From 90% to 99%: the economics of cascading geocoding
Every geocoding provider misses. Not the same addresses — that's the crucial part. An authoritative national source misses new developments not yet gazetted; a global commercial provider misses rural lot references the national source knows cold; open data misses where contributors haven't mapped. Industry experience with the cascading pattern puts single-provider success around 90%, rising to 98–99% with a two-to-three provider cascade — because different providers fail on different address types.
What a 9-point match-rate gap costs
At 250,000 lookups/month, nine percentage points is 22,500 addresses a month that either bounce to manual handling or — worse — enter your systems as locality-level guesses. In asset-heavy operations, each of those is a mis-dispatched crew, a mislocated inspection, or a data-quality ticket. Even a conservative remediation cost of a few dollars per record dwarfs the per-call provider fees everyone negotiates so hard over.
Cheap-first, premium-fallback
The cascade also inverts the cost curve. A well-ordered chain puts the authoritative or cheap source first, and the premium global source last:
AU chain: authoritative national source (flat licence)
→ mid-priced commercial provider
→ premium global provider (best global recall)
→ self-hosted open-data geocoder ($0, locality-level backstop)
Premium calls happen only for the residual the earlier sources missed — you pay premium prices for 2–10% of traffic instead of 100%. The routing engine also needs to understand that free tiers aren't comparable (one provider resets its free quota daily, another monthly, a third caps requests per second with no monthly ceiling) — a chain optimised purely on list price per 1,000 calls makes bad decisions.
The chain must survive vendor death, not just misses
Two real events inside the last 12 months: a UK address-lookup service was discontinued following a court decision, and a national address API in Europe retired its legacy endpoint. Any consumer integrated directly against either had an emergency re-platforming project. Consumers behind an aggregation layer had a config change. Provider continuity risk is not hypothetical, and a fallback chain designed only for "address not found" doesn't absorb it — the chain has to treat provider gone as a first-class cascade trigger.
One honest caveat
Aggregation is not novel IP — several vendors sell cascading geocoding. What varies is where the layer lives and what it's wired into. Our view: for asset-intensive and government estates, the value concentrates when the cascade is fused with authoritative-source licensing, zone/LGA enrichment, ERP field mapping, and an auditable ledger — one governed step inside the data pipeline you already run, rather than another external API to govern.