How to evaluate a new feature before adopting it: the six-step protocol and the monthly cycle

The Advanced level — and the course — closes with the method for the only thing that doesn't change: that Google will change. A six-step protocol for any new feature, an example, and the monthly cycle that keeps this module alive and your accounts up to date.

The six-step protocol

1. Read the official source

The announcements page or the blog, not the industry headline. Note down: what exactly it does, for which campaigns, whether it is beta or general, whether it switches on by default, whether it has a deadline.

2. Classify it

Mandatory · on by default · optional and relevant · irrelevant (lesson 1). The first two go into the monthly routine, no exceptions.

3. Check what switched itself on

In each account: campaign and account settings, change history by client_type, applied recommendations. If something switched on without a decision of yours, set the baseline from that date (module 3) before judging it.

4. Estimate the impact with your own data

Which campaigns does it affect? How much spend? What do the curve, the IS, the POAS and the judge say about those campaigns? A feature that promises "+15% conversions" is evaluated against your marginal return, not against Google's average.

5. Test with an experiment or a pilot account

• Campaign features: custom experiment (50/50, 4+ weeks, one variable).
• Account features, or where no experiment is available: a pilot account (in a manager account) or a pilot campaign with a capped budget, with the judge's TEST mode logging.
• Never the whole account at once.

6. Decide and document

Adopt, limit, discard or postpone; with the result, the date and the planned review in your news log. Update the architecture document if the design changes, and the scripts' configuration if what they watch changes.

Example: Smart Bidding Exploration in Shopping

1. Source: the GML 2026 blog — it explores auctions it would normally discard, with a more flexible target ROAS; it reaches Shopping.
2. Class: optional and relevant (ecommerce); watch out for defaults.
3. Switched on by itself: not in this account (checked in the settings).
4. Impact: two high-margin Shopping campaigns are overdelivering on ROAS with 30% lost IS (budget) → there is room to grow; the low-margin ones are not.
5. Test: a 50/50 experiment on the high-margin campaign, 5 weeks, primary metrics POAS and new converting users.
6. Decision: if POAS holds up and new users go up, adopt for high margin; discard for low margin. Document it; adjust the judge's band for the campaign with exploration on.

The monthly cycle (of this module and of your accounts)

1. The month's news (news agent / your sources) → a classified list.
2. Affected lessons in this course → they get updated (new date) and the monthly email tells you.
3. Accounts: check what got switched on by default; baseline where needed.
4. Evaluations under way: read the results; decide.
5. Log and architecture document up to date.
6. Scripts: configuration adjusted to what's new (bands, exclusions, reports).

What doesn't change

The whole course has been built on things new features don't touch: the auction rewards quality; honest measurement is the master lever; money follows the marginal return; the ad filters who comes in; Google automates what suits Google and leaves the tedious work behind; the deterministic goes to a script and the unpredictable to a person (or a supervised agent). When a new feature seems to contradict one of those principles, suspect the feature before the principle.

💡 Ninja trick: this module comes last because it is the only one that never ends. The Suite's news agent collects every week, the Academy updates every month, and the monthly email tells you which lessons have changed. And if a new feature opens a gap Google leaves uncovered, on the last Friday of each month the agent proposes the script that would fill it. That is how much of the Suite was born; that is how it will carry on.

What you should remember

• Six steps: source → class → what switched itself on → your own impact → experiment or pilot → documented decision.
• What is on by default is judged with a baseline; what is optional, with an experiment; never the whole account at once.
• Monthly cycle: news → lessons → accounts → evaluations → log → scripts.
• What doesn't change is what holds the whole course up.

End of the Advanced level and of the course. Thank you for making it this far. The Academy is updated every month; the 💡 button is there for whatever is missing.