10 steps to ensuring CX data is good enough for AI

As our Global State of CX research confirmed in 2024, as a driver of CX data is top of mind for CX practitioners around the world – and there is good reason for this.

In 2023, we found the top trends influencing the role of the practitioner were focused on digital CX, AI and automation. In 2024, we noticed these trends drop down the list of CX “priorities” and be replaced by data.

From data collection to data storage, utilization and security, data powers the digital, AI-driven and automated experiences that allow organizations to stand out against their competition.

As the old adage goes, rubbish in equals rubbish out. This means that for AI to work effectively, all data must be accurate, unified and accessible to the different systems that depend on data.

Yet CX Network’s research into the Global State of CX in 2024 found that only 42 percent of survey respondents trained CX teams in data utilization or management in the 12 months before the survey was fielded. This signals a potential challenge for a number of organizations; failure to ensure the quality of data can compromise the performance of AI models, resulting in bias, hallucinations and even brand damage.

As outlined in the CX Network report, Maximizing data strategy for enhanced CX with AI, this article sets out 10 steps CX practitioners in service, data and journey design need to take in order to prepare data for unification and therefore, utilization.

Centralizing customer data

Dominik Olejko, a senior CX executive and retail expert who has worked for IKEA, Decathlon and H&M says the “biggest winners” when it comes to using AI effectively, are not the organizations with the most data, but the ones that have the most organized data and have strong data governance policies in place.

“When it comes to data organization and utilization, the arrival of AI shows us the winners [and losers] in the market. As they say, when the tide goes out you see who has been swimming naked,” he says.

At the simplest level, Olejko says practitioners should maintain clear versioning to help manage updates and ensure data sets are fresh and accurate and all those in CX “need to understand how to talk to AI”. He says: “If you do not know what kind of questions should be asked, you can’t work with AI properly.”

On how to do this, Srinivas Pradeep principal data scientist for Glencore, says: “To begin, practitioners must determine all the sources from which customer data is collected, such as CRM systems, social media, website analytics and customer support platforms, then create a comprehensive inventory of data assets, noting the type, location and accessibility of each dataset. Choose a centralized data platform that allows easy access and sharing of data across the organization,” Pradeep says.

Other crucial first steps include eliminating redundant and duplicate data entries to ensure each customer is represented only once in the database.

To support strong data governance and security, Pradeep advises practitioners to clearly define data governance policies to ensure data accuracy, consistency and security, and designate individuals to take responsibility for overseeing data quality as well as compliance with the governance policies.

In the later stages, practitioners should conduct training to ensure that all relevant teams understand the importance of data unification and how to use the centralized platform effectively.

Finally, Pradeep says: “Practitioners should encourage collaboration between different departments to ensure that data are shared and utilized to improve CX holistically.”
However, unification is not a “set and forget” exercise. Maintaining the quality and integrity of data requires continued updates.

Keeping ahead on data compliance and security

Regulatory obligations also need to be met.

Hundreds of regulations and laws exist to protect and limit how customer data are collected, processed and stored by organizations – and these regulations change between different geographic regions.

Joshua Tye, senior customer operations leader for Cash App, says those who have been smart about data management in the past are now able to minimize the impact of regulations.

“When thinking about data storage, transfer and processing it is important to determine what is important and what data is critical for customer facing and enablement teams. After determining what is critical, a best practice is to then prioritize the data that need to be kept and work across the business to develop data handling practices,” he says.

“This type of governance model ensures that only significant regulatory changes would materially impact the way the organization handles data,” Tye adds. 

Tye advises practitioners join forums and other professional networks that can inform them of potential changes on the horizon.

“Other best practices include engaging in peer discussions on the regulatory landscape, visiting conferences where discussions on regulatory changes are a key topic, or developing relationships with regulatory bodies specific to your type of company,” he says. “Practitioners should consistently design their customer strategy and Objectives and Key Results (OKRs) to include intended changes to regulations.”

Overcoming customer concerns about data and AI in CX

Delivering on data quality, security and compliance can help keep regulators happy. However, to maintain customer trust, businesses must go above and beyond these basic requirements and commit to ethical data use and AI development.

Since the arrival of generative AI, consumers in all markets have become more aware of how AI models are trained, how data power them and the security and ethical risks such data collection and utilization can pose.

When CX Network asked its members to select the three customer behaviors that are influencing their roles at present, awareness of how AI works and uses customer data emerged as the third most selected response. Furthermore, 65 percent of survey respondents either agreed or strongly agreed that customers are concerned about ethical AI use and the future development of AI for CX.

When data quality is compromised, the mistakes that arise can quickly erode customer trust.
Adam Nowak, the global CRM and loyalty director for Electrolux, says: “Transparency, ethical principles, security measures and regulatory compliance are just hygiene factors when it comes to building real trust with customers about using their data, especially with AI involved.

“To truly build trust, organizations need to clearly demonstrate how providing data will tangibly solve problems and create value for the customers themselves,” Nowak adds.

The key to this, Nowak says, is to communicate the value exchange on offer. Practitioners should educate customers on specific use cases that clearly demonstrate to the customer how their data, paired with AI capabilities, can enhance their experience, for example through more convenience, personalized recommendations or new and valuable features.

“When customers can clearly see what is in it for them, they are more likely to feel the value exchange is worthwhile and opt-in to data sharing,” he adds.

10 fundamentals to ensure your CX data is good enough for AI

As outlined by Pradeep, here are the 10 steps practitioners need to take in order to ensure the data they work with will deliver the customer experiences they intend.

1. Data collection and integration

• Diverse data sources: Collect data from multiple sources such as customer feedback, surveys, social media, call centers, emails, and chat logs to get a comprehensive view of customer interactions.

• Data integration: Ensure seamless integration of data from different sources to create a unified dataset. This can be a big effort, but integration tools have advanced greatly over the last few years and the effort is critical for success of any AI model.

2. Data quality management

• Data cleaning: Remove duplicates, correct errors, and handle missing values to improve data quality.

• Normalization: Standardize data formats and scales to ensure consistency across the dataset.

• De-duplication: Ensure that repeated or redundant entries are removed to avoid skewing the model's understanding.

3. Data enrichment

• External data: Augment internal data with external data sources such as market trends, demographic information, and economic indicators to provide additional context.

• Metadata: Add metadata to enrich the dataset, including timestamps, customer profiles, and interaction contexts.

4. Data annotation and labeling

• Manual annotation: Use human annotators to label data accurately. This is particularly important for training supervised learning models.

• Crowdsourcing: Utilize crowdsourcing platforms for large-scale data labeling.

• Automated tools: Leverage automated labeling tools where applicable, but always review and correct labels to ensure accuracy.

5. Data governance

• Policies and standards: Implement data governance policies to ensure data quality, privacy, and compliance with regulations (e.g., GDPR, CCPA).

• Access control: Restrict data access to authorized personnel to maintain data integrity and security.

• Regular audits: Conduct regular data quality audits to identify and rectify any issues.

6. Bias mitigation

• Bias detection: Analyze datasets for potential biases and ensure diversity in training data to avoid skewed model outcomes.

• Fair representation: Ensure data represents all customer segments fairly to prevent biased predictions.

7. Validation and testing

• Holdout data: Set aside a portion of the dataset as “holdout” or “validation data” to test the model’s performance.

• Cross-validation: Use cross-validation techniques to ensure the model generalizes well to unseen data.

8. Continuous monitoring and improvement

• Data drift monitoring: Continuously monitor the data for changes over time (data drift) and update the training data and models accordingly.

• Feedback loops: Implement feedback loops to incorporate new data and insights, ensuring the model evolves with changing customer behaviors and preferences.

9. Collaboration with data scientists

• Interdisciplinary teams: Work closely with data scientists, engineers, and domain experts to ensure the data is relevant and accurately prepared.

• Training and workshops: Conduct training sessions and workshops to improve the understanding of data quality requirements among all stakeholders.

10. Leveraging advanced tools and technologies

• Data quality tools: Utilize advanced data quality tools and platforms that offer capabilities such as automated data cleaning, anomaly detection, and data profiling.

• AI and ML for data quality: Use AI and machine learning techniques to detect and correct data quality issues automatically.

Quick Links

Why your predictive analytics and AI projects are failing – and how to transform your success

How to grow your customer base, enhance UX and drive sales with data

Maximizing data strategy for enhanced CX with AI 

Special Report: The Global State of CX 2024

Now in its eighth year, this edition of our flagship annual report is also the first in the Global State series to examine the profit and loyalty benefits delivered by disruptor technologies such as generative AI and virtual reality. Download the Global State of CX 2024 now.

Download Now