Axad Capital Triples Clean Traffic With Anura and Everflow Stack

Axad Capital says it moved the share of high-quality clicks reaching advertiser landing pages from 31.8% to 89.6% in two months, a 2.8x lift. The headline number is the lift. The more interesting number is the baseline: for a self-described technology-driven performance marketing company running Insurance, Financial Services, and Legal traffic, roughly 68% of inbound clicks were not clean before the new stack went in.

That baseline is what senior performance and platform teams should stare at, because Axad is not a fringe operator. It runs millions of high-intent interactions a month across three of the most fraud-targeted verticals in performance marketing.

What Happened

On September 16, 2026, Anura announced the results of a joint deployment with Axad Capital and Everflow out of Middletown, Delaware. As Morningstar carried it, the two-month program processed more than 1 million inbound clicks, prevented more than 670,000 low-value or invalid clicks from reaching advertisers, and reduced bad traffic by 58 percentage points.

The architecture is straightforward. Anura, an ad fraud detection vendor that markets a 99.999% accuracy guarantee and holds a TAG Certified Against Fraud designation, sits in front of the click stream and classifies traffic in real time. Everflow, an attribution and partner tracking platform that reports 1,200+ brand clients including Playtika, JG Wentworth, and Evite, handles routing, attribution, and partner accounting. Every click gets one of three verdicts: good traffic is passed to advertiser landing pages, bad traffic is stripped out before it touches an advertiser, and warning traffic is held for additional qualification.

Krishna Tripathi, CTO of Axad Capital, framed it in the release: "Our goal has never been to generate more traffic; it has been to generate better traffic." He added that the pairing let the team "build a customer acquisition engine where quality comes first."

What the announcement does not disclose, and what matters for anyone trying to replicate the result: the mix of media sources feeding those 1 million clicks (paid search versus paid social versus native versus affiliate partners), the pre-filter cost per click, and how much of the "bad" bucket was outright bot traffic versus warmed-up human fraud. The bound is knowable: if the 670,000 filtered clicks carried an average paid CPC of even 50 cents, the wasted-spend exposure was material. We do not have that CPC in the source.

Technical Anatomy

The design pattern here is a pre-attribution fraud gate. It matters because most performance stacks put fraud detection after attribution, which means invalid traffic still consumes budget, still fires conversion pixels, and still poisons downstream optimization signals fed back to Google Ads and Meta. By the time a fraud vendor flags the click, the algorithm has already learned that a bot-shaped user is a good prospect.

Putting Anura in front of Everflow flips that order. Clicks are scored before they are routed to an advertiser landing page, which means three things technically. First, the conversion pixel on the landing page never fires for filtered traffic, so the ad platform's optimizer is not trained on garbage. Second, the partner or publisher generating the click is scored in near real time, which changes how Everflow can weight future routing decisions from that source. Third, warning traffic gets a second pass rather than a binary allow or block, which is the classification tier most operators skip and the one that usually contains the highest-value edge cases.

The three-bucket model (good, bad, warning) is not new, but running it at pre-landing-page latency across 1 million clicks in a reporting window is a non-trivial engineering constraint. Anura has to return a verdict fast enough that the click redirect chain does not visibly degrade user experience, typically inside a few hundred milliseconds end to end when you include Everflow's own tracking hop. The source does not disclose the observed p95 latency of the combined pipeline, which is the single most useful engineering number that got left out. For anyone evaluating this stack, that is the question to ask on the demo call.

Everflow's integration surface (Google Ads, Meta, HubSpot, Salesforce, plus Everflow Pay for global payouts) matters because the fraud verdict has to propagate. A filtered click that still triggers a partner payout defeats the purpose. The value of the two-vendor pairing is that the classification signal and the payout ledger live in the same system.

Who Gets Burned

Three groups should be reading this announcement carefully.

The first is any performance marketing network operating in high-payout verticals: Insurance, Financial Services, Legal, plus the adjacent debt, mortgage, and health insurance categories. Axad's pre-deployment baseline of 31.8% clean traffic is a data point, not a universal truth, but it sets a rough floor for what sophisticated invalid traffic looks like in verticals where a qualified lead can be worth hundreds of dollars. If your internal quality score is materially above that without a comparable filtering stack in place, either your measurement is optimistic or your traffic sources are unusually clean. Both deserve scrutiny.

The second is the affiliate and publisher partners feeding those networks. A pre-attribution fraud gate makes partner performance visible in a way post-hoc reconciliation never did. Partners who were coasting on a mix of legitimate and invalid clicks will see their effective payout drop, and the ones who cannot adapt will be culled. Expect friction, expect renegotiation, and expect some partners to route their weaker inventory to networks that have not deployed comparable filtering.

The third is advertisers directly buying from performance networks. The Axad announcement is essentially a signal to the buy side: ask your networks what percentage of delivered clicks pass a real-time fraud gate, and ask what happens to warning-tier traffic. If the answer is vague, or if the network cannot produce a click-level classification breakdown, that is diligence you should be doing before the next quarterly budget.

What we do not know from the source is how Axad's advertiser CPA moved after the filter went in. A 2.8x lift in clean traffic share is not the same as a 2.8x lift in conversions or a proportional drop in CPA. The testable bound: if the filter is doing what the release implies, advertiser-reported CPA on Axad traffic should compress within one to two quarters, and partner concentration should shift toward a smaller set of higher-quality sources.

Playbook for Performance Marketing

Concrete moves for the next two weeks, whether you run a network, a publisher, or an advertiser buying performance inventory.

Instrument before you filter. Before signing any fraud vendor, log a week of raw click data with source, partner, IP, user agent, and downstream conversion outcome. You need a pre-filter baseline to measure lift against, otherwise you are buying a black box. Axad's 31.8% baseline is only meaningful because they measured it.

Separate the three buckets in your own reporting. Good, bad, and warning are not just vendor categories, they are business categories. Warning-tier traffic is where the interesting margin lives, because it is the population most likely to contain both your best edge-case customers and your most sophisticated fraud. Treating it as a distinct cohort with its own qualification step is the part of the Axad design most teams will skip.

Push the fraud verdict upstream of the ad platform conversion pixel. If your conversion API is firing on filtered clicks, you are training Meta's and Google's optimizers on noise. This is a plumbing fix, not a strategy question.

Audit partner payout logic against fraud verdicts. If a click is filtered but a payout still fires because the partner tracking system and the fraud detection system do not share state, you have a leak. Everflow's pitch works precisely because those two systems are unified.

Key Takeaways

• The Axad baseline of 31.8% clean traffic is the number worth remembering. The 2.8x lift is a function of how bad the starting point was.
• Pre-attribution fraud gating changes what the ad platform optimizer learns from, which compounds over time in a way post-hoc filtering does not.
• The source does not disclose latency, CPC of filtered clicks, or advertiser CPA movement. Those are the three numbers to demand in any vendor evaluation.
• Partners feeding mixed-quality traffic to performance networks should expect payout compression as more networks deploy comparable stacks.
• Testable prediction: within two quarters, expect at least one large performance network in Insurance or Legal to publish a similar case study, and expect advertiser RFPs in these verticals to start requiring click-level fraud classification reporting as table stakes.