Predicting service disruptions really works – now let’s talk analytics algorithms
Most communications service providers (CSPs) rely on customer complaints (usually calls to the help desk) to learn about service disruptions because existing systems, like network operations support systems (OSSs), cannot easily identify issues with the access network, customer premises equipment (CPE) or applications. According to industry analysts, around 70% of the service problems that residential customers report to operator help desks originate from the access and home network. Digital Subscriber Line (DSL) technology, for example, causes an average of three trouble calls/ year/subscriber, at least half of which come from the customer premises. Up to 50% of all trouble calls are related to Wi-Fi; a vital element of the connected home experience.
Analytics capture large amounts of data from the network, CPE, trouble tickets and more. Through analysis of this historical data, algorithms can be developed to better predict service disruptions and take proactive actions to address issues before the customer notices or calls in.
Getting a heads-up on large service disruptions
Although it would be ideal for the network to always know when service degradations occur, it’s often the spike in customer calls that is the first indication of a problem. Based on this reality, the researchers at Nokia Bell Labs had an epiphany; why not correlate customer ticket records based on past service disruptions with network topology and services to detect (in real time) when outage spikes are starting to happen and what the offending network entity or service is? The result is a “call anomaly detection” algorithm.
The process starts with an examination of all calls received by customer service representatives (CSRs). When a sudden burst of calls (a spike) is identified, caller IDs are correlated with other network, service, and device data. Using real-time statistical signal processing algorithms, calls concerning possible service disruptions affecting multiple subscribers are categorized and separated from other calls being received. We are all too familiar with outages that popular services like Netflix and Facebook had in the last few months. This call anomaly detection can detect the outage within minutes and update interactive voice response (IVR) systems to play the appropriate message, thereby ensuring the call centers are not overwhelmed and can stay focused on solving the real issues.
Good enough is no longer good enough
The number of calls handled by your help desk represents just a small fraction of your dissatisfied customers. Even though most people don’t bother to call the help desk when there is a service disruption, the ones who do call drive significant call center costs and the service outages themselves can cause customer dissatisfaction and eventual customer churn.
Simple problem detection – which is characteristic of many descriptive analytics solutions – can be helpful in responding to service disruptions, but it is reactive in nature. Reacting to a service outage after it has occurred – no matter how thoroughly – is no longer good enough.
Predictive analytics address customer issues before they arise and go one step further – by providing management systems with what they need to automatically and quickly prevent service disruptions. To identify such issues a combination of algorithms are used – for anomaly detection, for ticket correlation and for domain-specific analysis. Examples of domain-specific algorithms include – algorithms to predict CPE failure, analyze home Wi-Fi data to predict poor Wi-Fi interference or coverage experience and algorithms to detect DSL line instability. This is where data science meets domain experience.
Going the extra mile: from reactive to proactive
Once an anomaly is identified or predicted a recommendation algorithm is able to trigger several proactive actions.
Examples of actions that can be initiated by the algorithm include:
- automatically re-configuring or rebooting the CPE before the customer notices an issue and calls the help desk
- optimizing the Wi-Fi configuration of a group of access points
- reconfiguring the DSL link, using dynamic line management (DLM), to find the optimal stability and speed of the line
- notifying customers of the problem and providing instructions/information
- shipping replacement equipment or mobilizing field technicians to undertake proactive repair actions
- annotating the customer’s account information, so that CSRs are aware of potential problems in case the customer calls the help desk
The algorithms are constantly tested, updated and refined using real-time data from the network and customer calls. Implementing the actions listed above using workflows also provides the CSP with the flexibility to adapt the action according to their desired business processes.
It’s all worth it
When customer care improvements are made – that is, issues are detected and proactively addressed before the customer is aware of any problem – agent-assisted care is more on point, self-care is more effective and help desk calls and field technician deployments are reduced, resulting in significant cost savings:
- Call Anomaly Detection can lead to an 85% reduction in help desk calls related to network outages and a 90 percent reduction of unnecessary truck rolls.
- A Dynamic Intelligent Workflow system leads to 5-15 % reduction in Average Handle Time (AHT).
- Dynamic Line Management algorithms lead to more than 15% reduction in customer calls and truck rolls in just a few months.
- Wi-Fi Self-Optimization Network analytics leads to a 10% reduction in Wi-Fi agent support calls.
More reliable service and support also results in improved customer perceptions and decreased churn rates.
Watch how it works in this video: Analytics algorithms powering customer care
Don’t miss our upcoming blog: “The technology behind improving customer care”, where we describe how actions are orchestrated using workflows in more detail.
Share your thoughts on this topic by replying below – or join the Twitter discussion with @nokianetworks using #CEM, #CSPCX