing the imported calls uses, at 1 credit per processed minute, and how many credits your plan has left. Approving saves the brief as written, sets the purpose, success line, endings and parts that your calls are measured against, and approves the call reasons under What each call is for. It then starts the analysis of the calls already imported. If the import is still running, the analysis starts when it ends. With no calls, approving only saves the brief, and you can import calls later from Import production calls on Agent config. With no credits to analyse the calls, on Free or once a month’s credits are used, the button reads Choose a plan to analyze. It approves the brief and opens your plans. The analysis starts on its own once your plan has credits.

​

Change it later

Open Agent config, select Edit, change the text, and select Save and re-approve. Saving does not change calls already on the board. They were read against the brief as it was at the time. To read every stored call against the new text, select Re-analyze on Agent config and confirm. Changes to Matters and to how endings are marked also reach the order of issues as new calls are analysed and when a new version arrives.

• A new Retell version does not touch your edits. SignalMD records the new configuration but does not redraft the brief.
• Draft again replaces everything. It writes a new brief from scratch and removes your edits and the conversation with them, after asking you to confirm. If calls are already imported, the new draft groups the parts by how those calls moved through your flow.
• Put designed behaviour in the brief. If SignalMD flags something your agent does on purpose, answer No on the issue, pick Meant to work that way, then add the behaviour to the brief so the next reading expects it.

​

Next steps