Built-in tool

Start an experiment

Log a new experiment: the hypothesis, the one variable being changed, its variants, the deciding metric and the minimum sample.

Freeno per-call chargeNo API key needed

Endpoint

Parameters

NameTypeRequiredNote
namestring · requiredstringrequired

Short name, e.g. 'Opener: question vs statement'.

Example: Opener: question vs statement

hypothesisstring · requiredstringrequired

What you expect to happen and why.

variablestring · requiredstringrequired

The ONE thing that differs between the variants, e.g. "first message opener". Everything else stays the same.

Example: first message opener

variantsobject · requiredobjectrequired

Two to six variants; the first is the control (what runs today).

primaryMetricstring · requiredstringrequired

The one number that decides it: 'accept_rate', 'reply_rate' or 'positive_reply_rate' are measured from the outreach funnel for variants bound to Workflows, steps or senders; 'custom' is any other rate you report yourself.

One of: accept_rate, reply_rate, positive_reply_rate, custom

minimumSamplenumber · requirednumberrequired

How many each variant needs in the metric's denominator (requests sent for accept_rate, conversations for reply rates) before it may be judged.

metricNamestring · optionalstringoptional

For 'custom': what is measured, e.g. 'calls booked per positive reply'.

Example: calls booked per positive reply

scopeLabelstring · optionalstringoptional

When not tied to one Workflow: the area the test occupies, e.g. 'linkedin content' or 'google ads'. One test runs per area at a time.

One of: linkedin content, google ads

workflowIdstring · optionalstringoptional

The Workflow (campaign) this test runs in. One test runs per Workflow at a time.

ownerAgentIdstring · optionalstringoptional

Agent responsible for this test. Defaults to the agent calling.

repeatReasonstring · optionalstringoptional

Only when this tool answered that the test repeats a logged one: what has changed since, so running it again is justified.

Get started

How to use this tool

  1. Connect your AI

    Add Connie to ChatGPT, Claude or Slack once, and this tool comes with it. There’s no API key or account to set up.

  2. Ask for the outcome

    Describe what you need in plain words. Your AI runs this tool when the job calls for it, along with any others it needs.

  3. Make it a workflow

    Add it as a step in a workflow that runs the same way every time, on a schedule or when something happens.

  4. Stay in control

    Choose whether it runs on its own or asks first. When it asks, you see an Allow or Deny card before any credits are spent.