Intelligent Systems incorporating Automatic Classification and Carbon Footprinting for Corporate E-Procurement
Submitting InstitutionUniversity of Reading
Unit of AssessmentElectrical and Electronic Engineering, Metallurgy and Materials
Summary Impact TypeEconomic
Research Subject Area(s)
Information and Computing Sciences: Artificial Intelligence and Image Processing, Information Systems
Economics: Applied Economics
Summary of the impact
Two Knowledge Transfer Partnership projects, carried out between 2006 and
2009, between an e-commerce marketplace provider (@UK plc) and the
University of Reading, led to the development of two software tools that
were launched in 2010. The tools, SpendInsight and GreenInsight, are the
first of their kind to use artificial intelligence techniques to handle
the extremely challenging data associated with purchasing in large
organisations. Since their launch, these tools have been used by @UK plc
to identify procurement savings and environmental costs of procurement
activities for governments, multi-national corporations, academic
institutions and healthcare providers. Over the last three years @UK plc
has benefitted from the launch of these products as it has provided them
with a competitive advantage over the market place, increased the quality
and efficiency of their spend analyses and led to multi-million pound
licensing agreements. An analysis of spending in some of the NHS Trust
Foundations has led to changes in procurement behaviours that have
resulted in hundreds of thousands of pounds saved to date — benefitting
not only the NHS, but also taxpayers.
Purchasing of goods and supplies by large organisations is a complex
process, which can be streamlined through an e-procurement system.
However, implementing e-procurement requires tools that can cope with vast
amounts of data from multiple and disparate sources, such as supplier and
product information, in order to analyse where the best value is and where
efficiency can be improved. The nature of e-procurement data (large
quantities, heterogeneous, sparse, unstructured, distributed and noisy)
makes it challenging to apply existing state of the art management
In 2006, three linked Knowledge Transfer Partnership (KTP) projects were
started to create innovative solutions to the challenges of e-procurement
data; the solutions needed to be cost effective and commercially viable.
Two of these were with the University of Reading, while the third took
place at Goldsmith's, University of London. The company @UK plc, a leading
e-commerce marketplace provider, giving support to over 1 million
businesses worldwide, was the industry partner on all three KTP projects.
The University of Reading was the knowledge base partner for two of the
KTP projects which focused on innovative approaches to data classification
and related artificial intelligence techniques; one of these projects
concentrated on classification of products, while the other dealt with
ranking web pages on the basis of textual search, as well as finding
property structure from product descriptions based on natural language.
Goldsmith's was the knowledge base partner on the third KTP project, which
concentrated on systematically spidering web pages to gather relevant
product information from them. These linked projects ran for three years
from 2006 to 2009.
Classification of data using artificial intelligence
The University of Reading team was comprised of: Dr Richard Mitchell,
Associate Professor (2013-present) previously Senior Lecturer
(1994-2013); Slawomir Nasuto, Professor of Cybernetics (2012-present),
previously Reader (2007-2012); Dr Virginie Ruiz, Associate Professor
(2013-present), previously Senior Lecturer (2004-2013); and Victor
Becerra, Professor of Automatic Control (2012-present), previously Reader
The Reading team brought decades of experience in classifying challenging
data to the partnership with @UK plc. When data are collected by a spider
— a program that "crawls" through the web looking for relevant data in
response to a query about, for example, a product — they need to be
analysed to determine the category to which the product belongs; this
classification can then be used to find equivalent products from different
suppliers. The products also need to be ranked to determine which are most
important or relevant to the search request. The Reading team have been
developing and applying methods for classifying data using various
artificial intelligence approaches, including neural networks [1,2,8],
evolutionary computing, swarm intelligence, adaptive signal processing,
and machine learning [1,2]. The team have developed and applied these
methods in response to challenges faced in a diversity of disciplines,
including health [2, 7 & 8] and economics .
There are several challenges with real world text classification,
including poor class structure, overlapping classes and blurred boundaries
between categories. Moreover, training data pooled from multiple sources
tend to be inconsistent and contain erroneous labelling, leading to poor
performance of standard text classifiers. In 2010, the Reading team looked
at how health service products were classified to specialised procurement
classes in order to examine and quantify the extent of these problems .
They presented a novel method to analyse the labelled data by selectively
merging classes where there was not enough information for the classifier
to distinguish them. Additional contributions made through the research
that lead to SpendInsight include: ensuring that the data pre-processing
and classification methods were scalable so that they were able to process
procurement data from large organisations (`big data') in reasonable time;
the development of a sophisticated rule engine for de-duplication ,
which is employed to automatically detect and eliminate duplicate items
from the procurement data; and finally the development of methods for
automatic detection of attribute data in textual descriptions of products
This research led to the development of an intelligent spend analysis
system known as SpendInsight which became a key component in the @UK plc
e-procurement and e-marketplace platform.
Methods to estimate environmental cost
With individual products classified through SpendInsight, the Reading
researchers then developed methods to estimate the environmental cost of
the product by mapping it to the ethical and environmental information
held for millions of products by the Centre for Sustainability Accounting.
The resultant system, GreenInsight, was launched in 2010 and is used by
procurers to assess the carbon footprint of their purchases; they can now
compare the cheapest price economically and environmentally, and thus
quantify the cost of `being green'.
References to the research
These outputs have been internally assessed as of at least 2* quality.
Those suggested for quality assessment are indicated with *.
 *Becerra, V. M., Galvao, R. K. H. and Abou-Seada, M. (2005) Neural
and wavelet network models for financial distress classification. Data
Mining and Knowledge Discovery, 11 (1): 35-55. doi:
 *Froese, T., Hadjiloucas, S., Galvao, R. K. H., Becerra, V. M. and
Coelho, C. J. (2006) Comparison of extrasystolic ECG signal classifiers
using discrete wavelet transforms. Pattern Recognition Letters, 27
(5): 393-407. doi: 10.1016/j.patrec.2005.09.002 Citations=16
 *Roberts, P., Howroyd, J., Mitchell, R. and Ruiz, V. (2010)
Identifying problematic classes in text classification, in 2010 IEEE 9th International Conference on Cybernetic Intelligent Systems, Sept
1-2, 2010, Reading, UK: Institute of Electrical and Electronics Engineers.
 Roberts, P., Mitchell, R. and Ruiz, V. (2009) Using triangulation to
identify word senses, in: 8th IEEE International Conference on
Cybernetic Intelligent Systems — IEEE Systems, Man &
Cybernetics Society, September 9-10, 2009, Birmingham, UK: Institute
of Electrical and Electronics Engineers.
 Roberts, P. J. (2011). Automatic Product Classification. PhD
Thesis. University of Reading: UK.
 Brown, M.(2011) Automatic production of property structure from
natural language. PhD Thesis. University of Reading: UK.
 Guo, Q., Shao, J. and Ruiz, V. F. (2009) Characterization and
classification of tumor lesions using computerized fractal-based texture
analysis and support vector machines in digital mammograms, International
Journal of Computer Assisted Radiology and Surgery, 4(1): 11-25.
doi. 10.1007/s11548-008-0276-8. Citations=18
 Bakstein, E., Burgess, J., Warwick, K., Ruiz, V., Aziz, T. and Stein,
J. (2012) Parkinsonian tremor identification with multiple local field
potential feature classification. Journal of Neuroscience Methods,
209 (2): 320-330. http://dx.doi.org/10.1016/j.jneumeth.2012.06.027
All citation values are from Scopus as of 3rd October, 2013.
Details of the impact
The classification techniques developed at Reading [5-8] became essential
components of the Eprocurement system, SpendInsight, which was launched in
February 2010 by @UK plc. The system uses artificial intelligence
techniques to enable e-procurers to analyse their purchases and identify
potentially significant savings. @UK plc subsequently launched the
GreenInsight system, which again incorporated the artificial intelligence
techniques developed by Reading. GreenInsight estimates the carbon
footprint of products and procurement activities in order to help
businesses develop environmentally friendly procurement policies.
@UK PLC benefits from the launch of SpendInsight and GreenInsight
SpendInsight immediately increased @UK plc's analysing capacity; within
the first two months of launching the SpendInsight website, @UK plc had
analysed over £35 billion in spend, which was an increase from the £16
billion it had analysed the previous year [a].
significantly improved @UK plc's efficiency, producing results faster, at
less cost and with better results than previous systems. @UK plc was
working with the NHS London Procurement Partnership (LPP) and using the
leading spend analysis system at the time, they had taken 2 years to
analyse less than 25% of all of London's spend, costing around £500,000
[a]. In 2010, they implemented SpendInsight and in 6 months had analysed
London's entire spend, with better results, for significantly less cost
On September 4th, 2013, @UK plc announced that they had agreed
a £3.4 million conditional licensing agreement with Tungsten Corporation
plc for SpendInsight [b].Tungsten will market the software to its 122
global clients as TungstenAnalytics and @UK will receive up front
establishment fees and installation costs of around half a million pounds
The @UK Chairman has stated that; "SpendInsight and GreenInsight have
revolutionised our business. SpendInsight is recognised as still being
globally unique 3 years after development, and we just won a £ 3.4m
contract where the customer is licencing SpendInsight. This is 1.5x our
current turnover. Additionally the SpendInsight capablity has been a major
factor in our relationship with Visa and the global roll out of our
marketplace technology where SpendInsight delivers an instant business
case for buying organisations." [c]
Taxpayers and NHS benefit from savings identified through SpendInsight
SpendInsight was used to identify considerable savings for the NHS London
Procurement Partnership (LPP) - both soft process savings and hard
cashable savings - and provided impetus to standardise and centralise
their e-procurement across the Trusts' back office. SpendInsight also
identified which were the key suppliers and key commodities, a process
that had the effect of driving both user and supplier acceptance of the
change to e-procurement.
Over the course of the project the researchers
obtained and processed NHS procurement data for 73 NHS trusts in the UK,
the majority centred in London. The SpendInsight system was used to
analyse the purchase orders, invoices and contracts of the different
organisations, identify equivalent purchases and benchmark the best
possible prices for each product, and identify savings.
The analysis found that significant savings could be made within the NHS
with relatively simple changes in procurement behaviour. For example, the
amalgamation of small ad-hoc orders into larger, less frequent orders
could standardise product choices and facilitate lower-costs for
guaranteed high-volume deals. One NHS Trust saved £320,000 in 2011 alone
by switching to one supplier of examination gloves and ordering 2 choices
of glove rather than 20 [d].
@UK PLC provided analysis services to the National Audit Office (NAO) to
assist them in the preparation of their 2011 report on the procurement of
consumables by NHS hospital trusts. The National Audit Office (NAO)
included @UK plc's SpendInsight analysis in their 2011 analysis of the NHS
Trust's procurement and spending on consumables [e]. The NAO concluded
that if the procurement system was utilised across all NHS Trusts in
England, they could "make overall savings of at least £500 million, around
10% of the total NHS consumables expenditure of £4.6 billion" [c].
After the initial analysis using SpendInsight in 2012, the Royal
Berkshire NHS Foundation Trust decided to continue using the system to
analyse their quarterly spend for 2013. The Principal Procurement Manager
stated that "with an addressable spend of £ 60 million, to be able to save
even one percent of that is a benefit". The Trust has received data from
their spending review and they "appear to have clawed back £100,000 as a
result of actions taken from [their] first pass" [f].
Businesses, governments, academic institutions and others benefit from
SpendInsight and GreenInsight analyses
@UK plc works with organisations all over the world and has now used
SpendInsight to analyse "over £67 billion in spend, and millions of
products from 100's of thousands of companies" [i].
The following is a
confidential list provided by @UK of users of SpendInsight and
GreenInsight, which shows that governments, health authorities,
multi-national companies and academic institutions have all benefitted
from the use of SpendInsight and GreenInsight.
|| Westminster City Council
States of Jersey
The Highland Council
Tower Hamlets Council
|Higher Education Purchasing Departments
|| University of Huddersfield
University of Reading
University of East Anglia
Goldsmiths' College (University of London)
||Steria UK Corporate Ltd
||New South Wales
|Health Authorities / Hospitals
|| London Procurement Partnership
Royal Berkshire Hospital
Basingstoke and North Hampshire NHS
In November 2011, at an event organised by the Business Application
Software Developers Association (BASDA), the Director of Solutioning for
the NHS made a presentation about the SpendInsight and GreenInsight
systems and claimed that they would be "saving the NHS £250 million over
10 years" and equated this to "the cost of 12,000 additional nurses".
Sources to corroborate the impact
[a] `SpendInsight', @UK plc [website] http://bit.ly/1fgJVPW
accessed 3 Oct 2013. Gives evidence of increased volume of analyses in
2010 over 2009 as a result of implementing SpendInsight.
[b] `SpendInsightTM Agreement', INTELLISYS. http://bit.ly/1fgJVPW
Gives value of financial gains for @UK plc for a conditional licensing
agreement for the software system developed with Reading researchers.
[c] Chairman, @UK plc (testimonial letter on the impact of the software)
[d] Denwood, A. (27 Sept 2011) `Hospital saves fortune just by swapping
rubber gloves', BBC News Health <http://www.bbc.co.uk/news/health-14971984>.
Provides figures for cost savings made by Barts and the London NHS Trust
by making simple changes in procurement behaviour as a result of
[e] National Audit Office (2 February 2011) The procurement of
consumables by NHS acute and Foundation trusts, Report by the
Comptroller and Auditor General, HC 705, Session 2010-2011. London: The
Stationery Office. http://bit.ly/187uszM
[f] `Case Study: Spend Insight', @UK SpendInsight Savings and
Benchmarking, Royal Berkshire NHS Foundation Trust. http://bit.ly/17odbzn
[g] Head of Purchasing and Supplies, Basingstoke and North Hampshire NHS
[h] Data Enablement Manager, NHS London Procurement Partnership*
[i] `Process', GreenInsight [website] <http://www.green-insight.com/process.html>
accessed 6 Oct 2013. Gives evidence of the extent and reach of
*Contact details provided separately