How authoritative latitude/longitude data — and a transparent coordinate provenance field — enables Geocoding at Scale turning raw UK property records into a spatially intelligent commercial real estate dataset.

Introduction

Geocoding at scale is the discipline of assigning accurate, attributable coordinates to millions of addresses — and in commercial property analytics, it is the difference between a usable dataset and a misleading one. This article explores how geocoding at scale works using AddressBase Premium coordinates, and how two deceptively simple fields — latitude and longitude combine with a rich commercial real estate dataset to create something far more meaningful than the sum of its parts.

The dataset underpinning this analysis is a UK-wide commercial real estate (CRE) view containing over 40 attributes per property record, spanning occupancy, ownership, valuation, business rates, floor data, and — crucially — precise geospatial coordinates. It is this combination that transforms a flat table of addresses into a spatially queryable intelligence layer. Geocoding at scale is the engine that makes that transformation possible.

The Underlying Data: What We’re Actually Working With

Before diving into the spatial mechanics, it is worth being explicit about the underlying dataset — because the credibility of any geocoding-at-scale analysis rests on the depth and structure of the records it operates on.

The dataset is a single, wide commercial real estate view (commercial_real_estate_creos01) in which each row represents one commercial property location, keyed by a unique reference (urn) and anchored to a real-world address via udprn, uprn, and full_address. It is not a thin coordinate table — it is a consolidated property intelligence record that stitches together several distinct data domains into one row per location:

• Identity & address — urn, udprn, uprn, full_address, postcode, dar/dars (address reference codes)
• Geospatial — latitude, longitude, lat_long_source, plus administrative geography (lsoa_2021, output_area_2021, local_authority, county, country)
• Physical form — number_of_floors, floor_levels, basement_present, location_size_m2, location_internals_m2
• Occupancy — occupant_name, occupant_company_number, occupation_date, occupant_sic_code_1/2, n_years_occupant_address, occupant_owner
• Ownership — commercial_owner, commercial_owner_number, commercial_owner_type, title_number
• Financial — rental_value, business_rates, average_cost_m2_rent
• Risk & compliance — companies_house_compliance, n_companies_registered_at_address, n_disqualified_directors, fsa_hygiene_score, gambling_license_type

Commentary on the data itself: the scale is substantial — the dataset holds well over 2.5 million commercial property records spanning the whole of the UK, from Inner London to Cornwall. What makes it analytically powerful is that it is pre-joined: rather than forcing an analyst to reconcile an address file, a valuation list, a business rates register, and a companies database, the dataset presents them as a single coherent record per location. The change_date field provides a temporal anchor, and the presence of both udprn and uprn means records can be cross-referenced against the wider UK address ecosystem. Crucially for this article, the coordinate fields sit alongside — not apart from — these attributes, which is exactly what allows spatial analysis to be enriched with commercial context rather than being a purely geometric exercise.

A note of caution on data quality: as with any large consolidated view, completeness varies by field. Financial and occupancy attributes are populated more densely for some property types than others, and — as we will see — coordinate provenance itself is mixed. Geocoding at scale is therefore not just about having coordinates, but about understanding the confidence and coverage of every field you build on.

The Three Pillars of Spatial Data

1. latitude — the north–south position

Latitude expresses how far north or south a point lies from the equator, measured in decimal degrees. In the dataset it is stored as a high-precision decimal(38,18) value, preserving sub-metre accuracy essential for distinguishing adjacent units in dense urban commercial districts — a core requirement of geocoding at scale.

2. longitude — the east–west position

Longitude expresses east–west position relative to the prime meridian, stored with the same decimal(38,18) precision. Together, latitude and longitude form the coordinate pair that anchors every property to a real-world point on the map.

3. lat_long_source — coordinate provenance

This is the field that too many analysts overlook. lat_long_source records where the coordinate came from — the provenance of the geocode as not all locations will appear with an exact lat/long it’s very useful to have proxies. Provenance matters because not all coordinates are created equal, and geocoding at scale only delivers value when that quality is transparent:

• A coordinate derived from AddressBase Premium (Ordnance Survey’s authoritative address database) carries a high degree of positional confidence. But not all locations have an exact lat/long.
• A coordinate inferred from a postcode centroid is accurate only to the postcode level — potentially hundreds of metres off in rural areas.
• A coordinate from a third-party or interpolated source may carry unknown error margins.

By exposing lat_long_source as a first-class field, the dataset lets analysts filter, weight, or flag records by geocoding quality — a best practice that prevents spatial analysis from silently inheriting positional error.

---

Provenance in Practice: What the Data Actually Shows

Geocoding at scale is only as trustworthy as its provenance mix — and the dataset reveals exactly that. Across the full commercial property view, coordinate provenance breaks down as follows:

lat_long_source	Property Count	Share	Interpretation
UPRN	2,108,153	~83.4%	Highest confidence — geocoded to the Unique Property Reference Number (building/unit level). Some locations may share a UPRN.
POSTCODE	369,495	~14.6%	Postcode-level centroid — accurate to street/area, not building
(null)	29,454	~1.2%	No provenance recorded — treat with caution
DERIVED	6,587	~0.3%	Inferred/interpolated coordinate
EXACT	5,973	~0.2%	Explicitly exact match

Key takeaway: Over 83% of records are UPRN-level geocoded — the gold standard for building-level spatial analysis. However, ~15% are postcode-centroid or derived, meaning any fine-grained radius or catchment analysis should filter to lat_long_source = 'UPRN' (or at minimum segment by provenance) to avoid positional error creeping into results. This is geocoding at scale done responsibly: not just assigning coordinates, but recording how each one was produced.

Why Provenance Is the Hidden Superpower

Consider two properties that appear identical in every attribute except lat_long_source. One is geocoded to rooftop-level precision from AddressBase Premium; the other is a postcode-centroid approximation. If you are:

• Mapping competitor density within a 200-metre radius,
• Calculating drive-time catchments for retail site selection,
• Joining to flood-risk or planning layers at building level,
• Running spatial clustering on high-street occupancy,

…then the provenance field is the difference between a defensible insight and a spurious one. Best practice: always segment or filter by lat_long_source before drawing spatial conclusions, and document the provenance mix in any downstream model.

Best Practices for Geocoding at Scale & Spatial Analysis

1. Filter by provenance first. Use lat_long_source to isolate rooftop-accurate records for fine-grained analysis, and treat centroid-derived records as lower-confidence.
2. Preserve precision. The decimal(38,18) storage supports sub-metre work — avoid rounding coordinates prematurely.
3. Validate coordinate ranges. UK coordinates should fall roughly within latitude 49–61 and longitude −8 to 2. Outliers signal data issues.
4. Join spatially, not just by postcode. Coordinates enable true distance-based joins to OS Open Data, flood zones, transport networks, and demographic layers.
5. Aggregate responsibly. When rolling up to LSOA, output area, or local authority (all present in the dataset), weight by coordinate confidence.
6. Audit the provenance mix. Report the proportion of records by lat_long_source in any published analysis — the mark of mature geocoding at scale.

Spatial Validation: Coordinates That Behave

A quick sanity check confirms the coordinates cluster exactly where UK commercial property should. Average coordinate centroids by local authority (top 15 by volume):

Local Authority	Properties	Avg Latitude	Avg Longitude
Inner London	183,927	51.51247	−0.10971
Outer London	128,447	51.50695	−0.14568
Greater Manchester	113,594	53.49647	−2.27485
West Midlands	99,717	52.49361	−1.91257
West Yorkshire	90,027	53.75550	−1.65781
Kent	54,924	51.26587	0.77012
Essex	51,571	51.75866	0.61217
Lancashire	48,520	53.79610	−2.60309
South Yorkshire	45,895	53.45510	−1.36255
Merseyside	44,381	53.43393	−2.94449
Hampshire	43,210	51.01443	−1.22042
Devon	41,335	50.72968	−3.70563
Norfolk	40,677	52.68552	1.13341
Tyne and Wear	37,452	54.96231	−1.54430
Cornwall	36,991	50.33871	−4.97869

Observations:

• All centroids fall cleanly within valid UK ranges (lat 50–55, lon −5 to +1.1) — confirming coordinate integrity across the geocoding at scale pipeline.
• Inner + Outer London dominate with ~312k properties combined, followed by the major northern and Midlands conurbations — consistent with commercial density.
• The centroids trace the UK’s geography correctly: Cornwall far south-west (50.34, −4.98), Tyne and Wear far north-east (54.96, −1.54), Norfolk east (1.13 lon) — a strong validation that the latitude/longitude fields are trustworthy.

Where the Dataset Creates Compounding Value

Coordinates alone are useful. Coordinates combined with the rest of this dataset become a genuine analytical engine. The same record that carries latitude and longitude also carries:

• Occupancy intelligence — occupant_name, occupant_sic_code_1/2, occupation_date, n_years_occupant_address
• Ownership structure — commercial_owner, commercial_owner_type, title_number
• Financial signals — rental_value, business_rates, average_cost_m2_rent, estimated_location_turnover
• Physical characteristics — number_of_floors, location_size_m2, basement_present, floor_levels
• Compliance & risk — companies_house_compliance, n_disqualified_directors, fsa_hygiene_score
• Administrative geography — lsoa_2021, output_area_2021, local_authority, county

This is the compounding effect: geocoding at scale is not an end in itself — it is the connective tissue that lets you ask questions like:

• “Which SIC sectors cluster within 500m of each other in this town centre, and what is their average rental value per m²?”
• “Where do high-turnover retail locations coincide with low hygiene scores and high business rates?”
• “Which commercial owners hold the most floor area across a local authority, mapped by coordinate?”

None of these are answerable from coordinates alone, nor from attributes alone. They require the spatial anchor plus the rich attribute layer — which is precisely what this dataset delivers.

Conclusion

Geocoding at scale is fundamentally about trust: trusting that a coordinate points where it claims to. By treating latitude as north–south position, longitude as east–west position, and lat_long_source as coordinate provenance, analysts can build spatial models that are both precise and honest about their uncertainty. Combined with the deep commercial, financial, and compliance attributes in this dataset — and a clear-eyed understanding of its coverage and quality — authoritative coordinates become the foundation for commercial property analysis that is genuinely decision-grade.