Virtual warehousing and market intelligence in online book retailing
Submitting Institution
University of BathUnit of Assessment
Computer Science and InformaticsSummary Impact Type
EconomicResearch Subject Area(s)
Information and Computing Sciences: Computation Theory and Mathematics, Computer Software, Information Systems
Summary of the impact
Bath research on the design of multi-agent software systems governed by
norms and institutions has directly influenced the development of the
essential business systems of an internet-based trading company, and been
instrumental in their success.
The Book Depository (BD) was founded in 2004. In 2005, their Chief
Technical Officer, Emad Eldeen Elakehal, sought the expertise of Julian
Padget in the Department of Computer Science at Bath, and began a
part-time PhD, working on the application of normative frameworks to the
design and implementation of business systems. Elakehal has applied these
principles in the design and construction of two key subsystems of BD's
software infrastructure: the catalogue maintenance system (live since
2006) and the price checker and setter system (since 2008). Their
effectiveness has underpinned the growth and success of the company by
providing robust software implementation of business processes that adapt
to changing market conditions. The company's turnover grew from £24M to
£120M from 2008 - 2011, and continues to grow. The software systems
enabled this growth to take place with no increase in the operations
team's manpower, and now handle a catalogue of over 8 million titles, from
120 suppliers, all available within 48 hours to customers on the Book
Depository's own web site or via Amazon's marketplace: all Amazon book
customers have seen offers of books generated by this software. The
software underpins BD's award-winning business, a unique offering in the
book retail sector which attracted takeover by Amazon in 2011. BD's
Managing Director states that "without the agent/norm based technical
systems not one of the business' USPs could have been effectively
realised."
Underpinning research
Norms are a well-established concept in law, social sciences and
economics that express constraints on individual behaviour or obligations
defining situations that are to be achieved or to be avoided. The
long-term objectives of this research are (i) the expression of norms in a
form suitable for both human and software comprehension and (ii) design
patterns for the creation of software that is capable of norm-compliant
behaviour.
Our work on the formalization of norms in computer science and artificial
intelligence began in the late 1990s [1]. Borrowing from work in the
domains noted above, Padget & colleagues proposed that the distributed
economy afforded by the Internet should exploit via mimesis the strategies
evolved over thousands of years of human commerce, e.g. norms such as
eligibility requirements and formal specifications of acceptable behaviour
which are policed to detect early problems. The major research
contributions that underpin the impact reported here are (i) the
definition of a formal model for normative frameworks [2] based on the
principle of real world events that "count-as" normative events, (ii) a
sound and complete translation process from the formal specification into
a logic program under answer set semantics, to create a model that permits
the formal verification of system properties through a form of
model-checking, and (iii) identifying the different requirements for the
use of such models at design time, when the objective is to explore total
system properties, and at run-time, when the objective is to provide
normative guidance for software components [3].
The work of the first two contributions was carried out by Padget (senior
lecturer), De Vos (lecturer, then SL) and Cliffe (PhD student jointly
supervised by Padget and De Vos) in the period 2001-2006 and that of the
third by Padget, De Vos, Cliffe (RA until 2010) and Balke (2010-2012, PhD
student jointly supervised by Padget and De Vos).
Further work combined these ideas with work on agent-oriented software
engineering to translate them into the business domain, informing first
the design and implementation of the catalogue maintenance system
developed by the Book Depository [4] and later that of their price checker
and setter systems [5]. Notable features of both these systems are (i) the
attention paid to self-monitoring and self-management and (ii) the
capacity to change component behaviour in response to a changing
environment (especially in the case of the price checker/setter),
reflecting the principles of norm-governed systems.
This latter work was carried out by Padget and Elakehal
(industrially-based PhD student, supervised by Padget, since 2005).
Subsequently, building on the third contribution (design vs. run-time
requirements) and as a result of evaluating the development processes for
the above systems, Padget and Elakehal have developed a new
business-oriented software engineering meta-model for the construction of
self-managing multi-agent systems [6].
References to the research
[2] * Owen Cliffe, Marina De Vos, and Julian Padget. Answer set
programming for representing and reasoning about virtual institutions. In
Katsumi Inoue, Ken Satoh, and Francesca Toni, editors, CLIMA VII, volume
4371 of Lecture Notes in Computer Science, pages 60-79. Springer, 2006
[DOI:10.1007/978-3-540-69619-3_4]
[3] * Tina Balke, Marina De Vos, and Julian Padget. Normative run-time
reasoning for institutionally-situated BDI agents. In Coordination,
Organizations, Institutions, and Norms in Agent Systems V, volume 7254 of
Lecture Notes in Computer Science, page tbd. Springer, 2012. [DOI:10.1109/wi-iat.2011.49]
[4] * Emad El-Deen El-Akehal and Julian A. Padget. Pan-supplier stock
control in a virtual warehouse. In Michael Berger, Bernard Burg, and
Satoshi Nishiyama, editors, AAMAS (Industry Track), pages 11-18. IFAAMAS,
2008. [http://dl.acm.org/citation.cfm?id=1402798]
[5] Emad Eldeen Elakehal and Julian Padget. Market intelligence and price
adaptation. In Proceedings of the 14th Annual International Conference on
Electronic Commerce, ICEC '12, pages 9-16, New York, NY, USA, 2012.
ACM.[DOI:10.1145/2346536.2346538]
* marks the outputs best indicative of the quality of underpinning
research
Details of the impact
The Book Depository is an online book retailer, founded in 2004. Book
retailing traditionally falls into two classes: high-street style
retailers selling the relatively small number of top-rated titles, all
available instantly, and specialist stores serving niche markets, with
many titles available but longer delivery times (2 - 4 weeks) and
relatively few sales per title (under 1 unit/month worldwide) — the "long
tail". BD's unique aim was to bring high-street style availability and
turnaround time to the long tail sector.
The research of BD's Chief Technical Officer Emad El-Akehal with
Padget has driven the design of two software systems crucial to BD's
operations: the catalogue maintenance system and the price
setter/checker. The operation of these systems enables BD's business to
run: the catalogue system maintains a "virtual warehouse" of books,
currently over 8 million titles, which are physically located at over 120
distinct suppliers; and the price system monitors supplier cost and
competitor prices to enable BD to offer their books at optimal pricing.
[7]
The adoption of the design principles developed in Padget's research has
enabled BD to deploy software systems that are highly distributed and
decoupled — the software comprises many separate software agents that run
on computers all around the world with minimal reliance on one another —
and that can monitor and repair themselves thanks to the norm-based
governance of the system. This means that the system is highly scalable,
for example no change to the running code was required when scaling from 5
to 20 suppliers early in BD's growth, as well as being modifiable,
so that for instance changing data format to suit a new supplier is easy,
and reliable, because many faults and problems can be addressed by
the system itself stopping the faulty agent and replacing it with a fresh
one. The distributed nature of the system allows agents to be placed where
they are most effective: for instance, agents that need to make a great
many requests for data to Amazon's servers in the US can be located in the
US, while agents holding data governed by UK law can be located in the UK.
BD's Managing Director explains the significance of Padget's work to BD's
operation as follows [7]:
"The system design was grounded on the research efforts of the company's
Chief Technical Officer; Emad Eldeen Elakehal during his PhD program under
the supervision of Dr Julian Padget at the University of Bath — UK. Both
the catalogue maintenance (Virtual Warehouse System) and Pricing System
are designed as multi agent systems and are responsible for the core and
crucial elements of our daily operations. These systems in particular are
responsible for the implementation of our pricing strategies and they
contribute to the management of the company's offers in both third party
marketplaces and our own website.
Both systems have shown high degree of resilience and scalability over a
period of more than six years and they are still in operation today. We
are able to change and modify the system attributes easily with minimal
amount of development effort, as a result of their norm/agent based
design. The company's operations and the number of transactions have grown
more than 10 fold since implementation and the systems were able to scale
up to support this growth. This acceleration is clearly reflected in the
company's annual revenue, turnover increased from £24M in 2007/8 — first
year of operating these systems — to over £120M in 2010. The company
continues to grow with expectation of reaching circa £150m turnover this
current year, 2013.
The size of the operations and technical support team who looks after
these specific systems has remained static over the last six years, due to
the reliability of the software and to its dynamic features."
Thanks to these systems, and Padget's work with Elakehal, BD has been
able to fulfil its goal of offering its 8 million titles with 48 hour
turnaround, creating a service in the long-tail sector that was
previously unavailable. It has rapidly become the market leader in
long-tail book provision. As the Managing Director notes, turnover has
grown from £24 million to over £120 million in the period since
2008.[7,8] Most remarkably, this has been possible with no increase in
the operations team: the software systems are still managed by just
two staff. The success of the business is directly attributable to the
effectiveness of their software systems: BD's managing director states
that "without the agent/norm based technical systems not one of the
business' USPs could have been effectively realised." [7]
Every user of Amazon who has searched for a book has been a user of this
software. The Book Depository sells its books via Amazon's third-party
"marketplace", and every book search returns "new and used" price results,
one of which is invariably from the Book Depository. This marketplace
offer is generated by BD's software; the fulfilment of orders placed via
Amazon is also handled by this software. The Book Depository became such a
significant player in the long-tail sector, winning the Queen's Award for
International Trade in 2010 [9], that Amazon chose to acquire the company
in 2011 [10]. Though now an Amazon subsidiary, BD continues to trade in
the same way under its own brand, using the same software.[7]
Sources to corroborate the impact
[7] Factual statement from Managing Director, the Book Depository.
[8] Book Depository data from Companies House. Report generated from FAME
database 25th July 2013.
[9] The Book Depository wins Queen's Award 2010, http://www.thebookseller.com/news/book-depository-wins-queens-award.html
(archived version of 29 August 2013 held on file)
[10] Amazon Purchases the Book Depository. http://www.retailgazette.co.uk/articles/23044-amazon-purchases-the-book-depository
(archived version of 29 August 2013 held on file)