The module closes with the practice: how the shield is deployed in an account, what modules it is made of, how to read its Dashboard and what is left in the client's hands. It is written around Ninja Shield, but it describes what any serious shield should have.
Deployment
Ninja Shield is one of the scripts installed by the team from the MCC (not self-installable): the code lives in the MCC and never in the client's account. The pieces:
- The client's spreadsheet (created from the master template): Configuration, scenarios, Dashboard, a tab per module, Safe Sites, False positives, Manual.
- The loader in the MCC with the account's licence.
- The visitor snippet on the client's website (WordPress plugin, or a proxy for Next and others): it captures the fingerprint of every visit (the same one that feeds Lead Rating) and sends it to a collector.
- The site analyser on our own server (not in Apps Script, because many farms block Google's IPs): it scores domains every night and writes the scores into the spreadsheet.
Typical installation: spreadsheet + loader + snippet in one day of the team's work (the goal: 15 minutes once sign-up is automated).
The modules
| Module | What it does |
|---|---|
| Site Analyzer | Scores every domain you appear on (farm and trust signals); writes Site Scores |
| Farms | Clustering by IP, NS, RDAP and the ads.txt pub-ID; excludes networks; a census shared across accounts |
| Visitor Shield | The fingerprint of every visit to the site: probability of being non-human |
| IP Shield | Account IP exclusions built from repeated fingerprints, with an expiry and respecting the limit |
| App Shield | Apps by score and category; "inefficient app" detection by CPA |
| Video Shield | Rubbish channels and video partners |
| Lead Shield | Stops spam reaching Google as a conversion (with Lead Rating) |
| Safe Sites + harvest campaign | The whitelist and controlled exploration |
Scenarios and Smart Mode
The Configuration holds scenarios (complete rules with their thresholds): each one decides what gets excluded and how aggressively. Smart Mode, per scenario, tunes the thresholds within validated ranges according to what it observes (at a configurable cadence, with raw data downloaded incrementally and, optionally, an AI analysis that proposes adjustments; if the AI fails, heuristics take over). The 1-5 aggressiveness selector and the retroactive setting (lesson 4) govern the whole thing. TEST mode as standard.
The Dashboard
A traffic light per module and, above all, fraud blocked in euros: the estimated spend that would have gone to the excluded placements, apps and IPs (using each one's historical spend as the basis), accumulated by month. It is the number that justifies the shield to whoever is paying, and the one that feeds the client panel ("€ of fraud blocked").
False positives: the human review
A tab holding the doubtful exclusions (a score in the grey zone, partial trust signals, a site a human might recognise as legitimate): each row with its score, its signals and a reinstate checkbox. Tick it and the shield withdraws the exclusion, adds the domain to Safe Sites and learns. The Agent watches this tab across every account and raises a warning when the doubtful cases pile up.
The client's routine
- Weekly (5 min): Dashboard (fraud blocked, modules in red), the false positives tab (reinstate where appropriate), the shield's email.
- Monthly: aggressiveness (raise or lower it, using the impact forecast?), Safe Sites, the click-through CPA of Display/PMax against the previous month.
- Never: touching the account exclusion lists by hand while the shield is managing them (two hands on the same list = conflict).
What the shield does not do
- It does not decide your Display/PMax strategy (which campaigns, what budget): it only cleans up where they appear.
- It does not replace measurement: Lead Shield needs Lead Rating (or some scoring) to exist.
- It does not guarantee zero fraud: it reduces the surface area and keeps it reduced.
💡 Ninja trick: the piece that adds the most value over time is the shared census: every domain scored once serves every client (farms repeat: the same network turns up in accounts from completely different sectors), and every new client starts with their inventory already scored. It is the advantage of a community over a farm, which is also a community — except ours learns faster.
What you should remember
- Deployment from the MCC: spreadsheet + loader + snippet + the analyser on our own server.
- Modules: Site Analyzer, Farms, Visitor/IP/App/Video/Lead Shield, Safe Sites.
- Scenarios + Smart Mode + aggressiveness 1-5, with TEST mode.
- A Dashboard with fraud blocked in €; false positives with a checkbox; a weekly 5-minute routine.
That is the end of Module 5. Module 6 moves from the traffic to the customer: lead quality with real data, and Lead Rating in depth.