Unboxing the Last Mile: Introducing Last Mile Insights

The Internet's Final Hop Deserves Better Visibility

Network engineers often say the last 20% of a connection requires 80% of the effort. The Last Mile — the link between end users and their ISPs — is the most heterogeneous and least controlled segment of the Internet path. Every last-mile network operates independently, yet all of them must function for the Internet to work as expected. Organizations running mission-critical services typically leave the health of these networks entirely in the hands of ISPs.

Cloudflare has introduced Last Mile Insights, a closed beta tool available in the Cloudflare Dashboard under the Analytics tab (Edge Reachability). It gives customers visibility into traffic that failed to reach Cloudflare, including where the failure occurred and why. Enterprise customers can sign up starting today.

Three Legs of the Internet Path

The journey between a user and content is generally described in three segments:

• The First Mile — path from the origin server to the requested data
• The Middle Mile — path from origin to proxies and intermediate network hops
• The Last Mile — final hop from the ISP to the end user

Cloudflare's architecture treats request life in two legs: end users to Cloudflare, and Cloudflare to origin. But the real-world path involves far more variable components. Last Mile issues are inherently difficult to detect: when users cannot reach a resource, the resource never learns about the problem. Multiply this across hundreds of thousands of independently operated ISPs and the detection problem becomes formidable.

How Detection Works

Last Mile monitoring combines Network Error Logging (NEL) signals with machine learning to identify errors and performance regressions. When a browser connects to a site with NEL enabled, Cloudflare passes headers instructing the browser to report any future connection failures to a specified endpoint. Reports arrive in real time after the browser experiences the failure.

The browser can report failures for various reasons: incorrect TLS certificates, ISP or transit issues along the request path, overloaded terminating servers, or unreachable data centers. The reports follow a W3C-specified structure that identifies the failure phase, reason (such as tcp.timed_out), ASN, and metro area. NEL does not capture personally identifiable information.

Cloudflare operates its own NEL reporting endpoint and is now opening that collected data to customers. This capability proved valuable in real incidents — for example, when Canva users connecting through Cox Communications in San Diego experienced connectivity issues. Cox was routing traffic destined for Canva incorrectly, affecting roughly 50% of mutual Cox/Canva users. Cloudflare's monitoring detected the anomaly within minutes, allowing rapid diagnosis and remediation despite the problem living entirely on Cox's network.

Proximity as Performance Strategy

Minimizing time on Last Mile networks is the most effective performance lever, since these networks can be unreliable. Cloudflare continues expanding into new cities and markets to reduce physical distance to users — nobody can outrun the speed of light, and distance directly affects latency.

Network health is another critical variable. Healthy networks show no downtime, minimal congestion, and low packet loss — conditions that keep latency down. Data from three comparable ISPs illustrates this: ISP A (healthy, directly connected to Cloudflare) and ISP B (healthy, not directly connected) perform similarly, while ISP C (unhealthy) shows latencies roughly 360% higher.

Expanding infrastructure shrinks the Last Mile's impact. After adding locations in Africa, latency for the Internet-connected population there dropped significantly. Cloudflare's efforts have brought 95% of that population within 50ms and 80% within 30ms of its network.

Connecting the Final Hop: Cloudflare Opens Last Mile Visibility

The Pareto Principle holds in networking as much as anywhere: the last stretch of the connection often demands the most work. Last Mile networks — the link between users and their ISPs — are heterogeneous, independently operated, and largely outside the control of any single organization. Yet for a service to perform reliably, every one of these networks must function. ISPs typically own this segment, but organizations running mission-critical Internet services that ignore Last Mile conditions effectively delegate their uptime and performance to third-party carriers.

Cloudflare's role as a proxy means it sits between end users and origins, so last-mile health matters deeply to its own service quality. The company is now announcing Last Mile Insights, a closed beta reporting tool available to enterprise customers in the Cloudflare Dashboard under the Analytics tab, in the Edge Reachability section. The tool shows customers traffic that failed to connect to Cloudflare, where the failures occurred, and why — surfacing connection problems that have historically been invisible because failed requests never reach the resource."

Three Segments of the Internet Path

When users connect to content online, the path is commonly described in three Legs:

• The First Mile — path from the origin server to the requested data
• The Middle Mile — path from origin to proxies and other network hops
• The Last Mile — the final connection from ISP to end user

Cloudflare's anycast architecture handles requests in two phases: users to Cloudflare, and Cloudflare to origin. But the broader Internet path involves more components than that simplified view implies. The real challenge is that when users cannot reach a service, the service never learns about the problem. With hundreds of thousands of ISPs across diverse regions, tracking all the ways the connection can break is a substantial observability problem.

How Last Mile Insights Works

Cloudflare's monitoring tools combine machine learning with browser-based Network Error Logging (NEL). When a user's browser successfully connects to a site that has NEL enabled, Cloudflare returns headers telling the browser to report future network failures to a specified endpoint. These reports are sent in real time the moment a browser encounters a connection problem.

Browsers categorize failures by cause per the W3C specification — for example:

• TLS certificate errors
• ISP or upstream transit issues along the request path
• Overloaded servers dropping requests
• Unreachable data centers

A sample report contains a JSON blob with details such as the failure phase (e.g., tcp.timed_out), the ASN the request originated from, and the metro area. This gives network engineers a view of where problems occur and why. NEL reports do not contain personally identifiable information, as documented in Cloudflare's knowledge base article on NEL.

Services can operate their own NEL endpoints, but Cloudflare also runs one and is now sharing the collected data with customers. This capability proved useful in a recent incident: Canva customers connecting through Cox Communications in San Diego experienced connectivity failures, with roughly half of Cox users unable to reach Canva or Cloudflare. The cause was Cox routing traffic destined for Canva incorrectly. Cloudflare's monitoring identified the issue within minutes, enabling early mitigation discussions with Cox and quick customer communication — a scenario that would typically have taken hours to diagnose, since Canva's own traffic metrics wouldn't have triggered alerts for an outage not on their side.

Proximity and Network Health

Beyond visibility, Cloudflare emphasizes physical proximity: minimizing time spent on last-mile networks improves performance, since these connections can be slow and unreliable. Cloudflare continually expands its presence into new cities and markets to get closer to users. Distance matters because network latency cannot exceed the speed of light, but distance alone does not determine performance — each ISP's network conditions also play a role. Healthy networks have minimal downtime, congestion, and packet loss; these factors add latency.

Data comparing three ISPs in a single country illustrates the point: ISP A (healthy, directly connected to Cloudflare) and ISP B (healthy, not directly connected) both show similar latency distributions. ISP C (unhealthy) shows significantly worse — roughly 360% higher — latency to Cloudflare.

Adding data centers in underserved regions reduces this effect. Cloudflare's expansion efforts have brought 95% of the Internet-connected population within 50ms of its network, and 80% within 30ms. Every additional location shortens the distance to users, particularly for the long tail of last-mile connections.

What Cloudflare Offers

Most cloud services optimize segments they can control, starting from the first mile. The last mile is harder to improve because it involves thousands of independent ISPs. Cloudflare is targeting performance gaps across the whole path, including the segments near end users.

Last Mile Insights provides the observability layer: identifying issues on networks outside a customer's control and reducing time-to-resolution. The ability to see the full picture helps engineers communicate with end users during incidents — even when the fault is on a third party's infrastructure. For customers, Last Mile Insights means the ability to see what is actually happening on the Internet, while Cloudflare works on infrastructure expansion to make the final hop shorter and more reliable.