Digital Marketing Campaigns Best Practices: Test Decisions

Digital marketing campaigns best practices start with one honest question: what are we changing, and what would we do differently if it wins? The split test answers the first half. The decision memo answers the second.

What to take away

Two primary metrics, chosen before launch

Google Ads custom experiments compare a changed campaign against the original, split traffic and exposure between them, and let you name up to two primary success metrics. The cap is the useful part. It forces the team to agree on what the test is for.

Custom experiments are one split-test design among several. An A/B/n test compares more than two variants at once, a multivariate test varies several elements together, and a geo-holdout compares matched regions. The same two-metric discipline applies when the test runs in email, social, display or retail media, so name the design in the brief before launch.

Choose one metric that moves during the test window and one that confirms the change did not damage something else. A cost-per-lead test pairs with a qualified-lead rate. A landing page test pairs with revenue per session.

Write both into the analysis plan with the population, the exclusion rules and the date the outcome matures. A metric named after the results arrive is not a success metric.

The outcome the platform cannot see

Custom experiments report what the ad platform recorded. Accepted opportunity, first product value, collected order, retained subscription and completed renewal live somewhere else.

Reconcile the two. Export the experiment segment, join it to your own outcome table on the same identifier, and state the join rate. If a fifth of experiment clicks never appear in the outcome table, the test result carries that gap.

Add the business outcome to the analysis plan alongside the two platform metrics. It is the number the budget decision rests on.

Economics the experiment does not include

Media cost is the visible line. Creative production and tooling sit outside the experiment report. Agency hours and discounts also sit outside it. Sales time, service time and refunds sit outside it too.

Give the arithmetic in your own variables: contribution per qualified outcome equals revenue per outcome minus media cost per outcome minus fulfillment and service cost per outcome. Substitute your figures. No published benchmark substitutes for them.

Review marginal performance, not average. The tenth thousand dollars rarely behaves like the first.

Close it with a decision memo

A campaign that ends without a memo gets relitigated. The memo is short: what was tested, what the two primary metrics did, what the business outcome did, what uncertainty remains, and what changes next.

Archive the brief, the analysis plan, the settings and any mid-flight deviations, the source exports, the approvals and the incidents. Stop expired offers and automation. Remove obsolete assets and revoke access the campaign no longer needs.

Assign the closure review to a named owner with a deadline. Unowned closure slips indefinitely.

Worked example: a two-metric test

A team wants to know whether a shorter lead form raises qualified volume. They run a custom experiment with form length as the only change.

Metric one is cost per lead. Metric two is qualified-lead rate, measured against the CRM thirty days out. The analysis plan fixes the population, excludes existing customers, and sets the read date.

At the read date, cost per lead falls and qualified-lead rate falls further. The form collected more names and fewer buyers. The memo records the result, the thirty-day lag, and the decision to revert. Next quarter's planner starts from that, not from the cost-per-lead headline.

Records that support the review

The NIST Privacy Framework starting guide describes a voluntary process for identifying and managing privacy risk. Use it to assign owners for the data a campaign collects and the responses it triggers. It is not legal clearance, and state privacy obligations — California's CCPA, for instance — vary, so confirm specifics with counsel.

The CISA logging guidance explains how event records support security review. Preserve access, change, failure and correction records for the campaign. Logging shows what happened; it does not validate a metric.

Before launch, screen the journey with W3C's preliminary accessibility checks. That resource covers selected issues and is not a conformance evaluation. Complete the testing your standard and risk require.

Common questions

These are the questions that come up at the end of almost every test.

How long should a custom experiment run?

Long enough for the slower of your two primary metrics to mature, and long enough to cover a full business cycle. Set the read date in the analysis plan before launch. Stopping early on the faster metric is how teams scale noise.

What belongs in the decision memo?

The hypothesis, the two primary metrics, the business outcome, the uncertainty that remains, and the next action. Add the archive location and the owner. One page is usually enough, and a longer memo tends to bury the decision.

Can we skip closure if the campaign worked?

No. A winning campaign still needs its settings, source exports and deviations archived, and its expired offers and automation stopped. The memo records why it worked, which is what makes the result repeatable. File it with your digital marketing campaigns examples, so the next planner finds the decision and the evidence in one place.

Who owns the closure review?

A named person with a deadline, written into the brief at kickoff. If the owner is a team, nobody closes it. Recurring questions about ownership and cadence are collected in the common campaign questions.