Measurement
Outreach metrics: separate activity from customer outcomes
By ColdPegion editorial · · 4 min read
A trustworthy report distinguishes what happened, what is inferred and what remains unknown.
Define stages and units before the pilot
Research, verification, import, preparation, queueing and sending are separate events. If your system exposes only a queue count, report that count. Do not promote it to delivery because the stronger claim sounds more useful.
Keep the unit clear. Companies, addresses, unique recipients and individual messages are different counts. Several addresses may belong to one company, and a sequence may contain several messages per person. Decide which unit answers the question before comparing campaigns. Preserve the observation window and the campaign identifiers in the report.
Match the denominator to the question
For a human-reply rate, define a group of unique recipients and a fixed observation window. Today’s replies may belong to last week’s sends. Dividing those figures without checking the cohort can produce a misleading result even when the arithmetic is correct.
Publish both numerator and denominator when reporting a rate. One reply among five recipients and twenty replies among one hundred recipients have the same arithmetic rate but different amounts of evidence. Neither establishes what a much larger audience will do. Avoid treating a small pilot as a delivery or sales forecast.
Read replies before claiming interest
An inbound message can be automatic, an opt-out, an unrelated solicitation or a real conversation. A classifier label is a triage aid. Inspect the content and its relationship to the campaign before counting an interested prospect. A message in a connected inbox is not automatically an outreach reply.
An opt-out is also an operational event. Apply the supported suppression process and keep the contact out of later campaigns. Do not continue a sequence simply because additional steps were planned before the reply arrived.
Connect conversions with evidence
Use campaign UTM links where appropriate and preserve attribution in the signup flow when your system supports it. Keep account creation, activation and payment separate. Define activation in advance—for example, connecting a mailbox or preparing an inspectable campaign.
If attribution is unavailable, say so. A new account during a campaign is not automatically a campaign conversion. You can still report observed sends, conversations and product events without inventing causation. End with a concrete next experiment that follows from the evidence rather than a claim that queued work has produced customers.
Sources and useful links
Practical guidance from the ColdPegion team. Examples are illustrative, not customer results. Use the current product documentation for supported operations.
Try the workflow with your own brief.
Connect your AI agent to ColdPegion, prepare a small campaign and inspect the messages before launch.