How to Apply Predictive Analytics to Post-COVID Recovery, Part 1

FROM THE AUGUST 2020 ISSUE

We are now into the final quarter of a year that will surely be remembered as one of the most chaotic in history. Many of us would like nothing better than to simply fast forward to 2021 or beyond. But how to get there? As contact center leaders plan for the post-pandemic recovery—both short- and long-term—where can they turn for reliable insights for informed decision-making? The pandemic outbreak created volatile market conditions and unpredictable customer behavior. What impact do the unique circumstances that businesses experienced in 2020 have on the reliability of recent customer data and predictive models? How will it influence strategic planning for the post-pandemic contact center?

To learn the answers to these questions, I reached out to several predictive analytics experts in the contact center industry. For this article, I wanted to zero-in on three areas key to strategic planning for the post-COVID era—collecting and using customer data, human resources (hiring, performance management and retention), and data security.

Providing insights and advice on the data challenges caused by erratic customer behavior and changing market conditions during the pandemic are: Daniel Foppen, Senior Principal Product Manager, Oracle CX Service Strategy; Jeff Gallino, CTO and Founder, CallMiner; and Alex Martinez, Senior Product Marketing Manager, OpenText.


Customer Data: Is It Reliable; Best Use Cases for Recovery

Q. How can companies evaluate data and predictive models that may no longer be reliable due to erratic consumer behavior and rapidly changing market conditions during COVID-19?

DANIEL FOPPEN: Predictive models are based on past patterns. But many patterns have abruptly changed over the last six months during the COVID-19 pandemic. To adapt to these changes and avoid making inaccurate predictions, predictive analytics or machine learning (ML) systems can be adjusted or redirected by feeding them a new set of learning data directly from this time of change.

That said, the question remains whether companies should base any forward-looking strategic actions based on the data, patterns and behavior from the last six months. True, the situation may seem dire today in countries like the U.S., Brazil and Russia. For example, in the U.S., the unemployment rate is running into the double-digits, whole swaths of the economy are standing still, and government support checks are likely to run out at some point. However, we are experiencing a downturn created by a pandemic, not a structurally misaligned economy that will lead to a prolonged depression. When we look at how other economies across the world are cautiously picking up pace after suppressing the virus outbreak, we should correspondingly plan for gradually returning to normalcy, especially as vaccines become available. Some industries will recover faster than other industries.

Thus, when thinking about using data and predictive models, I would recommend identifying different scenarios and timespans for planning. In the short-term, consumer behavior will be similar to what we have seen over the last six months. In the long-term, it will be somewhat similar to pre-COVID-19 times with some common-sense adjustments (e.g., curbside pickup, digital experiences, etc., will be more common than before). How long the transition lasts between these two phases will depend on the industry. 

JEFF GALLINO: Customer data and insights shift by the month, week, even day—and brands got a wakeup call on its ever-changing nature during COVID-19. The most crucial action a company can take to collect reliable customer data is to keep a pulse on 100% of conversations, 24/7, with evolving contextual analysis applied in real-time through predictive analytics.

For example, many brands are only analyzing 1%–3% of their customer interactions. Because of this, they receive a very limited and skewed vision into consumer insight. With the right tech, companies can listen to and analyze every single interaction—widening their data pool for a more accurate depiction of their audience. With contextual analysis, brands can assess location, social and economic factors that may affect interactions, which is especially important throughout this pandemic.

The key to finding actionable insights through this plethora of data is the ability to sift through endless amounts of noise to discover trends, common issues and changing preferences instantly. The right tech is imperative, as the volume of analysis and level of categorization goes beyond human capability.

ALEX MARTINEZ: COVID-19 changed the way organizations gather and look at customer data. In the prior face-to-face and transactional world, customer data was readily available and gathered. However, now customer data is scarce and heavily gathered by titan eCommerce players, like Amazon, Shopify and Walmart eCommerce, as consumer buying has shifted online across all of demographics.

Today’s stay-at-home reality demands for new or recalibrated predictive models and forecasts as the old models may be unreliable. Companies need to get creative to find new sources of customer data. This may require incentivizing customers to engage in things like a newly minted organizational digital zone. Here, customer insights can be gathered from real-time analytics across all customer touchpoints, whether structured or unstructured (e.g., voice).  Also, incorporating Voice of the Customer (VoC) programs can enable organizations to identify the latest trends. Finally, social media data has become more important than ever as organizations can no longer rely only on point-of-sale data.  


Q. Which other types of feedback or data can provide timely insights into changing customer behaviors to help guide decision-making?

DANIEL FOPPEN: First, let’s look at digital analytics. Digital analytics is the best tool to keep track of digital customer behavior across different devices. A tremendous amount of insight can be obtained from how customers use apps, what they do when they visit companies’ websites, what they click on, what they like, what they download and so forth. This proves especially telling when we can look at the data in real-time, combined with identity resolution.

It’s important to note that most digital experiences happen when the customer is not logged in—meaning they are unknown web visits. A lot of generic insight can be obtained from unknown web visits, but if those visits can be tied to an identity, it will be more powerful. In doing so, a brand can generate a single digital behavioral profile of a customer that includes data from different devices that the customer owns, web visits on those devices that lead to subsequent purchases (e.g., in a store) or the absence of a subsequent purchase, and digital behavior tied to customer service visits and more. 

We can also add data from customer-owned IoT-connected devices to the mix as well. Think about an IoT sensor indicating an impending failure and combining that with the information that the customer owns a range of expensive devices and is a big spender with the business. For that customer, it may make sense to proactively reach out to offer a replacement part or service.

There many more different sources that may provide clues to customer intent and behavior, including feedback management, social networks, third-party (ad-tech) data market places and so on. With so many different potential sources of customer data the challenge for businesses is not just to identify these sources, but instead to consolidate all these different formats, unstructured and structured data flows and source systems, into a cohesive dynamic customer profile. When interpreted, this data can be used in real-time to activate and tune messages, respond to signals and personalize interactions across the entire customer journey.

Data lakes have been touted as the solution for this problem over the last decade or so, but data lakes come up short because they import data in batches, e.g., over the weekend. This may serve the needs for the IT persona, but when the customer data is not available in (near) real-time it becomes difficult for the CX professional to do something useful with customer insight, especially in today’s fast-changing environment.

At Oracle, our point of view is that the only viable way to guide decision-making in our dynamic landscape is with a Customer Intelligence Platform that both consolidates a broad variety of data sources around a dynamic customer profile and also makes that information available in real-time. In doing so, a business can act intelligently and agilely.

JEFF GALLINO: One of the core values of predictive analytics is its ability to extract valuable insight that powers better, more strategic decision-making and process improvement across the entire enterprise. This includes employee performance and experience, sales, marketing, operations, product development and more.

Emotional and sentiment analysis derived from analytics can reveal the types of questions your customers are asking, the topics they care about, and what they want from your brand. This intel can uncover the unknown, identify root problems and guide the entire enterprise. In uncovering insight more rapidly, organizations can make better, more informed decisions on staffing needs, product deficiencies, messaging shifts and more. The most mature organizations operationalize the customers insights for strategy, company-wide performance improvement, and bottom-line value.

ALEX MARTINEZ: Point-of-sale customer data has been the bulk of the traditional pool of data. Now, organizations must expand and collect customer data from different sources. This can include conversational data that occurs, beginning with the call center or sales 1-800 number. It continues with the shift to online engagement happening across generational thresholds due to COVID-19.

The type of data—either structured or unstructured—is also very important. Structured customer data can be easier to capture across multiple channels, for example, via website peer review or social media because digital text is digital text. Unstructured data, on the other hand, is harder to capture. It includes data in the form of video or voice information and requires a layer of additional technological capabilities to capture that information in a usable format to be processed by analytics tool. Other digital systems of engagement can be used as a reliable source. For example, interactive chatbots can provide a wealth of customer sentiment, feedback and behavioral response patterns.


Q. What are the best use cases for predictive analytics in the post-pandemic business recovery?

DANIEL FOPPEN: The current crisis offers many challenges—but also many opportunities—to change the way we think about processes, approaches, culture, technologies and more. We strongly encourage businesses to reflect on inefficient processes and use the COVID-19 crisis as a pivotal moment to rethink the way things have always been done.

For example, the pandemic has encouraged the adaptation of new business models that place service at their core. One of the most promising changes involves predictive customer service, turning the very basic paradigm of service upside down. The idea of service has always been a reactive model. A customer has an issue with a product or service and will contact customer service to get the problem solved. 

Today, however, there is such a wealth of information available to businesses that customer service teams can predict if a customer will have a problem before the customer even knows it. Imagine how amazing it would feel to receive a call, a message or an email from customer service to inform you that you were about to have a problem, but that it has already been fixed? Smart organizations are figuring out how to differentiate from their competitors with predictive service models, and predictive analytics plays a huge role in enabling this transition.

Let’s take that idea a step further. Imagine a business that sells IoT-connected industrial devices to small- and medium-sized businesses. Traditionally the vendor would sell these devices as one-off purchases with a big manual, sending prime material every month, taking calls when the products need fixing, and after a few years, sell a customer a new device. A predictive, service-centric business model flips this idea on its head. Instead of selling these devices, the vendor offers them with a monthly subscription fee—a hardware-as-a-service model. Instead of printing a big manual, a trainer goes on-site to teach staff. Prime materials are shipped and replenished automatically when IoT sensors indicate low levels, and customer service is actively monitoring for impending errors announced by predictive analytics that reports on sensor data. If the predictive analytics indicates a likely failure, the service team can proactively ship a replacement part and schedule a technician for installation. A model like this it is in the best interest of both the customer and the vendor, as it maximizes usage and uptime of the devices. A win-win.

Even if a business is not dealing with IoT-connected devices, we still believe that predictive analytics can drive similar innovative business models.

Due to the wealth of customer data available, there are ways to derive to similar insights. For example, take a company that sells simple mechanical devices with a locking mechanism. Say the company sold 1 million of these devices over the last year and 1% of those customers contacted customer service with a very trivial question of how to unlock the locking mechanism. At an estimated $10 a call, times 10,000 calls, it costs the business $100,000 to answer that simple question that customers could have solved by looking at the FAQ section on the website. So, could a business prevent these calls using predictive analytics? That depends on the data.

When you combine tools like Oracle Data Cloud’s audience analytics you can actually compare the 10,000 customers that called with 8 billion other profiles. Here, you can find all the commonalities that allow your business to create highly accurate customer segments that are very much alike. Using this predictive analytics segmentation data, a business can then proactively reach out to all the customers that bought the devices who will most likely contact customer service to ask how to unlock the device with the solution to that question. This is another example of solving a customer’s problem before they know they have a problem. We at Oracle are very excited imagining what that will do to customer loyalty and how experiences like this will be driving new, innovative business models in the future.

JEFF GALLINO: Recovery from the pandemic will be a long and rocky road, and we’re sure to experience a number of prominent changes in consumer behavior throughout this new normal. Because of this, some of the best use cases for predictive analytics will involve the ability to identify the customers still vulnerable in the wake of the pandemic. Through assessing vulnerability by identifying keywords, phrases and acoustics, agents can be guided to respond to consumer stress levels more appropriately and empathetically. In measuring vulnerability, predictive analytics can also alert organizations about a customer’s propensity to pay. The pandemic has put many people in a tough spot financially and recovery won’t happen overnight. The ability to extract customer insight into future payments will not only assist in developing better collection strategies, but also help your organization mitigate compliance risk and losses to increase recovery rates.

Overall, in order to make the biggest impact, it’s crucial to use tech that goes beyond discovering trends based on human-fed searches. The best predictive solutions build off initial hypotheses to automatically identify related triggers that brands should analyze and take action on. This is how organizations can facilitate more efficient and complete engagement analytics, enabling organizations to swiftly drive enterprise-wide value and bottom-line performance improvements.

ALEX MARTINEZ: There are numerous and ever-growing applications for predictive analytics in today’s business world. Multiple industries have very specific needs. However, there are some uses cases that can be applicable across multiple industries:

  1. Customer profile and segmentation for an enhanced and more precise match-up of product and services.
  2. Identification of consumer behavior and buying power.
  3. Voice of the Customer, and subsequently, buying decisions based on customer satisfaction, current sentiments, Net Promoter Scores (NPS) and more.
  4. Increase of customer wallet share by identifying preferences, capacity to buy, and predictability of what they need. This can lead into more cross-sell or upsale opportunities.
  5. Predicting product inventory and consumer demands based on macro and micro economic information data.

Check out Part 2 of this series for insights on agent hiring, performance and retention issues, and keeping data secure in a work-from-home environment.