Skip to main content
Your agents can keep three things as data in the workspace, instead of in a message someone has to remember: the goals you set, a log of the experiments they ran, and the outreach funnel behind both. You don’t open a page for any of it. Ask an agent in its DM or in a room, for example “Set a goal of 150 connection requests a day” or “How is each sender doing this week?”. These work in every workspace, on every plan, and use no credits beyond the message you send.

The outreach funnel

Ask an agent for the funnel and it reads the results of your outreach Workflows: connection requests sent, accepted, declined, withdrawn and still unanswered, messages sent, conversations, replies and positive replies, with the accept, reply and positive-reply rates.
  • Split it by Workflow (campaign), sender account, A/B variant, step, day or week, up to three at once. Totals are always included.
  • Pick the period: a number of days or a date range, in the time zone you name.
  • Small samples are marked. Every rate comes with the number it rests on. Under the minimum sample (30 sends unless you say otherwise) the agent is told the rate is not yet evidence and how many more sends it needs.
  • A reply or an acceptance is counted on the day its message or request was sent, so recent days keep rising while answers arrive.
  • Replies are counted once per conversation, the same way as on Analytics.
The funnel covers the Workflows the people in the conversation can open. Booked calls are not a Workflow step; track them with a goal (below).

Goals

Tell an agent a target and it saves it as a goal: a metric, a target, a period (day, week, month, quarter or a fixed date range), a time zone and, if you want, the agent responsible and one Workflow. Connie measures these itself every time the goal is read: Any other metric (post comments, followers gained, ad conversions) is a reported goal: an agent reads the number from its source and records it for the period, with a note saying where it came from. A goal on a metric Connie measures itself takes no reported value: to track a number you count yourself beside it, ask for a second goal under its own metric name. When you ask “Are we on target?” the agent reads every goal with its value so far, the target, what is left, how much of the period has passed, where it will land at the current pace, and last period’s value.
  • There is one active goal per metric, period and Workflow. Setting it again changes the target; it does not add a second goal.
  • A spending goal (“no more than 2,000 credits this month”) is a limit to report against. It does not stop anything from running.
  • Any member can set a goal or record progress; guests can’t. Progress on a credits goal is shown only to owners and admins.
  • To retire a goal, ask the agent to archive it.

Experiments

When an agent wants to try a change (new message copy, another audience, a different send time), it logs an experiment first: the hypothesis, the one thing being changed, the variants, the metric that decides, and the minimum sample each variant needs.
  • One test at a time. A second experiment can’t start in the same Workflow (or the same named scope) while one is running, because the two could not be told apart.
  • No repeats. A proposal that matches a test already concluded is not started. The agent gets what was tried, the result and the decision taken then. It can run it again only by saying what has changed since.
  • No early verdict. Until every variant reaches its minimum sample, a read-out shows only how many more are needed and roughly how many days that takes. It shows no rates and no leader. A result of won, lost or inconclusive can’t be recorded before then either; a test can always be abandoned.
  • A real comparison. Once the sample is reached, the read-out gives each variant’s rate, the difference to the control and whether that difference is statistically significant.
  • Variants tied to a Workflow, a step or a sender account are measured from the outreach funnel. For anything else (content formats, ads, landing pages) the agent records the numbers it read from the source.
Logging an experiment changes nothing by itself. The agent still creates the variant and applies the decision, with the usual approvals. Ask “What have we tested so far?” for the log: every experiment, running or closed, with its result and the decision.

Credits by agent

Owners and admins can ask an agent what each agent spent: credits per agent and per day over a date range, with the work that belongs to no agent (Workflows, MCP and API calls) shown separately. See Credits.

Good to know

  • Goals and the experiment log belong to the workspace, not to one conversation. Any agent a member talks to can read them.
  • Guests can’t read goals or the experiment log. In a room that a guest reads, or a conversation shared by public link, an agent does not show them; ask in a conversation only members read.
  • Days for the funnel and for goals follow the time zone given; credits by agent uses UTC days.
  • Nothing here sends a message, spends credits or switches a Workflow on.