Geo Strategies Case Study
The Challenge
Geo Strategies / Romanian Government - National Postcode & Address System
The Top Gear Romania trip from Series 14, Episode 1, featured the hosts driving Aston Martin, Ferrari, and Lamborghini cars to find the "World's Best Road" but their SatNav did not even recognise Romania as there were few standard roads and no postcodes, leading to hilarious navigational mishaps. Today if you drive in Romania your satnav will function due to the work done by RD Research.
As a condition of joining the European Union, Romania had to implement a formal national postcode and addressing system. Outside Bucharest, streets had no official names, the same road might be known by several different local names, and many addresses were little more than descriptions given to the local postman. Utility companies relied on local staff who simply “knew” where people lived. Another contractor had already attempted to solve the problem and had concluded that it was practically impossible.
RD Research were asked to design and develop that engine. The system had to cope with multiple data sources, many-to-one street names, missing validation, and frequent street renaming by local mayors, and it had to run at national scale for logistics, mapping and sat nav providers.
The Development Process
We began with a technical review of the existing datasets and the failed first attempt. It became clear that there was no single-step solution. Instead we designed a staged process built around a central database that combined Romanian postal data, voting registers and government locality hierarchies into structured reference and variant tables.
Geo Strategies Preliminary Spec
The new system was specified initially in Microsoft SQL due to its support for the Romanian language. We then re-engineered it in MySQL and implemented a multi-phase pipeline:
- Input data from Geo Strategies was normalised into a common structure, with Romanian diacritics handled consistently with premises (block, stair, floor, apartment).
- Sophisticated pre processing routines standardised abbreviations, removed noise and converted thousands of local variants into canonical forms.
- Matching logic then combined exact, rules based and machine learning techniques, and soundex style phonetic matching tuned for Romanian, to link each input address to the best candidate in the Knowledge Base.
The process was automated, running on dedicated servers that processed millions of records at a time and wrote back confidence scores and reasons for each match.
Where results were ambiguous, the system flagged addresses for manual review and for later incorporation into the Knowledge Base, so the platform improved over time.
Throughout the 12 month project we worked closely with Geo Strategies’ data specialists. The system was engineered so that new official data releases, such as updated postal files or locality lists, could be ingested and reconciled without redesign.
Key features
- National address matching and data cleansing for Romanian streets and premises
- The final database combined official postal data, voting registers and government locality hierarchies into reference and variant tables
- Handling of complex multi part premises such as block, stairwell, floor and apartment, mapped into a consistent structure
- Batch processing framework capable of running long multi day jobs on dedicated servers and logging detailed audit information for each address
- Grid based postcode allocation that linked matched addresses to geographic co ordinates for use in logistics, mapping and sat nav systems
- Tools and reports for Geo Strategies to review ambiguous cases, correct data and continuously improve match rates over time
Outcome
The project delivered what had previously been considered unachievable. Geo Strategies asked us to achieve a minimum automated match rate of 50 percent; the RD Research engine consistently delivered around 67 percent fully matched and validated addresses across the country. The remaining third were resolved by field teams visiting properties, capturing the locally used street names and feeding them back into the system, where our gridding approach assigned each a stable postcode.
"This was a complex project and I despair of what might have transpired without their level of expertise."
- Bill Metcalf, Director, Geo Strategies Ltd