Turn Campaign Studio into a measurable experiment: a field-by-field guide

Begin with an experiment contract

Before using Campaign Studio, write: “For [customer in a specific situation], [message] should produce [useful action], because [evidence]. We will review [period] and protect [quality or cost limit].” A campaign called October launch is difficult to evaluate. A campaign testing whether a reading checklist attracts suitable consultation requests has a decision built into it.

The studio does not buy placements, deliver messages, or verify your revenue. Its job is to hold a plan, create exportable artwork, and summarise measurements you enter. Inspection date for this guide: October 11, 2026.

Brand fields establish who the experiment serves

Treat proof and restricted claims as approval boundaries. A claim about the service’s included deliverable is different from a claim that children will improve by a certain amount. The person responsible for delivery should validate the former; the latter needs appropriate evidence. Do not let a confident tone disguise an untested proposition.

Brief fields define the eligible customer and destination

Add the decision rule to your working notes: what counts as qualified, how much evidence you need, and what would cause you to pause. For a lead campaign, a booked call from an unsuitable customer is not success merely because the form worked. Select the outcome you can reconcile with an independent business record.

Creative fields define the intervention

An angle comparison only helps if the competing treatments are recognisable. Keep a record of the actual exported file, copy, launch date, placement, audience, and destination version. The tool’s three angle labels are categories, not an immutable version history. If you edit an angle halfway through the reporting period, document that change before pooling its results.

Tracking labels are a measurement contract

The tracked URL has four core labels: source is the referrer, medium is the delivery category, campaign identifies the test, and content identifies the creative. The studio generates source and medium from the channel; for most social options the presence of a positive budget selects paid_social rather than organic_social. Google Search, email, and community options use cpc, email, and referral respectively. Review the output against how distribution actually happened.

UTMs describe acquisition; they do not prove incremental impact. Keep them out of internal navigation and avoid names or email addresses in URLs. A copied link can move to another surface, consent can reduce observability, and another device can break the recorded journey. Reconcile ad records, analytics, CRM qualification, and order data without pretending they are identical systems.

Each performance input needs a source and scope

Choose daily, non-overlapping rows for each angle wherever practical. The reporting date does not protect you from overlap. A $50 Monday row followed by a $90 Monday–Tuesday row yields $140 recorded spend even though only $90 was spent. Separately summed nonblank fields can mix coverage periods. Keep a completeness register outside the studio when the dataset is partial.

Read the summary through the economics

Suppose the two $120 orders generate $240 revenue but require $140 in variable fulfilment and fees. After $120 media spend, contribution is negative $20 before fixed costs. The 2× ROAS has not changed; its business meaning has. An apparent CTR winner can also create lower-quality inquiries and more sales effort. Pair the lead rate with qualification reasons and contribution per customer.

Make the record useful for the next decision

The current CSV export excludes the angle field, so preserve a version-to-angle map before analysis. The JSON campaign brief retains measurement records and their angle identifiers. There is also no dedicated subscriber metric, experiment randomisation, significance test, or automatic reconciliation with your CRM. Record these limits in the brief instead of using the interface as a substitute for them.

Choose a review action that matches the uncertainty. Broken tracking warrants repair, not an audience pivot. Several relevant visits with no usable next step warrant an experience review. A promising small sample warrants a limited follow-up test, not a claim that an angle has won. Preserve local drafts and exports; optional cloud sign-in does not automatically upload guest work or its local photo. End with a named owner, a decision date, and the observation that would change your mind.

Continue the traffic and growth learning path

Find the next useful guide in this 16-part series →