Troubleshooting Persistent Inaccurate IP Geolocation Data for Specific ISPs on Our Web Tool

We've developed a "What is My Country?" web tool providing real-time IP geolocation, displaying user country and IP address. It's generally robust, but we're encountering a persistent issue.

The core problem lies in the inconsistent accuracy of IP geolocation data for users connecting through specific Tier-1 ISPs, particularly those with large-scale NAT deployments or anycast networks. For instance, users in country A are frequently misidentified as being in country B, even when their local ISP assigns them a geographically correct IP range.

Our current backend utilizes a combination of MaxMind GeoIP2 databases (both Country and City) updated weekly, supplemented by a custom caching layer and a secondary lookup via ipinfo.io as a fallback. We've implemented strict cache invalidation policies and cross-referenced with multiple public IP lookup services.

Attempts to resolve this have included:

• Adjusting MaxMind's confidence thresholds.
• Implementing a rolling average of location data for repeat IPs (which caused more issues than it solved due to dynamic IPs).
• Directly contacting some of the problematic ISPs (with limited success in getting actionable data).
• Analyzing traceroute data from affected users to identify routing anomalies, but this is not scalable for every user.

Despite these efforts, the inaccuracies persist for a statistically significant subset of our user base, impacting the tool's perceived reliability. We're looking for advanced strategies or less common solutions to improve our IP geolocation accuracy, especially concerning large ISP networks and their routing complexities. Are there any specific techniques for handling anycast IPs or better methods for correlating ISP-provided data with public geolocation databases?

Waiting for an expert reply.