The "address" column in a customer, collateral, or ATM table looks like location data. It is not. A computer cannot calculate anything from the text "Jl. Melati No. 12, RT 03". Distance, radius, nearest branch, flood-prone areas: all of it needs a point on the surface of the earth, not a sentence.

Our view is simple. Data only deserves to be called location data when three things travel with it: the coordinates, the reference system those coordinates use, and a record of how confident we are in the point. An address string without those three is just a label.

This article explains, from zero, how geospatial data works behind the scenes. It is written for IT and operations staff who handle data every day but have never had to deal with maps.

What geospatial data is

Geospatial data is ordinary data with one addition: a position on the earth's surface. "Customer A" is ordinary data. "Customer A at latitude -6.2, longitude 106.8" is geospatial data. Those two numbers are what a machine uses to measure distance, find the nearest item, or check whether a point falls inside an area.

That sounds easy. The difficulty sits in four steps that run quietly behind every map.

Step 1: turning an address into a point

The process is called geocoding. The system first cleans the address text, for example treating "Jl", "Jalan", and "Jln" as the same word, then matches it against a reference dataset of building points, street lines, or boundaries. The Wikipedia article on address geocoding explains that when only a street segment matches, the system estimates a position by interpolation: number 742 on a 700 to 799 block is placed slightly short of the middle of the block.

Interpolation assumes houses are spread evenly along the street, and the same source admits that this is almost never true in real life. As a result, points in a single table can differ wildly in quality. From most to least precise: building or parcel points, interpolation along a street, street segment centres, postal code centres, administrative area centres. If the method is not recorded, they all look the same on the map: one blue dot.

Five geocoding steps from address text to a stored point carrying its method and confidence

Step 2: the digits after the decimal point

Coordinates are written in decimal degrees, and every digit after the point sets how fine the position is. According to the table in Wikipedia's article on decimal degrees, at the equator three decimal places equal 111 metres, four equal 11.1 metres, and five equal 1.11 metres.

Compare that with the source. GPS.gov states that GPS-enabled smartphones are typically accurate to within a 4.9 metre radius under open sky, and that accuracy worsens near buildings, bridges, and trees. So five decimals is already enough for a phone-sourced point, and the eighth digit is decoration. Precision is not accuracy: a point with eight decimals looks millimetre-exact, yet it came from an officer's phone outside a shopfront.

The consequence is practical. A rule such as "a visit counts as valid if it is within 5 metres of the collateral point" will reject visits that really happened. A realistic tolerance comes from the accuracy of the point's source, not from how many digits are stored.

Comparison of three, four, and five decimal places against the 4.9 metre accuracy of smartphone GPS

Step 3: the coordinate reference system

Coordinates without a reference system are like numbers without units. In a spatial database that system is recorded as an SRID. The PostGIS documentation states that functions comparing or combining two geometries require them to share the same SRID, and that `ST_Transform` converts between systems. For geography columns, if no SRID is given, the default is 4326 (WGS84).

Indonesia has its own system, SRGI2013. The EPSG registry lists a transformation from SRGI2013 to SRGI2013 epoch 2021.0 that uses the INADEFORM deformation model to move coordinates from epoch 2012.0 to 2021.0. In other words, the official reference is tied to time: the same coordinates can mean a slightly different position depending on the reference year. We are not saying the difference is large for a branch map. We are saying it has to be recorded. Keep the SRID in the column next to the coordinates, and for high-precision survey data, keep the epoch too.

Step 4: the spatial index

"Find every customer within 2 km of this branch" without an index means the database inspects every row. Per the same PostGIS documentation, the spatial index is an R-Tree built on GiST that stores the bounding box of each geometry. Queries run in two passes: a primary filter uses bounding boxes to narrow the candidates, then a secondary filter tests the exact condition. The database does not compute distance millions of times, only for the candidates that remain.

If a radius query feels slow, check whether the location column has an index before adding servers.

Five layers behind a map, from point sources and reference systems to storage, indexing, analysis, and reports

One map, not ten

Indonesia has tackled this problem nationally through the One Map Policy, which according to Diskominfo Central Java (23 February 2018) is a mandate of Law Number 4 of 2011 on Geospatial Information. Its goal has four parts: one geospatial reference, one standard, one database, and one geoportal.

The same pattern applies at small scale inside a bank. The credit team has its own location sheet, collections has another, the network team a third. One collateral item can end up with three points and nobody knows which is right. A single location base with method and accuracy recorded can serve all of them.

One location base at the centre used by field verification, the branch and ATM network, collections, and risk concentration

Five questions before touching a map

  1. Where did each point come from: geocoding, officer GPS, or manual entry?
  2. Is there a column recording the method and the accuracy level?
  3. Which reference system is stored, and is it consistent across tables?
  4. Is there one source of location data, or does each team hold its own copy?
  5. Does the location column have a spatial index?

The first audit or field check that finds a point hundreds of metres off will ask one thing: where did this point come from? A table that cannot answer makes every other point doubtful too, including the ones that are right.

This is the part we pay attention to in Space Technology, XETUP's geospatial line, which we are still developing. If you want to talk about the state of location data at your organisation, message us on WhatsApp at +62 823 1499 5005.