Using AI in CX: Why You Need High-Quality Customer Experience Data

Authors: Stephanie Caravajal & Alex Champagne

Does AI have value in Customer Experience (CX)? We resoundingly believe it does, however, it must be implemented the right way. What do we mean when we say the ‘right way’? Only by using quality data as the foundation of AI can we trust the results and make informed decisions. 

What are examples of AI in CX? 

There’s no question that AI has many uses in CX. Some common examples include: 

• Chatbots and Virtual Assistants: AI-powered chatbots and virtual assistants can provide instant customer support, answer common questions, and assist with basic tasks, ensuring 24/7 availability and reducing response times. 
• Sentiment Analysis: AI-driven sentiment analysis tools assess customer feedback and social media posts to gauge customer sentiment and identify trends, helping companies respond to issues or capitalize on opportunities. 
• NLP (Natural Language Processing) Models and Text Analytics: AI can automatically analyze large volumes of customer feedback, reviews, and survey responses to extract actionable insights, saving time and providing valuable information for improvements. 

The above list is not exhaustive; many use cases exist for implementing an AI model for CX Analytics. However, we must be careful when building these models and interpreting their results. 

Common Pitfalls of AI in CX 

Failure to interpret model results carefully can cause false assumptions and poorly informed decision-making, even by senior leadership. This is especially true if the data used to train the model is bad. Bias, lack of randomization, and many other issues can lead us down a slippery slope.  

We must use caution when extracting insights from any model, as results could reflect hidden biases. For example, survey data is often inundated with biased results, including non-response bias, selection bias, order bias, and acquiescence bias (i.e., when individuals are likely to agree with something regardless of how they actually feel). These are often coupled with leading questions and poor survey design.  

An AI model cannot separate biased data from unbiased data, therefore training them on biased data will only perpetuate issues and provide inaccurate recommendations. 

Another common pitfall is relying solely on AI. Depending too heavily on AI can lead to a lack of ‘the human touch’ in CX. Companies should ensure that AI complements human efforts rather than replacing them entirely. This is especially true when using results of AI models.  

At the 2023 Women in Technology conference held by Colorado Technology Association (CTA), a panel of AI leaders discussed the current state of AI and future considerations for the field.  

Carolyn Ujcic, Director of AI Services at Google Cloud, and fellow panelist, Ananta Nair from Dell agreed, that “AI will continue to need human intervention.” They went on to emphasize that datasets need to be strong at their core, and beyond this, humans need to help bring in context that cannot be embedded accurately or without biases into AI modules.  

Data Quality Checklist 

So how does one avoid the common mistakes and succeed in their AI models?  

Data Governance needs to be foundational for any organization collecting and using data. Of course, there are industry-norm privacy, security, and compliance laws and practices which must remain constant. A robust Data Governance presence within an organization follows these and keeps the business in good standing. 

While no plan is foolproof, the checklist below will steer you in the right direction.  

• Data Collection and Sources 

Are the methods used to collect data transparent, ethical, and consistent with data privacy regulations? Are the sources of the data well-documented, reliable, and trustworthy? Is there enough volume of data to train AI models effectively, and is it scalable for future needs? 

During the CTA AI panel, Rose Lindauer, an AI/ML engineer for Lockheed Martin, discussed the importance of ensuring that your data source can support the output you are after, simply summarizing it as, “the conversation should start with, ‘how good is your data?’” 

• Data Quality and Completeness 

Has the data maintained its integrity throughout its lifecycle, including during collection, storage, and processing? Is the data free from errors, inconsistencies, and inaccuracies? Is the data complete and consistent across different sources and time periods?  

Data integrity is the core of this point, as it encapsulates data accuracy, consistency, and completeness. These pieces are necessary for the data to be usable for AI purposes. 

• Data Relevance 

Does data align with your organization’s specific CX objectives and key performance indicators (KPIs)? Is data up-to-date and does it cover all relevant aspects of the customer journey including touchpoints, channels, and interactions? 

Often, the most challenging part of finding relevant data is trimming down massive datasets. The sheer volume can be overwhelming, so we must temper our expectations and understand the data we need may not exist. This is where the human touch guides the way. 

• Data Preprocessing 

Has the data been cleaned to remove duplicates, outliers, and irrelevant information? Has data been transformed or standardized to ensure consistency and compatibility with AI algorithms? 

If the data is not already formatted so that an AI model can pick it up and run with it, somebody needs to prepare it. There is often a need to extract, transform, and load (ETL) the data to prepare it for model ingestion. 

• Data Quality Monitoring 

Is there a process for ongoing data quality monitoring to ensure data remains accurate and relevant? Is the data validated for quality before being used in AI models for CX decision-making? Is this overseen by a Data Governance entity? 

Together with Data Governance, data quality must be monitored over time to ensure it is accurate and usable. A pre-model assessment can help validate data readiness and provide benchmarks for future monitoring and identifying data drift – the evolution of data that invalidates a model. 

• Data Documentation and Metadata 

Is there a source of truth, like a data catalog or inventory, of all CX data assets, including descriptions and usage? Are metadata and data lineage well-documented and maintained? 

Tools like Collibra allow shared understanding and ensure consistency across the organization by being the one source of (data) truth. A well-maintained data catalog, including metadata details, sets the stage for spectacular AI modeling and compelling data storytelling. 

With strong data and a strategy to avoid the common AI pitfalls, you can easily leverage AI in the CX space. Utilizing AI within CX can set you apart from competitors as it will increase your ability to gain customer insights and improve your company’s performance.  

While implementing AI in CX can be challenging, RevGen Partners is here to help. We have industry experts who can help you achieve your organization’s goals and develop a plan to fine-tune your CX. Let’s get your data ready for AI so you can use it to enhance CX findings. 

Learn more about our Customer Experience expertise on our CX page. 

Alex Champagne is a consultant at RevGen Partners specializing in data storytelling and customer experience. He is passionate about helping organizations find hidden insights in their data and enhancing the way they engage with their customers.

Stephanie Caravajal is a senior manager at RevGen Partners specializing in customer experience. She is passionate about helping organizations create and manage successful CX programs while aligning CX initiatives to digital transformations.