Build a 90-day plan that leaves you wiser

A long optimisation checklist can create a busy quarter without making the account easier to understand. When several settings change together and nobody records the hypothesis, even a positive result can teach very little.

A stronger plan is a sequence of decisions. It identifies the important constraint, proposes a limited intervention and defines what evidence will determine the next move.

Choose a small portfolio

The Strategy Atlas suggests three to five initiatives as a practical starting range, adjusted to the business and team. That is not a quota. A low-volume account may support one principal performance hypothesis at a time; an urgent measurement problem may dominate the first phase.

Each initiative needs an outcome, an owner, a hypothesis, linked changes, dependencies and an evaluation gate. “Improve account performance” does not identify a testable mechanism. “Repair the reproduced mobile enquiry failure, verify tracking and review qualified enquiries” is much more concrete.

Sequence by dependency

Use the opening period to establish the brief, commercial limits, measurement and baseline. The next phase can address verified corrections and launch suitable tests. The middle of the quarter is for observation, evaluation and refinement. The final phase is for justified rollout and the next backlog.

These windows do not require a bidding change in a particular week. A long sales cycle may mean the quarter produces credible pipeline evidence before customer outcomes mature. Preserve that distinction in the plan.

Reserve time for monitoring, data preparation and unexpected repairs. A schedule that allocates every hour to new interventions leaves little capacity to notice when an assumption fails.

Write the decision rule before the result

Record a primary outcome, guardrails, exposure limits, maturity conditions and rollback options. Distinguish configuration verification from outcome evaluation. The first confirms that the change was applied; the second asks what followed and how confidently it can be attributed.

Where the platform offers a suitable experiment, use it for a coherent question. For example, Google documents Performance Max experiments for several scenarios. A supported feature still needs a sensible scope, adequate evidence and attention to interference.

Keep an honest learning record

In a fictional quarter, a test may produce too few mature outcomes to distinguish a useful improvement from noise. “Inconclusive” is more valuable than inventing certainty. Record what prevented the decision and whether to extend, redesign or stop.

Seasonal events, measurement changes and concurrent work can also alter interpretation. Preserve the original hypothesis and annotate those changes instead of rewriting the plan to make the eventual result look inevitable.

At the end of the quarter, ask what the business now knows, which changes are worth retaining and which uncertainties deserve investment next. A good plan can produce fewer edits and better decisions at the same time.

Atlas reading: S19 — Controlled growth and experimentation; S20 — Operational discipline and learning; S18 — Seasonal offers and demand timing; 90 Day Planning; Learning Record.