How AI makes our telephones work

March 2018

Desk telephone systemGilbert Owusu FBCS and Sid Shakya MBCS, from BT, lend a telco’s perspective to the process of putting artificial intelligence into operational use.

AI provides an opportunity for products and services to become smarter and more responsive to customer needs, an extremely important factor for any service organisation such as a telco.

BT has been using AI for many years, both in its products and services, as well as in its systems and networks. We see a range of applications for introducing smarter technologies that can increase automation and better tailor products to customer needs.

AI underpins many of BT’s processes. BT was the first European telco to use AI techniques in workforce scheduling in the 1990s, to maximise the efficiency of our workforce. Since then we have extended the use of AI across workforce and resource management in BT, and received a number of awards for this innovative work.

We have also used AI in other areas such as automated network design, process optimisation, cyber-security threat detection and ‘nuisance call’ detection.

Developments and advances

Recent developments in AI include advanced machine learning techniques such as deep learning, mainly used for pattern recognition, classification and forecasting. Fault detection, predictive maintenance, demand forecasting, and service level prediction are some of the applications where BT has used data mining and machine learning techniques to enhance the decision-making process and to provide better service to its customers.

A common thread in these AI-based operational systems is the need to model decision-making for better service and operational outcomes. Our models can be classified into three main approaches. The first approach focuses on using models to analyse past performance so as to gain analytical insights to improve current operations.

We refer to this as retrospective modelling. The second approach focuses on optimising in-life operations, with the focus on ascertaining how best to utilise current resources to satisfy incoming demand. The third approach is what we term prospective modelling and this is used mainly to model future states; it is also appropriate for strategic transformational decision-making.

Case in point, BT has a large and complex field force of over 20,000 engineers. The effective management of this operation is key to both BT’s customer experience and the efficiency of the business. To address this challenge, we developed FOS, a system that is used to plan and coordinate the field force, and draws on a range of AI techniques including meta-heuristics, machine learning and fuzzy logic.

Making a forecast

The field force management process starts with gaining visibility of demand. Forecasting is a complex and tedious task as there are variations in demand types (i.e. new installations, faults, planned maintenance and capital build work).

The scope of demand forecasting may have several dimensions. The entire business space tends to be divided into regions, domains, etc., and the product requirements may have several types. Time is another dimension which impacts the forecast. Forecasting in all these dimensions will end up in multiples of combinations of data.

Once we have visibility of demand, advanced capacity planning is done to make sure that on any given day, we have enough resources to deliver service. The capacity planning process takes into account any expected absences / shrinkages and expected productivity. Finally, workforce scheduling is done on the day to make sure the right skilled technician is assigned to the right task to minimise cost, such as travel, and maximise productivity and customer satisfaction.

A working model

AI techniques are used across the workforce management process. Demand prediction is carried out using machine learning techniques, including neural networks. Capacity planning is modelled as an optimisation problem and solved using heuristic search techniques, such as evolutionary algorithms, that are used to optimise the capacity plan and to make better resourcing decisions. These include decisions such as optimal overtime, recruitment, training, etc. Shrinkage prediction and productivity analysis is also done by analysing past trends using AI models.

Finally, workforce scheduling is modelled as a ‘travelling salesman problem’ and optimised using an array of techniques including genetic algorithm, fuzzy-logic, simulated annealing, particle swarm optimisation and case-based reasoning.

The use of AI techniques in field force management have been key enablers for major transformation programmes in our field engineering teams, which have led to service improvements and operational savings.

Another interesting use case for AI application in telco service delivery is inventory and spare parts management. BT’s large network infrastructure requires constant monitoring and regular maintenance. It has vast amounts of spare parts, both in central warehouses and in local sites.

Ensuring that we have the right number of spares in the right place at the right time is key to delivering service on time and in a cost effective manner. The computational objective here is, firstly, to proactively place the spare parts at sites that are likely to require maintenance. Secondly, there is the need to optimise the location of central warehouses so that travel is minimised in case a spare part is required.

These objectives are, however, not easy to achieve, mainly because of the large variations of the spare parts (over thousands) and the large number of locations (over 5,000 sites across UK) that need to be managed.

An AI-based tool, called Intuitu, has been built to tackle these problems. It models spares management as a combinatorial optimisation problem, and uses machine learning and evolutionary optimisation techniques to produce optimal plans for moving spare-parts across the network, and also produces designs for optimal warehouse locations.

Beware the black box

One of the key challenges to deploying AI in service industries is its lack of visibility on how decisions are made. The ‘black-box’ nature of decision-making can sometimes make it difficult to engage end-users and stakeholders.

To date, we have used a number of approaches including advanced visualisation to provide insights into AI models and trials to address this challenge. One of the areas of focus in AI research is making the AI models more explainable and we are actively working with university partners on this front.

There are many other areas where service operations can benefit from AI technologies. With recent advancement in use of AI for virtual/augmented reality (AR), unmanned aerial vehicles (UAVs) and drones, new opportunities to exploit these technologies are emerging.

Service operations can enhance their provisioning and delivering of products and services with these new technologies, for example, by remotely assisting fault-fixing using AR, or mapping a complex terrain or underground sites for equipment installation using drones, or by using UAV for spare parts delivery for technicians in remote sites. These technologies are key enablers to empowering the workforce.

BT is working with its partners to develop technologies which will help shape new trends in future AI-based service operations. We are exploiting AI-based immersive decision-support systems using mixed-reality technologies to provide recommendations based on real-time contextual information and hands-on training. This approach drives employee empowerment.

Employee empowerment is a powerful management concept that helps to improve morale and productivity. At BT, this concept has been introduced and applied to a subset of engineers by giving them the power to choose their own tasks using a set of tools.

After the introduction of this set of empowerment tools, the travel has been reduced by 17 per cent and the engineers’ utilisation has increased by 10 per cent. The tools are based on a personalised task recommendation system that learns implicitly from the completed task history of individual engineers and recommends tasks similar to their previous choices.

About the authors
Gilbert Owusu FBCS, heads the Business & Operational Transformation Practice in BT, and Sid Shakya MBCS, is a chief researcher at EBTIC, and a principal researcher at BT. Both are members of BCS AISG, and co-chairs of the annual BCS Real AI event.
 

Image: iStock.com/BrianAJackson