AI Automation Is Only as Good as the Data Behind It

AI Automation Is Only as Good as the Data Behind It

Lessons from Industry Leaders

Artificial intelligence is becoming part of business operations across almost every industry. While the technology is still relatively new, its use is advancing rapidly.

Automation is one of the fastest-growing applications of AI. Companies all over the world are exploring how to reduce the time employees spend on manual tasks while preserving accuracy and privacy. Despite this push, 88% of AI pilot programs reportedly don’t reach production.1

The technology itself is only part of the equation. AI automation depends on the quality, accessibility and governance of the data flowing between systems. Because automated systems can act on information at scale and with limited human intervention, inaccurate, incomplete or poorly governed data can quickly become a significant business risk.

When AI Learns the Wrong Lesson

When automation is built on flawed or incomplete data, the entire system can be compromised. Unity Technologies learned this cost in 2022, when it experienced problems with one of its machine-learning-powered advertising products, contributing to a 37% drop in its stock price and an estimated $110 million impact on the business. 2

Bad data from one of Unity’s large customers was uploaded into the platform, which helps game developers optimize advertising and monetization through ad placement. This inaccurate data affected performance, reducing effectiveness and requiring the company to make an emergency fix.

This case demonstrates why data quality is critical to AI performance. Even a sophisticated machine-learning system cannot produce reliable results when the data feeding it is inaccurate. AI can magnify the value of good data, but it can also magnify the consequences of bad data.

The Power of AI Starts with the Data

Despite these challenges, industry giants are showing what AI can accomplish when it is supported by strong data management. Their success demonstrates that effective AI isn’t just about having the right technology. It depends on how organizations manage, govern and integrate the data behind it.

Lesson 1: Reliable Automation Starts with Quality Data

Google Cuts Data Center Energy Use With AI

The Challenge
Google wanted to increase energy efficiency and reduce CO2 emissions in its datacenters. Cooling was one of the data centers’ largest energy demands, but automating these systems presented two challenges. Cooling needs could change quickly based on internal and external conditions, such as weather, and each data center had a unique operating environment.

The Solution
Working in partnership with DeepMind, Google developed a fully automated AI system using real-time sensor data and a large cache of historical data. Every five minutes, the system takes a snapshot of thousands of sensors to predict temperature changes and take action to reduce the cooling systems energy consumption.

Alongside the sensor snapshots, Google used 80 million examples of historical operating data and implemented an eight-layer safety check to evaluate recommendations. 3

The Result
Google reportedly reduced the amount of energy used for cooling up to 40%.4  The results demonstrated four data management practices that contributed to the system’s success:

  1. Reliable real-time data: The system received a continuous flow of accurate sensor data.
  2. Quality historical data: A large collection of well-managed historical data was available to train the model.
  3. Usable information: Data was organized and structured so the system could put it to work.
  4. Ongoing governance: The system was continuously monitored and operated within extensive safety controls.

Lesson 2: AI Automation Depends on Connected Systems

SolutionHealth Streamlines Clinical Documentation

The Challenge
SolutionHealth, a health care system in New Hampshire, wanted to reduce the amount of time clinicians spent manually entering notes into electronic health records. Creating an automated system was difficult because patient conversations contain valuable but unstructured information. That information needed to be organized and made available within the systems clinicians already used.

The Solution
SolutionHealth deployed an AI ambient listening system that converts conversations between clinicians and patients into draft medical notes. Rather than creating information in a separate tool, the notes are integrated into the electronic health record system, where clinicians can review and approve it as part of their established workflow.

The Result
Across nearly 60,000 patient encounters, clinicians using the AI listening system reported spending average of 56% less time documenting during encounters.5

The results demonstrate that generating information is only part of the equation. By connecting the AI system with its existing electronic health record platform, SolutionHealth made that information immediately accessible within the workflows where clinicians needed it.

Lesson 3: Real-Time AI Requires Reliable, Accessible Data

Visa Uses AI to Fight Fraud

The Challenge
A top-tier U.S. credit union was experiencing significant losses due to fraudulent online transactions. Its existing fraud prevention systems were struggling to identify suspicious activity without also declining legitimate customer transactions.

The Solution
The credit union partnered with Visa to use AI-powered fraud detection risk models. The models use machine learning to analyze transaction patterns and identify higher-risk activity. Instead of relying on manual review, AI-generated risk scores are incorporated into the transaction process, allowing the system to evaluate risk and identify suspicious activity as transactions occur.

The Result
According to Visa, the credit union blocked more than 18,000 fraudulent transactions and prevented approximately $2.5 million in fraud losses after implementing its AI-powered risk models.6

For AI to support decisions in real time, the underlying data must be reliable, accessible and available when the system needs it. When information is delayed, disconnected or difficult to access, even sophisticated AI tools can be limited in their ability to act on it.

Build a Stronger Foundation for AI

As AI is given more autonomy in business operations, stronger data management becomes even more important. Without reliable, well-governed data, AI can amplify existing problems. With the right data foundation, organizations can reduce manual work, improve efficiency and get more value from the technology they already have.

Having the right tools is only part of the equation. The data flowing between those systems needs to be accurate, accessible and connected for AI to work effectively. Big Data Management Services helps organizations address the data quality, governance and integration challenges that can stand in the way of successful AI initiatives.

Not sure where your data stands? A Data Readiness Audit can identify gaps in your current data environment and help determine what needs to be addressed before moving forward with AI. Access your data readiness today. 

Sources

  1. 88% of AI pilots fail to reach production — but that’s not all on IT | cio
  2. The Impact of Bad Data and Why Observability is Now Imperative | IBM
  3. Safety-first AI for autonomous data centre cooling and industrial control | Google DeepMind
  4. DeepMind AI Reduces Google Data Centre Cooling Bill by 40% | Google DeepMind
  5. SolutionHealth builds patient-focused workflows with Dragon for two health systems | Microsoft
  6. Top-tier U.S. credit union reduces fraud by 35% with Risk Advisor | Visa