Data-driven decision making demands a holistic view

Communications service providers (CSPs) are now well aware of the power of the data they collect regarding their customers and see it as a key means to run their operations more efficiently, achieve compliance, improve customer experiences and ultimately enhance their profitability. The telecoms industry, however, is composed of a complex web of organisations created through mergers and acquisitions (M&A) and infrastructure that involves multiple generations of technology. This makes the process of enabling data-driven decision making multi-layered and encompasses cultural as well as technological challenges.

Mel Prescott is a principal consultant in the telecommunications practice at FICO, the data analytics specialist. He helps companies to use the most advanced analytics, decision rules, orchestration, mathematical optimisation and other systems and technologies so they can achieve the potential to optimise all aspects of their operations that lies in their data. He has more than 15 years of experience working in analytical telecoms roles to provide subject matter expertise and has held credit risk management positions at EE, Orange and Bank of America. He holds an MSc in Analytical Credit Risk Management from Sheffield Hallam University in the UK.

Here he tells George Malim how FICO is helping its CSP customers through the use of big data and mathematical algorithms to predict consumer behaviour. The company focuses on enabling CSPs with tools and software that allow them to optimise operations through fighting fraud and managing risk more effectively, complying with government regulations and creating more profitable customer relationships.

George Malim: What are the challenges communications service providers (CSPs) face in making effective use of data to drive their decision making?

Mel Prescott: We’ve been talking about use of data and ability to create new solutions by exploiting data for a long time but the reality on the ground is that a lot of CSPs are hamstrung by operational silos. They are still having to process data in large batches rather than in near real-time.

The data is sometimes inconsistent and different departments have different data warehouses. They are sometimes protective of these and suspicious of other business functions using that data. This means there isn’t the holistic view of the customer that is needed to make effective data-driven decisions.

For example, when I worked at EE in the UK, we started as Orange, then merged with TMobile. We had no way to link a broadband customer with a mobile customer, or an Orange mobile customer that also had a TMobile mobile subscription, or to provide seamless experiences.

There’s an old Indian parable about blind people in a village being asked to identify an animal by touch. The animal’s an elephant but because the people all operate in isolation they can’t provide a coherent description of the animal. CSPs’ fragmented structures and the barriers between departments result in the same incapacity.

GM: How does the process of collecting and analysing the data to inform the outcome work in practice?

MP: There isn’t one single answer that brings all the data from disparate systems together. Some CSPs, like Telefonica, are further along the path of platform reinvention than others. They’re going down the path addressing one use case at a time and as each new one comes up they can decide whether to ingest it into the platform.

CSPs have been investing in data lakes and big data infrastructure for several years now but once you have the data, what does it mean? Just having the data is one step but then you need to transform it into something useful – the information. This is where the power of analytics comes in.

What is useful, what analytics can be applied and then how to operationalise the insight created is enabled by decision management platforms. These help you to make use of your data.

GM: Is it too soon to envisage automated decision making enabled by data analytics allowing CSPs to transform their personalisation and marketing capabilities?

MP: The vision is heading towards automated decision making but there is definitely a nervousness about letting machines run the network. To let a virtual function take over without a human touch is daunting and one reason for that is that there are numerous examples of bias creeping into machine learning models if they have not been trained properly or have issues with data selection.

Last year, we saw controversy when a credit card provider was accused of gender bias when users noticed it offered smaller lines of credit to women than men. Higher credit lines were given to men because historical data introduced that bias into the business model and kept on perpetuating it.

In future, without doubt, regulation will come in to ensure models have to be explainable and in FICO we’re at the forefront of that effort but it is difficult because the more predictive the algorithms the more difficult they are to explain. This is an emerging area of data science, but it is necessary and FICO has been innovating in this area for some time. FICO has dedicated resources to the refinement of Explainable AI – new types of AI and machine learning models that make it easier for business generalists to understand what is going on inside the black box, and to satisfy regulators that decisions are made ethically through these models.

GM: What do you see as FICO’s role here?

MP: We’re a well-established company. We’ve been operating since the 1950s having been founded by a mathematician and an engineer who figured out they could use maths and science to predict outcomes. They built a credit score based on this which was pioneered across the US and is ubiquitous now.

We have a heritage in pioneering advanced analytics which underpin our Falcon fraud solutions. Two-third of credit transactions worldwide pass through Falcon. It uses artificial intelligence (AI) and neural networks as well as other technologies to spot anomalous behaviour. We will flag this up to the credit card issuer and they can make a decision on what action to take. Falcon utilizes global consortium data which allows the models to learn patterns of behaviour from all of the Falcon users and makes the overall data pool much broader. This is a huge advantage when trying to find fraud.

GM: Is the real meaning of AI getting diluted by the hype?

MP: There has been a lot of talk in the last two or three years about the adoption of AI and machine learning but these aren’t that new and we’ve been doing both for more than 20 years. AI from some of the newer companies gets badged as AI when it’s just data analytics. This makes it difficult to cut through the noise.

The revitalised market for AI and machine learning has spawned hundreds of new venture capital backed start-up companies. But we’re one of AI’s most innovative players and invented many of the tools and methods that make AI effective.

FICO’s core business is in helping businesses make decisions by assessing risk. AI was a natural place for FICO to invest, and has grown into a core competency. The company has a large and growing patent portfolio in AI and machine learning, and a long history of operationalising AI. The technology is taking FICO and its clients beyond its base in risk-focused applications. We’re good at this and proud of our capabilities. The combination of AI and machine learning with a decision management suite enables a modular approach that can be deployed across the entire CSP. This means you can operationalise the decisions and outcomes at scale. That could be in the call centre to help customers to get to the answer they’re looking for in just a few steps or it could be for credit management, loyalty and retention teams, collections and recovery, network operations and supply chain. The list goes on.

GM: Where does this end up? Can you see data being applied across multi-party business models to enable far more granular, automated decision making even between the different enterprises in a business chain?

MP: There’s no doubt that this kind of ecosystem is becoming more and more complex. CSP business models are evolving and opening up of application programme interfaces is happening. Huge amounts of data are being transferred at any one time which makes it difficult from an infrastructure point of view.

The really advanced companies are therefore looking at streaming data in real-time. This takes enormous processing power but, if you’re able to look at data as it passes through systems you can do a huge amount with it.

For example, being able to target a customer with the right message, at the right time and in the correct context means you are far more likely to have a successful outcome – whether that’s an upsell or an enhanced experience. However, that’s only possible if you’re able to stream and process large amounts of data and deliver insight back to the customers or service providers in the ecosystem at high speed.

We are working with a major mobile money provider in sub-Saharan Africa to develop credit scores for the unbanked population. This enables huge volumes of GSM data to be analysed, alongside the transactional data that comes through the mobile money service, in order to create predictive models to be able to offer micro loans to customers. Given the huge volume of unbanked people in these countries, this will open up access to credit for people previously excluded from this system.

Another current example is national authorities working with CSPs to analyse location data. This is anonymised and aggregated, but it enables authorities to monitor and track population flows in times of natural disaster, pandemics or other forms of crisis. Having access to multiple CSP data sets, potentially across national borders, has its own challenges in terms of data structure normalisation, access and ability to process in near realtime. However, it can be a very powerful asset to enable governmental decisions and inform strategy.

We also work with a European CSP that provides wireless internet to large rural areas where it is cost-prohibitive to implement cable or fibre. This CSP uses FICO optimisation solvers to route wireless traffic through the best possible pathways so it can meet the bandwidth needs of customers. It is able to engineer the traffic flowing through the network and, taking into account different weather conditions and daily variations in consumer traffic, the CSP can predict and react to demand and redirect bandwidth accordingly. This ensures a smooth user experience and optimises the way the network routes traffic.

Having the ability to connect the different decisions that different parts of the business are making is the key to this improved ability to drive further datadriven decision making across first a single business and later those involved in an ecosystem or business chain. However, there is some distance to go before the concepts involved in this are fully mature.

There is very little understanding right now that if I make a decision based on data within the loyalty and retention department there will be a knock-on effect in other parts of the business. Understanding how decisions impact across the entire business is a vital next step in educating organisations and users.