You are Home   »   News   »   View Article

PIDX - how e-commerce standards should evolve

Friday, November 30, 2018

Oil and gas e-commerce standards organisation PIDX had discussions about how commerce technology is evolving and how e-commerce standards should evolve with it, at its 2018 European Conference in London on June 5.

The basic idea is that there is an enormous amount of transacting which goes on inside the oil and gas industry, and companies would reduce the administration cost of this if they all used standard systems. Companies do not get any competitive advantage from handling transactions in their own way.

But we are seeing big evolutions in the way that companies are working with trading partners, including transactions handled on a software-to-software basis (rather than from sending electronic documents), suppliers providing richer information about their products which can automatically populate oil company systems (and help analyse purchases), and effort to standardise part numbers between suppliers for the equivalent item.

PIDX sees that the quest for more digitalisation in industry should go together with a quest for more standardisation, and PIDX provides a platform for 'mature discussion' about how that should be done.

The three core services of PIDX are providing standard legal frameworks (behind e-business), providing standard ways to manage digital catalogues, and supporting systems integration between buyers and suppliers.

PIDX standards describe how electronic documents such as invoices can be transferred between buyers and suppliers in a standard XML format. But we may see a growth in communications made between one software package and another.

Chris Welsh, board member with PIDX, suggested that PIDX could also develop API standards, describing how different software systems should integrate together. Some other standards bodies have moved in this direction - he cited FHIR Healthcare (Fast Healthcare Interoperability Resources) a healthcare data standards body, which has extended its scope from just XML standards to also doing API standards.

PIDX could also broaden its standards to include communications with suppliers in the after-sales, including about performance of their equipment and analytics, he suggested.

PIDX is also interested in in finding ways to make it easier to map together different part number systems and taxonomies. Currently every buyer has their own taxonomy. There needs to be a format to manage one supplier's part numbers within another company's taxonomy, he said.

Andrew Mercer, BP

Andrew Mercer, CIO of BP for Middle East and Africa, and also a PIDX board member, emphasised that the oil and gas industry is massively interconnected and so there is a lot of communication between different partners, and the standardising this communication can make life more efficient.

Oil majors like BP only get competitive advantage out of a small part of their overall activities, where they get better overall business results as a result of doing something better than their competitors. For the rest of their activities, there is no reason for an oil major to do things its own way - it would be easier for everyone if it adopted standards. Included in this is the way of creating a purchase order number or sales order, he said. There is no reason for an oil company to do this its own way.

He noted that BP is keen to move to more general-purpose software tools, rather than specialist packages. It has many specialist software packages it has bought from different companies, and has challenges integrating it all together. 'You end up with all these siloed systems,' he said. 'The number of data solutions we've got and data standards is a real barrier.'

Another theme with BP's technology development is moving systems to the cloud, but it is proving 'not as easy as people make out,' he said.

Sparesfinder - business translation

Sparesfinder has set up a spares numbering translation service, which aims to connect numbers in a buyers' system with numbers in a suppliers' system.

It scans the description of items and makes suggestions of what might match with something else, for example comparing every centraliser casing in the database.

It is too common to have suppliers asking questions of buyers about different parts, and nobody knows they are talking about the same thing, said Tom Cave of Sparesfinder.

The international retail industry solved this problem in just 4 years, after realising how much money was being spent on managing data about items (including stock keeping). It led to standard bar codes, and standard devices to read them. But the oil and gas materials industry has not yet managed to do the same thing, and it is leading to a significant cost, Mr Cave said.


Power and automation company Siemens conducted a survey of people from upstream oil and gas operators, and found that 50 per cent think digital technology and leads to faster decision making, 45 per cent think it leads to better asset management, and 46 per cent think better real time decision making. 59 per cent think it can help improve productivity and 25 per cent think it can improve training.

Siemens' 'AX4C Cloud' software is intended to help companies in a supply chain better collaborate. It maps the processes along the delivery chain. Companies can use it to share information.

Phil Lavin, development consultant for IT at Siemens AXIT, sees the main challenges to digital roll-out as improving collaboration between business and IT, overcoming reluctance to share data, fear of change, challenges convincing people to participate, and a 'wait and see what others are doing' attitude.

He sees a varying degree of maturity of digital business models in different sectors - perhaps with the highest in media and trade, and lowest in process management and energy.

Automatic classification systems

Preminor of Canada has developed a system to automatically classify purchased items in a rich way, using machine intelligence.

The purpose of the classification is that it enables companies to analyse their spending in different ways. For example, they might want to work out how much they spend on a certain sort of valve every year, which they can use as part of a negotiation with a supplier. They might want to see the value of components in a certain completion, or how much they spend on renting certain items each year and whether it would make sense to buy them, said Andy Ross, founder and principal of Preminor. (The company was formerly known as ACT Consulting).

Purchasing systems commonly allow companies to categorise or 'bucketise' spend into different areas, but doing the sort of analysis above requires much more granular classification.

Some companies have developed rules based classification, for example if the item contains the word valve it is categorised as a valve. The problem with a purely rules based approach is that it is hard to define something in the real world just using rules - you can end up with rules conflicting, saying that an item should be described in two different ways, or no applicable rule at all.

An algorithm / machine learning system can be much more expressive than a rules based engine, better able to find the right solution when there is a conflict (an item could be in 2 different categories, or it isn't obvious which category it should go into).

Another approach has been to send the data to an offshore processing centre (such as in the Philippines), which apply a combination of rules and lower cost labour. But doing this is still 'time consuming, expensive, inconsistent and inaccurate,' he said.

Human categorisation is not particularly consistent, with different people classifying in different ways, he said.

The experiment by ACT involved taking a year's worth of purchase data from one company, 3m line items, and trying to build an automated classification system. ACT developed an algorithm for how items should be classified.

Ultimately, it led to a 10 per cent increase in accuracy from labelling by computer, vs labelling it by people.

There was quite a lot of work involved in training the system - it can be more labour intensive training an AI based classification engine than doing the classification yourself, Mr Ross said. But of course once it is trained, the need for human support reduces.

A basic AI system can be put together in 2-4 weeks, and then taken another 4 weeks to be improved. There may be a need to write some rules on top of that, stating that in a certain situation the machine should take a certain answer.

To 'tune' the model, ACT put together a matrix showing when the choices between people and machine are most likely to diverge - then an human expert could check who was actually making the best choice and tweak the computer model If the computer's choice was the worst one.

One challenge was that the classification used the vendor's description of the product (as shown on the invoice) as the starting point, and many vendors did not have a very precise description of their product. However, the project did make it possible to identify which vendors provided the worst product definitions on their invoices, so they could be asked to improve.

After many years of looking for a 'use case' for classification for machine learning, this is the first time ACT has found one which looked like it might provide strong real world value, he said.

Associated Companies
» PIDX International
comments powered by Disqus


To attend our free events, receive our newsletter, and receive the free colour Digital Energy Journal.


Latest Edition May-June 2021
May 2021

Download latest and back issues


Learn more about supporting Digital Energy Journal