Practical Applications of Machine Learning and AI in Field Service

Feb 19, 2019 • Features • Artificial intelligence • Future of FIeld Service • Machine Learning • Emily Hackman

In the third article in our current series of articles from field service solution provider Astea focussing on Artificial Intelligence and Machine Learning we now take a look at some of the practical applications...

Is Machine Learning a key topic for your organisation?!

There is a detailed white paper on this topic authored by Emily Hackman and Liron Marcus which is available to fieldservicenews.com subscribers within our premium content library...

1. Dynamic Scheduling

Dynamic scheduling provides optimised schedule and planning of field service agents. It’s constantly updating in response to events in real-time.For example, when a new high priority service ticket is received, it needs to be scheduled for immediate execution and all other tasks must be postponed.

There’s an element of AI in this process because the scheduling system must perform its task of scheduling this service order intelligently. In other words, it must think and thus deduce that simply prioritising this service ticket is not enough. All other tickets must be postponed in order to ensure that the high priority one is resolved first.

For example, let’s say a call has arrived in the system and agents Steve and Karen have the same skills and are both available. The system decides to assign Steve as the agent for the call by considering the nearest agent rule. However, by looking back at history and identifying patterns based on call attributes and agent performance, we find that while Steve is located closer to the customer site and is more experienced, Karen is the right agent to dispatched.

Why? The pattern shows that Karen has a higher probability to resolve this call on the first visit as compared to Steve. This prediction can be done only if the scheduling system reviews historical data and learns from it. We can even find that having more experience and arriving earlier to the site doesn’t necessarily deliver better results.

2. Customer Contact Center

Call center agents play a key role in customer satisfaction. They are usually the front line of the customer’s experience, so it’s crucial to invest in systems that improve the way call centers are evaluated.

Two of the important indicators to assess the efficiency of call centers are “first call resolution” and “average time in queue”. Calls are typically analyzed manually by the agent who is listening to the call and determining the context. Agents also have access to a knowledge base to help them.

Unfortunately, having a human evaluating each call to determine the course of action is time-consuming and does not guarantee the best solution.

3. Service Contract Renewal

Service contracts are one of the most important parts that a field service organisation can ensure ongoing revenue from customers. However, these services are provided for a limited timeframe

only, after which customers have to renew their contract in order to continue receiving the services. It is very critical for the businesses to understand the underlying reasons that impact contract renewal. Traditional ways to evaluate the chance of renewal are based mostly on client satisfaction survey results. However, a lot of important information is captured as free text from emails, notes, problem descriptions, etc. while call center and field service agents interact with customers in trying to resolve service requests.

Using this unstructured textual information, with the aid of machine learning algorithms, we can perform a sentiment analysis to detect a customer’s emotions about the service or specific product. This allows companies to classify the root cause of problems that lead to a customer not renewing a contract.

Machine learning algorithms can perform a sentiment analysis to detect a customer’s emotions about the service or specific product. This allows companies to learn why a customer does not renew a contract.

EXAMPLE: “Your customer support is great, however battery only lasts for 30 minutes.”

When analyzing the sentiments in this sentence, we can classify that the cause that could potentially lead to contract non-renewal is not related to the service, but rather is based on the product itself. As a corrective action, we can propose an upgrade to a newer model in which the battery life has been significantly improved.

With sentiment analysis, companies can respond quicker to signals, track changes in customer sentiment over time, determine if particular customers feel more strongly about their services, and predict the likelihood of contract renewal.

4. Inventory Management

Efficient inventory management revolves around knowing your demands. On one hand, companies must ensure that they have enough stock in their warehouses to meet the ongoing demands. On the other hand, they must preserve a small quantity of slow-moving items. Traditional solutions rely on formulas to reach static sets between pessimistic and optimistic values, and because of that, the chosen buffer level does not guarantee that the stock level fully satisfies the demand.

The way in which data can be incorporated from different external sources, including non-business data, helps machine learning models to find a correlation between the different variable factors and to optimize stock levels dynamically.

For example, feeding weather conditions and inventory data into a machine learning model can identify that specific parts tend to fail when the temperature outside is above 90 degrees Fahrenheit. In case the weather forecast shows that a heat wave is approaching, several proactive actions can be initiated, like ordering a sufficient quantity of items that are likely to break down so that those items can be used for predictive maintenance visits. Essentially, the machine learning model predicts the necessary stock level and determines how much inventory to order by using the past records of weather and failures.

The machine learning model can predict the necessary stock level and determine how much inventory to order by using the past records of weather and failures.