Address Matching

Toronto-street-mapBespoke’s Fractal® Matching Algorithm provides unparalleled accuracy at parsing and organizing messy address data. Bespoke employs a hybrid algorithm-human approach, with a combination of two unique apps – Fractal Pattern Matching and Recursive Knockout Classification to highlight areas where the two algorithms disagree, signifying the need for human input. The Bespoke Fractal process can achieve accuracy above 99% based on the automated process alone, and near perfection with the hybrid approach. This product is available as either a one-time batch process on current records, or as a permanent feed-back loop to maintain data hygiene.

Incomplete, messy or inaccurate customer addresses can be a big drain on company resources and be a major impediment to efficient targeted marketing. Bespoke Fractal® helps our clients overcome these challenges by unifying multiple disparate and discordant address databases into one, clean, properly formatted address database. The messier the addresses, the more fun we’ll have fixing it! How do we do this better than anybody else? Bespoke Fractal® employs two proprietary algorithms that are augmented by human intelligence; we let the computers do what they do best (do the heavy, grunt work for the majority of cases) and have humans do what they do best (adapt and assess unusual or unexpected situations).

Fractal Pattern Matching | Algorithm #1

Bespoke’s proprietary Fractal Pattern Matching (FPM) builds out a series of potential patterns that each address could match, then applies the best pattern match to each address. We leverage this unique RegEx-based matching process which is flexible enough to capture close to 99% of all patterns, with a high degree of accuracy. Our pattern matching algorithm acts as our foundational process, giving us the baseline assumption as to the correct address. This process is in perpetual development and improves with each client that leverages this ground-breaking technology.

Recursive Knockout Classification | Algorithm #2

Recursive Knockout Classification (RKC) takes an entirely different approach to address matching/parsing. RKC begins by searching for a foundational element such as a street name, then recursively goes backward and forward to determine what each surrounding element is (address number, street number, unit number, etc). This recursive approach provides a valuable, unique perspective that enables us to have two algorithmic perspectives on each address fed into our system.

Combining the Two Algorithms

Diversity of opinions and perspectives can lead to better information; employing two philosophically unique algorithms enables us to have two powerful, automated perspectives that lead to unparalleled accuracy. Each process on its own can achieve greater than 98% accuracy on most data sets, together, they can achieve even greater results; this is accomplished by assessing where the two processes agree and where they diverge, infused with information on where each algorithm fails. Where the two algorithms agree, it is virtually guaranteed that the address has been parsed correctly; where the algorithms disagree, we ship it off to our team of experts who quickly go through the remainder and tick off which algorithm processed it correctly, or if manual intervention is required.

Approaching Perfection

Our unique FPM and RKC approaches can provide cost-effective automated address parsing, cleaning and classification. We can scale this up with human analysts to achieve near perfect results, depending on your budget.