Monday, December 9, 2024
HomeCustomer Experience NewsPitfalls To Avoid When Applying Predictive Analytics For CX

Pitfalls To Avoid When Applying Predictive Analytics For CX

Predictive analytics can be a powerful tool for your business as it will help you anticipate customer behaviors and align more closely to their expectations. This asset can continually improve your customer experience (CX) in real-time. You will also be able to preempt any problems in a customer journey, meaning reducing customer issues and the growth of their loyalty. 

However, an effective predictive analytics system can be prone to pitfalls that prevent your business from getting valuable CX insights. Here are six pitfalls to avoid when applying predictive analytics for CX. 

Not Linking Predictive Analytics With A Problem You Need To Solve 

Many individuals feel that predictive analytics can help them automatically find a CX solution. All they need to do is collect the data and then using an intelligent tool to analyze everything. However, this isn’t the right approach, as using predictive analytics could reveal useless patterns that your company cannot use to improve business outcomes or CX. 

John Dumo from Softchoice also considers this a pitfall as he states that you should never assume that predictive analytics will do all the work for you. He says, “ Be clear about what business problem you want to solve and what data will be needed as an input to solve it. Machines do not understand your business or its objectives.”  

Similarly, Nilakanta Srinivasan from Collaborat also feels that no one should treat predictive analytics like a magic wand. He states, “ Building predictive models without [a] good fit to customer needs or issues and without strong linkage to the overall CX roadmap and VOC [Voice of the Customer] can lead to highly accurate predictive models that have low utility. So always choose the right problem and define the scope.” 

Instead, before using an intelligent analytical tool, your business needs to have a problem to solve. Think about what you want to understand about the future by analyzing the past. You will also need to consider how you will use the predictions you gather. Will you make any decisions or take any actions as a result of the data? 

In other words, when using predictive analytics for CX, you need to know how you will use these insights to improve business outcomes. These outcomes could include:

  • Increasing customer retention as a result of your company delivering good customer service by solving problems quickly.
  • A decreased cost of serving customers as a better quality of customer experiences means fewer issues.

You can start tying CX strategies to a tangible business value by reviewing major sources of opportunities and pain points across different customer journeys. Then, think about how using predictive analytics can enhance existing solutions or create new ones to improve business outcomes like customer loyalty or cost to serve. 

Not Having A Central Database of Customer and Business Data

Different business functions often have their own goals and priorities in terms of their department and often work in siloes. However, when these departments work alone and don’t share their customer and business data with the rest of the organization, they cannot make business decisions in line with CX.

This is because one department’s data represents only part of a customer’s journey. Alternatively, a business that encourages each team to pool their respective data sets into a centralized hub is more valuable as analysis of the entire customer journey can occur.

Research from the Harvard Business Review (HBR) supports this. It found that cross-functional functional teams armed with frequent, real-time data and analytics are the key to driving experiences that result in stronger customer loyalty and retention. 

To prepare a more suitable environment for using predictive analytics for CX, your business needs to start by creating an integrated, comprehensive database of customer, financial, and operational data. These data sets should contain data on individual customers and aggregate data. 

In this way, it’s easier to track your entire customer base and the associated customer journey when it contains all of your business’s datasets across transactions, operations, and interactions. 

As Dr. Kashyap Dave from Miele states, “ Having a central repository of all customers [and] capturing [the] points and preferences of these customers [is] very important. [A] 360-degree view of the customers can be used to sell, cross-sell, service [and] market.”   

For instance, Air BnB discovered that they couldn’t make accurate business decisions when departments worked alone. So, they moved to a system where sub-teams of data scientists partnered with engineers, designers, product managers, and marketers. Doing this encouraged all teams to learn from each other and solve problems that address a customer’s actual needs.   

Treating Data From Every Channel The Same

While all customers need to receive the same level of service, it doesn’t mean that they should be treated the same by a customer service department. This is because each customer is unique depending on their age, preferences, and the channel they use. 

Research by Zendesk shows that customer expectations do vary by channel as 51% of consumers expect a response by phone in under 5 minutes compared to 28% do by live chat. More millennials/Gen Z customers prefer to resolve their issues by chat than baby boomers/silent gen. 

This highlights different customer journeys across your organization, depending on the channel used, meaning that you should treat customer data from every channel differently to improve CX. 

Unfortunately, Rashmi Bhambhani from Summatti highlights that this isn’t what businesses always do in practice. 

She notes, “CX has a broad challenge of determining what customers need, what their issues are, and whether or not these criteria were satisfied with a customer support agent. The easiest trap to fall into with any team is treating every channel the same, “

Rashmi continues to say, “The language from chat, email, to phones varies so greatly that if you apply the same generic analytics to each of these, you’ll find agents won’t be able to derive meaningful insights from the feedback they can get from these powerful tools.”

As events in a customer journey by email will be very different from those through chat, your company should use predictive analytics to generate CX insights for a customer’s journey, depending on the channel.

By treating customer data differently depending on its source, you can generate specific scores for individual customer satisfaction and business value outcomes like loyalty, cost to serve, and revenue. Additionally, using predictive analytics to analyze data by its channel can segment your customer base into groups based on their attributes so that they are offered a more personalized CX.     

Lack Of Data Preparation Before A Predictive Modelling Exercise for CX

Once you’ve identified the business issue you need to resolve, you also need to prepare a training dataset for your predictive analytics software to analyze. But, many individuals might not understand that they need to cleanse and prepare data before a predictive modeling exercise. 

Not only is this type of preparation a time-consuming process, but creating this type of dataset requires the skill of someone who understands both the business problem and the data an organization has. 

As Nilankanta Srinivasan from Collaborat also says, “Once a relevant problem is selected, most organizations fail to take into consideration the complexity associated with input data such as its quality, scope, and quantity. [So] preparing the training dataset is in fact the biggest challenge and impending factor for the success of predictive analytics in CX.”

Likewise, David Carmell from DealRockit agrees that challenging input data and determining what you want or hope to achieve from a predictive analytics project is challenging. “Frankly, at best, predictive analytics is a can of worms only as good as the bias, hard work, and granular detail once is willing to put into [getting] truly predictive results.”

Your business needs to establish how much training data you need before you begin the predictive analytics exercise. In general, the more high-quality training data you have, the better your CX insights will be. John Dumo from Softchoice agrees with this statement as he feels that you should never choose to work with smaller datasets. 

“Simply put, more is more. These [predictive] models feed off lots of data and get better as you add more. Small datasets can lead to broad variability and deliver models that are unreliable,” he says. 

Next, if individuals do not prepare or clean their training dataset to ensure quality and consistency, a business could use bad data for the modeling exercise. To ensure that your training data meets quality and consistency standards, you could check whether all records like dates or monetary values follow the same format or style. In addition, you might also have to break down complex values like large comments into simpler ones like keywords.

Moreover, John Dumo from Softchoice highlights other issues that make your data bad which will compromise your CX predictive analytics exercise if not rectified:

“Bad data can mean that you have inaccurate, incomplete, or missing information. Perhaps not all of a customer’s transactions are present or specific fields, or characteristics about an account, a product, or transaction are inconsistently filled in by users.”

He continues to say, “Are you sure you have their sentiment properly categorized? Perhaps you have several special cases or outliers you did not take out of the model.”

As you can see, the list of data preparation techniques mentioned in this article is far from exhaustive. But, the main goal of this process is to ensure that this data gets the most accurate and efficient CX insights from the predictive analytics process.

Ignoring The Human Element

Although predictive analytics software can find patterns among large amounts of data, it cannot interpret these patterns according to human behaviors. Decisions about which patterns to employ, the relevant sources, and presenting the predictive findings are human judgments. 

Like Coreen Merryweather from Blue Prism says, “It is important that you view this data only through the lens of what you are trying to achieve, and then complement it with human experience and insight.”

With the above in mind, it can be tempting to simply hire one or a selection of new employees who can decipher what your data has to offer and the business outcomes you need to achieve. However, predictive modeling for CX requires a broader team approach. 

In other words, it is better to build a collaborative team made up of people who collectively possess the necessary skills. You will need data scientists who can develop and apply complex analytical modeling and creative designers and developers to ensure that the final product or service is user-friendly and aesthetically pleasing. 

Additionally, IT architects will help provide technical insights while those in leadership can connect the departments of IT and analytics with management and business decisions. For instance, Intermountain Healthcare used human-focused analytics to drive better healthcare outcomes by ensuring that each of its frontline clinical departments has its own data teams. 

This meant that its analytics and frontline staff could work together to improve their knowledge of analytical processes and encourage them to act on any new insights. Additionally, it also meant giving doctors training on data processes to consume and understand analytical findings. 

Not Regularly Updating Your Predictive Analytics Model For An Issue

Once your predictive model goes live, it can be appealing to forget about it and move onto your next issue. However, any predictive model is based on business-as-usual assumptions and historical data based on a real-time environment at that time. 

In addition, the models created before the pandemic were built in an environment that was stable. This means that your business cannot accept the results of a historical model without ensuring that it is still accurate in line with the current business environment and current customer needs. 

Research by Capgemini indicates that there were changes in CX expectations as a result of Covid-19. For instance, once the pandemic is over, 54% of customers have said that they will prefer local or regional products. Michael Hinshaw from McorpCX also echoes the danger of only focusing on historical data in today’s climate and not regularly updating the data that your predictive model relies on. 

He says, “The reality is, our customers, our markets, our channels, and our competitors continue to change at a breakneck pace. One only has to look at the events of 2020 to recognize the truth of this.”

“Applying a predictive model based on past results in an uncertain world can easily lead to unintended and undesirable outcomes. Suffice to say that predictive analytics is not a quote set it and forget it solution. They require continual, thoughtful, and well-informed optimization to deliver on their CX promise,” he adds.

For example, by embedding analytics within their business processes, Intermountain Healthcare created a learning loop that encourages new data to be shared. In this way, sharing new data on every treatment of heart attacks as part of its data-driven process, helped reduce the average treatment time from 90 to 57 minutes.  

Wrapping It Up

It’s easy for businesses to make mistakes when adopting new technology, especially as pitfalls aren’t easy to see in advance. When it comes to applying predictive analytics for CX, you need to start with a problem you need to resolve. Finding this problem requires a complete understanding of the customer journey achieved by encouraging all employees to pool their departmental data into a central database.

It is also essential to treat each channel differently as a customer’s journey can vary depending on the channel. Before you begin the predictive modeling exercise for your CX issue, you need to prepare and clean your existing dataset for this purpose. Human insight is imperative here as the process of cleaning and interpreting data in line with customer expectations is a team effort between data scientists, IT professionals, and leadership. Lastly, once you create a predictive model, it needs to be updated with new CX data to meet customer expectations continually. 

About the author

Efrat Vulfsons
Efrat Vulfsonshttps://www.prsoprano.com/
Efrat Vulfsons is the CEO & Co-Founder of PR Soprano and the editor of CXBuzz parallel to her soprano opera singing career. Efrat holds a B.F.A from the Jerusalem Music Academy in Opera Performance.

RELATED ARTICLES

Gartner Disunited Engagement Channels To Result In Siloed Customer Experience

Gartner Disunited Engagement Channels To Result In Siloed Customer Experience

53
Omnichannel Customer Engagement Protects Critical Business Revenue Gartner reports that despite all the efforts to push companies to combine their customer-engagement channels, half of the...

Verint Announces New AI Real-time Agent Assist Tool

59
Verint, the customer engagement platform, recently announced their newest product addition, AI-driven real-time agent assist capabilities added to their Verint Customer Engagement Cloud Platform. The...

Most Popular