Fighting Noise In Email Tracking

Email 'Opens' Are Broken 💔

Email engagement is a crucial signal in every business’ outreach pipeline. Email opens & clicks help drive and trigger automated workflows. They often determine how sales teams prioritise leads & optimise their workflows further. But recently, open rates have become increasingly unreliable—and the root cause lies in how aggressively mailbox providers scan and filter emails.

From Google to Microsoft and Apple, providers are deploying automated proxies to open emails for spam detection, antivirus scanning, or content analysis. This behaviour by bot servers leads to messy, inflated open rates that paint a distorted picture.

Teams keep chasing ghosts, wasting time, and often getting frustrated when real engagement is nowhere to be found.

The Customer Frustration Is Real

Apollo users began noticing mass opens within seconds of sending emails, often from the same locations. These were clear signs of bot behaviour which inflated open rates and created confusion around what was actually happening.

One of our users raised this:

On top of that, noisy notifications became a pain point. High-volume senders were getting flooded with open alerts, many of which were triggered by bots, making it hard to trust which ones were real.

The Solution Journey

Stage 1: Advanced Bot Detection

We started by building sophisticated detection systems for automated opens. Our goal was to separate human behaviour from bot behaviour with high precision.

We leaned into known patterns:

• Google loads & caches tracking pixels in emails via their proxy servers, sometimes instantly. The pixels are served from cache through Google Proxy IP whenever Gmail user opens emails.
• Microsoft takes it further—emails can be opened by their filters even if the user never opens them. They have a wide range of proxy IPs & keep on rotating.
• Apple’s Mail Privacy Protection takes it a notch further - it might fetch content minutes, hours or even days after delivery, and uses generic user-agents!

We designed a service that processes incoming open and click events, runs them through several modules to categorise each event as either bot or human activity, and then persists the events along with the categorisation data. The categorisation modules do the following in 4 broad steps:

1. Analyse Timing and Frequency of Opens: We analyse the timing of each open event in relation to when the email was sent. Opens that occur almost immediately after delivery are flagged as potential bot activity, since they often indicate automated scanning by proxy servers. Additionally, we monitor the frequency and deduplicate repeated opens received in quick succession for the same email.
2. Map IPs, Hostnames, and User Agents to Known Proxy Servers: We identify automated proxy-based activity by mapping IP addresses, resolved hostnames, and user agents to known proxy servers from providers like Google, Microsoft, Apple, and others. Our pool of network signatures is continuously updated using publicly available data (including from Email Service Providers) and patterns observed in our internal event data. We've gone through multiple iterations to refine this proxy server detection process and continue to update it regularly.
3. Compare Events from Google Proxy and Apollo Chrome Extension: Our system tracks and compares events generated via the Apollo Chrome Extension with those coming through the Google Proxy. This helps us determine whether an open is likely from a bot, a self-open by the user, or a genuine open by a prospect.
4. Compare Network Signatures to Identify Self-Opens: We compare the network signatures of open events with those of the “original” senders to detect and flag self-opens. These self-opens, which occur when a user previews the email they sent, are not counted as genuine opens and are excluded from reporting.

With this service in place, bot opens were excluded from being surfaced to users or impacting their workflows. This gave us a significant edge over competitors who either don’t exclude bot activity at all or have only minimal filtering in place. But with great accuracy came a side effect...

Stage 2: Bridging The Perception Problem

Because our bot detection worked so well, our open rates appeared lower than those of our competitors. It wasn’t lower engagement, just less noise.

The downside? It looked like our performance was worse.

We realised we had solved the “noise” problem, but introduced a new challenge: “transparency.”

So we expanded our approach. While we continued to exclude bot events from impacting users’ workflows, we began surfacing both filtered and unfiltered (including bot events) open rates. We aggregated both sets of stats at the campaign level and gave users the ability to toggle between views for easy comparison.



This shift brought clarity and significantly improved users’ confidence in Apollo’s email engagement data.

Impact & Learnings

• With advanced bot detection in place, Apollo now delivers reliable email engagement tracking. Approximately 55–60% of all open events are triggered by bots, and we're able to filter these out to minimise disruption in our users’ workflows.
• Users now have the ability to compare filtered and unfiltered rates for full transparency. Approximately, 9.9% of email users toggle between open rate views every month.

1. Mailbox providers are constantly evolving

Tracking email engagement is an ever-evolving challenge. Google, Microsoft, Apple—all have their own quirks and patterns. A lot of complexity lies in recognising these patterns, and we have to continuously iterate to stay relevant. What works today might break tomorrow.

2. Transparency is king

Even the most advanced filtering means nothing if users don’t know what’s happening. That’s why we surface both filtered and unfiltered data, giving users full visibility and the ability to decide what matters most for their workflows.

Final Thoughts

Sales teams rely heavily on email engagement, but distinguishing real intent from noise is crucial. By eliminating the noise and focusing on transparency, we’ve improved how users analyse and act on email engagement data.

Our email tracking system helps users focus on meaningful interactions and reduce confusion from unreliable signals. As email ecosystems are continuously evolving, we'll keep iterating and refining our approach to ensure users have clear, trustworthy insights to support their workflows.

🚀 .....and we are looking for you!

Problem statements like these are in plenty at Apollo, and we are betting big on AI for building products for our customers. Our engineering team thrives on solving complex problems, pushing the boundaries of what’s possible with data, and delivering cutting-edge solutions that drive "impact".

We are looking for smart engineers like you to join our "fully remote, globally distributed" team. Click here to apply now!