
Envoy Media Group runs a lead generation operation that matches consumers facing financial challenges with vetted financial services partners. Each click in that funnel is bought with the company's own advertising budget and only pays off when it converts into a real enrollment for a client. That economics makes automated traffic an expensive problem rather than a nuisance.
Inside Revstr
Envoy's platform, Revstr, is an in-house multichannel system covering content management, marketing automation and business intelligence, built and maintained by an internal team on a multi-million line PHP codebase alongside its cloud services and infrastructure. It supports A/B testing of any combination of designs with any set of business rules, so each visitor can be shown a tailored experience that adapts to their responses. Revstr tracks every step through the company's pages and forms and integrates with ad platforms, client CRM systems and third-party marketing tools to build a 360° view of performance and customer experience. That data lands in a proprietary warehouse, which also feeds cleaned, normalized and labeled data to the machine learning pipeline where models can be tested, trained and deployed.
The cost of bad clicks
Traffic quality is a central challenge, because fake visits drain spending on advertising, UX optimization and customer service. The competitive logic of digital marketing is that whoever can afford to pay the most per click wins the auction, and bot traffic weakens Envoy's position in those auctions and its ability to buy. Media buyers on the business team monitor cost, quantity, engagement and conversion rates, looking for anomalies that signal fraudulent clicks.
Scoring requests instead of blocking them
Cloudflare assigns each request that reaches Envoy a bot score from 1 (automated) to 99 (human), produced through several methods including machine learning trained on the volume of traffic Cloudflare observes across millions of requests per second. Because a paid click is too costly to discard outright, Envoy does not block on a low score. It ingests the score as a custom header and feeds it into its rules engine, which can serve longer, more qualifying forms to suspicious visitors and streamlined ones to high scorers; in an extreme case, suspect visitors are asked to call by phone instead of filling out an online form. That preserves the chance to convert leads of uncertain origin while keeping the experience clean for the best ones.
The score also flows into the data warehouse for the marketing team. If a campaign or traffic source shows a low average bot score over time, Envoy can cut or eliminate spend on it, pursue refunds from providers and shift budget toward more profitable segments.
Feeding the models
Envoy uses ML to predict the lifetime value of new visitors from known data about past leads and passes that signal to its advertising vendors, skewing conversion values upward for leads with better predicted value so that pay-per-click platforms' smart bidding algorithms hunt for better-qualified traffic. One model in that process predicts backend conversion rate — the likelihood a lead becomes an enrollment for the client company. Adding the bot score and behavioral metrics produced a significant accuracy gain, and that gain carries through media management, form design, lead delivery integrations and email automation.
Measured results
Combining Cloudflare's bot score with internal data yielded several findings:
- Precision of the conversion rate prediction model rose 15% once bot score and behavioral metrics were added — a large improvement, given how hard even a 1% gain is in a carefully tuned model.
- Bot score now appears in 76 reports used by the media buying and UX optimization teams and in 9 ML models, and is a standard component of all new UX reports.
I use bot score in PPC search reports. Before I had access to the bot score our PPC reports were muddied with automated traffic - one day conversion rates are 11%, the next day they are at 5%. That is no way to run a business! I spent a lot of time investigating to understand and justify these differences - and many times there just wasn’t a satisfactory answer, and we had to throw the analysis out. Today I have access to bot score data, and it prevents data dilution and gives me a much higher degree of confidence in my analysis.



