THE CASE
Since merging with Northwich Building Society and other, smaller companies throughout the 70s and 80s, the Cheshire Building Society's (CBS) network has grown. The mutual now has 51 branches and 13 property service offices throughout the North West.
CBS relies on the quality of its customer service, so it wanted to create a holistic view of its customer base. This would involve merging its online customer data with that from its affinity partners, which provide products such as health insurance, loans and credit cards. Preference data also needed to be integrated to allow CBS to identify customers' preferred channels of communication.
"We wanted to build a dataset to communicate relevant messages to all our customers, but also respect their privacy," says CBS customer knowledge manager Howard Ormesher.
Diagnosis
The main problem was merging data that had been captured in various formats across multiple databases. "When humans are involved there will always be errors," says David Beardmore, director at CBS database services supplier thinkdata. "We had long strings of data, rather than data split into separate fields, and many title and gender fields were unclear."
Treatment
Thinkdata used its proprietary tools to merge and deduplicate (see box) the various databases. This involved the use of Royal Mail's Postal Address File (PAF) and other syndicated datasets, such as Equifax's Insight database of 160 million credit agreements. Thinkdata could thus fine-tune the deduplications and validate elements such as birthdays and National Insurance numbers.
The data was normalised using look-up tables to identify misspellings and abbreviations. Merged records, fields, and other data inconsistencies were put into a common format. Matching algorithms were implemented to generate scored matches. This allowed for the tightening or loosening of matches, depending on whether communications were important regulatory or policy information, or marketing.
CBS's half-million-record database now offers a detailed and accurate view of all its customers, including their product holdings. It also reveals the best customers to contact, when, and how.
"It has allowed us to manage our marketing budget and time more efficiently," explains Ormesher, "and it has also reduced complaints. Some of our customers were complaining that we weren't communicating with them enough, and others were saying they didn't want us to communicate with them at all. Now we can identify all these people."
There has also been a clear reduction in mail wastage. As the number of products held across each household is now available, CBS can contact a married couple with one mailing regarding their mortgage, but contact them individually about ISAs or saving accounts.
"The savings in postage and literature run into hundreds of thousands of pounds because CBS is no longer mailing duplicates," says Beardmore.
Cleaning kit: Deduplication tools
WHAT IS DEDUPLICATION?
As your database grows, so will the number of duplicate records. Deduplication (dedupe for short) is the process of detecting and merging/removing duplicate records.
How does it work?
Accurate deduplication can be tricky. Common pitfalls include: field misalignments (eg a postcode in the address line), phonetic variations (eg Norton/Naughton), plurals, miskeyed data and abbreviations (eg Automobile Association/AA).
'Data standardisation' and 'fuzzy matching' techniques are required to locate these duplicates. Potential matches can be given a 'confidence' score based on likelihood of match. With large databases, it is usual to automatically purge records scoring above a defined threshold, thus minimising any manual verification.
What do I need to ask of a dedupe vendor?
For bureaus, question their expertise, software products used and timekeeping.
If you choose in-house software, evaluate to ensure it meets requirements and fits in with your IT infrastructure.
Gareth Fearn is marketing manager at helpIT systems.