Data Quality and Diversity: The Critical Battlefield Beyond AI Models

1. Introduction: The Key Battlefield Beyond Models

On May 27, 2025, Salesforce completed an all‑cash acquisition of data‑management company Informatica for $8 billion, the largest deal since its 2021 $27.7 billion purchase of Slack . Salesforce stated the acquisition will "create the most complete, full‑stack data platform for intelligent agents" and strengthen the data foundation for its AI tools, highlighting that B2B AI competition has shifted from models to data.

The news reflects a broader trend: B2B enterprises are moving their AI focus from model capabilities to data quality and effectiveness of digital transformation.

Large language models such as ChatGPT , Claude and Gemini demonstrate AI's potential, but a critical factor often overlooked is data . As one AI researcher put it, " **Models are tools; data is the soul.** " Without high‑quality, diverse data, even the most advanced models cannot reach their full potential. This article explores why data quality and diversity are essential for corporate AI strategies and how companies can adopt a data‑first approach.

2. Data Quality: The Foundation of AI Success

Garbage In, Garbage Out

In AI, the principle "Garbage In, Garbage Out" (GIGO) is especially crucial for large models. No matter how sophisticated the architecture or how many parameters a model has, poor training data inevitably leads to poor results.

For example, a customer‑service chatbot trained on outdated product information or inaccurate support dialogues will produce misleading answers, damaging brand reputation and customer trust.

Key Dimensions of Data Quality

High‑quality data should exhibit the following characteristics:

Accuracy : Data must correctly reflect reality without errors or misleading information.

Completeness : Data should be as complete as possible, without missing critical information.

Consistency : Data from different sources must be consistent and free of contradictions.

Timeliness : Data should be up‑to‑date, reflecting the latest situation.

Relevance : Data must be relevant to the specific business scenario.

Before introducing AI capabilities, enterprises should assess whether their existing data meets these standards. Data cleaning, standardisation, and validation are essential steps to ensure data quality.

3. Data Diversity: Enhancing AI Adaptability and Fairness

Why Diversity Matters

Data diversity means the training set should cover the full range of possible situations in the target application. Diverse data helps to:

Improve model robustness : Enable the model to handle complex, less‑common scenarios.

Reduce bias : Prevent the model from favouring specific groups or situations.

Enhance generalisation : Allow the model to perform well on unseen cases.

For instance, a global customer‑service AI trained only on English data would struggle to serve non‑English users, and a model trained on data from a single region may fail to understand cultural nuances of other markets.

Strategies for Building Diverse Datasets

Enterprises can increase data diversity through:

Multi‑source data integration : Combine data from different channels, departments, and regions.

Deliberate inclusion of edge cases : Ensure the dataset contains sufficient non‑mainstream examples.

Continuous data collection : Establish mechanisms to constantly gather new, varied data.

Data augmentation techniques : Use technical methods to create diverse data variants.

4. Enterprise Data Strategy: The Winning Key in the AI Era

A Shift to Data‑First Thinking

Successful AI strategies start with a data strategy. Companies must move from "We have an AI model, now we need data to train it" to "We have high‑quality data assets, how can we leverage AI to unlock their value?"

This shift requires enterprises to:

Treat data as a strategic asset : Manage data with the same rigor as financial assets.

Establish a data‑governance framework : Define ownership, quality standards, and usage policies.

Cultivate a data‑driven culture : Encourage decisions and innovation based on data.

Building Data Infrastructure

Infrastructure that supports AI applications should provide:

Data acquisition capability : Efficiently and accurately collect data from all sources.

Data storage and management : Secure, scalable storage and management systems.

Data processing and analytics : Powerful processing and analytical capabilities.

Data sharing mechanisms : Enable safe data flow and sharing across the organisation.

5. Practical Advice: A Data‑Driven Path to AI Implementation

Start with a Data Audit

Before adopting large AI models, enterprises should conduct a comprehensive data audit:

Assess existing data assets : Understand quantity, quality, diversity, and coverage.

Identify data gaps : Pinpoint missing or low‑quality critical data.

Develop a data‑improvement plan : Define how to fill gaps and raise data quality.

Establish Data‑Quality Assurance Mechanisms

Ongoing data‑quality management is vital:

Define data‑quality standards : Set clear metrics and standards.

Implement data‑quality monitoring : Continuously monitor and promptly address issues.

Automate data validation : Use automated tools to verify accuracy and consistency.

Co‑evolve Data and AI

Data strategy and AI strategy should develop together:

Iterative optimisation : Refine data collection and processing based on AI feedback.

Domain‑expert involvement : Involve business experts in data labelling and validation.

Closed‑loop management : Create a loop from data collection to AI training to application feedback.

6. Conclusion: Towards a Data‑Driven AI Future

In the era of booming large models, we witness a pivotal shift: competitive advantage moves from "who has the best model" to "who has the highest‑quality data." As a data scientist notes, "In the AI era, algorithms may become commodities, but data remains king."

Actionable Questions

If you are considering a data‑first strategy, start with these questions:

Has your company built a complete data‑asset catalogue?

Are your current data‑quality assessment mechanisms sufficient?

Can your data‑governance framework support future AI applications?

Interactive Discussion

Share your experiences in the comments:

What challenges have you faced in data‑quality management?

What lessons have you learned while advancing your data strategy?

How do you think data will reshape your industry’s competitive landscape in the next 3‑5 years?

Let’s discuss, share, and inspire each other in this data‑driven AI era. If you found this article helpful, feel free to share it with peers to contribute to China’s enterprise data‑strategy transformation.