Skip to content
DocsSearch the docsEnglishBack to siteOpen console

Sensitivity and false positives

When you believe a verdict is wrong you have three levers: the Allowlist, which clears an address or device at once; a false-positive report, which turns a slice's wrong flag into a label for retraining; and the sensitivity dial, which moves the bar for nightly scoring. Writing detection rules is not one of them — deliberately.

The levers you have#

In Fraud Protection, AdPix does the detecting and you cannot write detection rules. You do have three ways to say a verdict is wrong, or that the bar should be stricter or looser:

  • Allowlist — the fastest fix for a wrong verdict on a click. From the Click evidence drawer on the Clicks tab, or from the Allowlist card on the Settings tab.
  • Report false positive — for a traffic slice's verdict from the nightly analysis. In the Flagged entities table on the Sources tab, and in each entity's evidence drawer.
  • Sensitivity dial — at the bottom of the Settings tab. It decides how much evidence the nightly scoring of slices needs before it flags.

The same Settings tab also holds the protection mode, industry preset and thresholds of the real-time click check — those are covered in protection settings.

AdPix detection rules are global and affect the whole network, which is why authoring them belongs to the platform team rather than to businesses. In exchange, every label you give goes straight into how the model's precision is measured.

The fast fix for a click's verdict: the Allowlist#

Every paid click gets a verdict the moment it lands, unless protection is off. If that verdict is wrong for traffic you know — your office, your testers, a partner — do not wait for any review:

1
On the Clicks tab, select the click's time to open the Click evidence drawer.
2
Select "Not fraud: allowlist this IP" or "Not fraud: allowlist this device". The IP button is there only while the click's raw IP is still kept (90 days), and the device button only when the device was identified.

To add one by hand, go to the Allowlist card on the Settings tab, set Type to IP address, IP range (CIDR) or Device, fill in Identifier and Reason, and select Add to allowlist. Remove takes a row off the list.

An address, range or device on this list gets a Clean verdict from then on, with the reason Allowlisted, and never enters blocks or the exclusion lists. The real-time check picks it up within a minute. Adding and removing needs edit access to the website and is written to the audit log.

The Allowlist clears things silently

Anything you add here is never flagged again, even if it genuinely turns into a fraud source one day. A wide CIDR range means every address in it. Add only what you actually know, and write the reason down, so that six months from now someone can decide whether to keep it.

The sensitivity dial#

The Sensitivity dial card sits at the bottom of the Settings tab, below the Allowlist. It has three settings and is stored against the selected website. Changing it needs edit access and is written to the audit log.

Setting What it does Who it suits
Conservative only the top 10% of your traffic by anomaly score is a flag candidate, and at least two detectors must agree the default. When a false positive costs you more than missing a real case
Balanced opens the candidate set to the top 20%, still on two-detector agreement large media buys spread across many channels
Aggressive opens it to the top 35% and accepts a single detector's agreement when you know you are under attack and would rather see more cases

A website that has never been set is scored Conservative, and the control opens on that setting, even though the Balanced hint begins with "Default". What decides is what you saved.

1
On the Fraud Protection page, open the Settings tab and scroll to the Sensitivity dial card.
2
Choose one of the three settings.
3
Select Save sensitivity. The current value is always printed beside the page title, after Sensitivity.

What the dial does not change#

Take this part seriously, because it is where the misunderstanding lives. The dial does not retrain the model and does not touch data that has already been scored; it moves the bar for the next run. Beyond that:

  • The real-time verdict on clicks has nothing to do with it. The protection mode, industry preset and thresholds on the same tab tune that.
  • Structural signatures are independent of the dial. Fingerprint rings, data-centre traffic, click flooding, IP rotation and dwell-variance collapse flag on all three settings. The dial only affects the "model opinion" path.
  • No setting removes the precision guards. A slice that converts, a slice with engaged sessions and a slice under three sessions are not auto-flagged even on Aggressive.
  • What is already clean stays clean. Payment gateways, hosts you excluded yourself and your own domain and brand never receive a fraud score on any setting.

So going Aggressive does not open the guards — it only lowers the amount of evidence an entity needs to leave the Clean tier.

When AdPix declines to decide: the Review tier#

Between "flag it" and "let it go" there is a third tier, and it is intentional. If an entity's score lands just under the bar — and it does not convert, does carry at least one corroborating behavioural signal, and has defensible volume — AdPix does not guess. It routes the case to Review, where a human on the platform team picks it up.

Several other situations go straight to review:

  • A slice that both converts and carries a structural fraud signature. Conversion never overrides a structural signature — otherwise any ring could whitelist itself by faking a few key events — but it is never auto-flagged either.
  • A slice with sound evidence and too little volume. Under three sessions nothing exceeds review.
  • A slice whose sessions count as engaged, where the only complaint is statistical outlierness.

Sessions in this tier sit in the invalid rate's denominator only and are never counted as invalid. That distinction matters: the numerator of your invalid rate contains only positive claims, not the cases still undecided.

Reporting a false positive#

If you recognise a slice and know the traffic is real — your own QA bot, an agency verifying placements from a shared IP, a known partner — say so.

1
On the Sources tab, find the slice in the Flagged entities table, or reach its evidence drawer from the breakdown table at the top of the same tab.
2
Select Report FP at the end of the row in the Flagged entities table; the evidence drawer carries the same action as Report false positive.
3
Under "Why do you believe this is legitimate?", explain. The more specific the better — tell the reviewer what they cannot know and you do.
4
Select Submit report.

Your explanation is stored alongside the entity ID; the text is not discarded and it is what the reviewer reads.

A false-positive report is not an immediate exemption

The report does not change scoring on the spot and does not guarantee the slice will never be flagged again. If you need an immediate, guaranteed exemption, the right tool depends on the case: for a specific IP address, range or device, the Allowlist; for return traffic from a payment gateway, referral exclusions.

What happens to the report#

Your report creates a legit label. Labels are the only human ground truth in this system, and four things are done with them.

Stage What happens
Human review a reviewer in the platform console reads the report and labels the same entity fraud or legit
Supervised member once enough labels of both kinds exist, a small supervised model is fitted on them and its opinion is blended into the final score at a fixed weight
Precision measurement those same labels are the yardstick for every run's precision — the share of flags that were right
Promotion gate a run is only adopted as the new baseline if its precision is at least that of the last adopted run

That last row is the backbone of the design. Because the model trains on network-wide data, any change can spill into other businesses; the promotion gate guarantees that a version which lowers precision never becomes the baseline. The pleasant side effect for you: every accurate report you file lowers the chance of the same mistake for everyone.

One more guard runs behind the scenes: if an active detection rule starts catching sources that convert or that carry a legit label, it is automatically demoted to shadow mode and stops flagging anything.

Timing#

The Allowlist reaches the real-time check within a minute. Nightly scoring runs roughly once a day by default; a sensitivity change, a new label and a new rule all take effect from the next run. If you check the page immediately after a change and nothing moved, nothing is broken — the run has not happened yet.

Before you touch the dial at all, it is worth opening a few current flags and reading their evidence. Why was this flagged explains what each signal means — and usually that alone tells you whether the problem is sensitivity or a missing exclusion.

Frequently asked questions#

Why can't I write my own detection rules?

Because detection rules in AdPix are global and affect every business on the network, so the platform team owns them. What you own is the Allowlist, the protection mode and thresholds on the Settings tab, the sensitivity dial, false-positive reports and your own website's referral exclusions. Together those cover nearly every real case.

How long until a sensitivity change shows up?

Until the next scoring run. The dial does not touch data that has already been scored; it moves the bar for next time. The automatic run happens roughly once a day by default.

A click from our own office came back Confirmed. What is the fastest fix?

On the Clicks tab, select that click's time and, in the Click evidence drawer, choose 'Not fraud: allowlist this IP' or 'Not fraud: allowlist this device'. That address or device is Clean from then on and is never blocked, and the real-time check picks it up within a minute. For a whole IP range, add it on the Allowlist card of the Settings tab.

If I report a false positive, does that source go clean immediately?

No. Your report records a legit label that the platform team reviews and that feeds retraining; scoring does not change on the spot. If you need an immediate exemption, use the Allowlist for an address or device, and referral exclusions for return traffic from a payment gateway.

Build with the APIUnderstand where revenue comes from.
Was this page helpful?