Leg of lamb is the most stolen item at Iceland. Thieves
also like cheese, bacon and coffee. With the UK in recession,
shoplifters appear to be switching their sights from alcohol,
electric toothbrushes and perfume to food. Tesco, Marks &
Spencer and Iceland have all reported an increase in shoplifting
since the economy began to contract in the second quarter of 2008.
Tesco alone caught some 43,000 would-be thieves in the first half
of 2008, up 36% from the same period in 2007.
The impact of the recession on retailers is yet to be reflected
in any of the major surveys of shoplifting.
The
Centre for Retail Research's Retail Theft Barometer only has
figures up to the end of 2007. Those figures show that shrinkage -
losses from crime and waste - cost retailers 1.3% of sales in 2007,
down from 1.34% in 2006. Even though 2007 was the peak of the boom,
the losses were still huge. Customers stole some £1.6bn and
employees another £1.3bn. Suppliers took £209m fraudulently. Some
£73m was lost through card fraud and another £39m through robberies
or burglaries. Retailers lost £666m through waste. The systems and
security guards intended to reduce losses cost £785m. The total
bill was £4.6bn.
Retailers have used a variety of technologies to reduce their
losses. Closed-circuit television (CCTV) and electronic article
surveillance (EAS) - the tags attached to individual items - are
all visible in stores. Some retailers, however, are using a
different type of technology to reduce losses:
data mining.
During the summer of 2008,
British clothing chain Jaeger went live with a data mining
application in an attempt to identify where it was losing
money. The application, which is located centrally, interrogates
data held on different systems throughout the business, including
both the head office and the company's 90-plus stores and
concessions.
In common with every other data mining application,
Jaeger's LossManager
system, from IDM Software, uses a feed from the company's
electronic point of sale (Epos) system to spot potential fraud such
as excessive discounting by a single member of staff.
"It's a centralised system, but every single store is feeding
into it. We have got time and attendance feeding into it as well,"
says Steve Hearn, head of safety and security at Jaeger.
Jaeger had used a data mining application from SFR, a small
British supplier, before it took the decision to implement the IDM
system.
"We had SFR Storescan, but
it had long since been defunct and we were not using it because our
till architecture had changed. I started to wonder if that was the
right piece of software for us. It was quite cumbersome," says
Hearn.
Jaeger does not disclose its net profits because it is privately
owned. However, it is a mid-sized clothing retailer lacking the
colossal IT budgets available to, say, Tesco or Sainsbury's.
Hearn's first choice supplier was too expensive for his budget.
"I would have gone with an IntelliQ product [another British
software supplier], but for the price," he says. IDM Software is a
start-up aimed at mid-sized retailers and Jaeger was its first
retail customer.
Unlike CCTV and EAS, which are designed to catch thieving
customers, data mining applications are supposed to catch thieving
employees. "We do an awful lot of work on internal fraud," says
Hearn.
"The data mining system was put in to separate the usual from
the unusual," he adds.
Jaeger set up an audit team when the system went live in June.
The team's job is to use the new application to identify losses
wherever they occur - from dishonest employees to working practices
that waste stock.
Like other data mining applications, LossManager generates
exception reports. However, it would be misleading to rely solely
on these reports, according to IDM's chief executive officer Khuram
Kirmani.
"It's very easy to get swamped with false positives," Kirmani
says.
The employees responsible for loss prevention (in Jaeger's case,
the audit team) use their data mining application to generate
exception reports as usual. Then they continue to use the
application to ask more questions of the data so that they can
understand whether the system is reporting a false positive or a
genuine loss.
"Each question is based on the answer to the previous question,"
says David Snocken, IDM's commercial director.
Any project's success is limited by the user's willingness to
extract as much value as possible.
"It depends on the amount of effort the retailer has put in,"
says Kirmani. IDM says its system has reduced losses as a
percentage of sales below Global Retail Theft Barometer's 1.3%
average for UK retailers.
Although Jaeger has only had the system since June, it already
expects a return on investment in its first financial year. Hearn
says, "Data mining is widely accepted as having one of the fastest
returns on investment of any technology. We are still in the early
days in terms of assessing the benefits, but we are almost
double-counting our results to check they are right."
One of the earliest discoveries was that theft by employees was
only a small part of total losses at Jaeger.
"We have not gone out en masse and started arresting staff
members for fraud, but we have identified considerable numbers of
erroneous transactions. That is not to say that they are all
fraud," explains Hearn.
Data mining is helping the clothing retailer to manage its
stock, thereby reducing the need for markdowns when items go out of
season and reducing the number of items that go missing
altogether.
In a recession that has already claimed the scalps of
established retailers such as Woolworths and MFI, any initiative
that helps a retailer conserve cash will receive management
support.
"Data mining is even more important now in terms of being able
to understand margin erosion. Shrinkage is the last free margin on
the table. We have got to keep the stock current," says Hearn.
At the start of the data mining project, Jaeger forecast that it
would make a return on investment within six to nine months of the
project going live. That target will be met. Jaeger now expects
both a significant improvement in margins and a substantial benefit
to its net profits.
"The sheer opportunities to improve margin - it's not just about
fraud, it's about putting the wrong stock in the wrong place at the
wrong time. As a result, the decision to go with data mining was
very quick. I had no resistance from Jaeger," Hearn says.
In Jaeger's case, the difficulty with implementing its data
mining application did not come from the management it came from
the complexity of setting up data feeds between Jaeger's existing
store applications and its new centralised system. The company
decided to buy a data mining application in the summer of 2007.
"It was nearly a year," says Hearn. "It was nothing to do with
IDM, but to do with Jaeger. Our data was very complicated because
we have had so much in-house development of our systems. For
instance, at just one meeting, we had to review at line level the
data we used in over 800 fields."
Jaeger's data mining project will make a positive contribution
to profits at the most important part of the business cycle. As the
recession worsens in 2009, retailers will need to develop similar
projects that produce rapid returns on investment those that make
sustained improvements to net profits year after year will stand
the best chance of winning management approval. As money strains
lead more customers and employees to steal from retailers,
applications that can reduce theft will become increasingly
important.
How data mining gathers information
A data mining application becomes more powerful if it uses a
greater number of feeds from the retailer's other systems.
LossManager was built in the Microsoft Development Environment and
was written in C++ so it can be used to accept feeds from as many
different systems as possible.
"We can take feeds from almost anything. We can use that
information to ask if there is a correlation between a store that
loses a lot of product and EAS deactivations and alarms. One of the
departures from previous approaches is that for an application to
be truly effective, we have to integrate multiple sets of data,"
says Hearn.
Several data mining applications already use video feed from
CCTV cameras to make sense of Epos data. Many retailers would like
to use the two technologies together, but they are unable to do so
because their CCTV cameras use analogue rather than digital film.
For most retailers, the cost of replacing analogue cameras with
digital cameras far exceeds the financial benefits that they expect
to gain from reducing their losses.
Retailers with
radio frequency identification (RFID) projects could even use
the information from tagged pallets or individually tagged items
within their data mining applications. Unfortunately for advocates
of RFID technology,
the only retailer with a public RFID project in the UK is Marks
& Spencer, which tags different ranges of clothing in most
of its major stores.
Audit teams use a type of network theory called link analysis to
understand the patterns between data on different systems. Auditors
look for symmetric patterns between two sets of data, or more
likely asymmetric patterns, to understand the relationships between
different types of information.
Retailers are not the only organisations that use data mining to
look for correlating information. Governments have used data mining
to sift through huge amounts of data to identify potential
terrorist attacks. In 2002, the Pentagon started a secret project
called
Total Information Awareness in an attempt to identify
terrorists. Total Information Awareness was a data mining project
on a massive scale. In 2003, it was cancelled after Congress
removed funding over fears that it was too intrusive.