6 Insights to Make Your Data AI Ready with Teresa Tung | Microsoft Azure Blog

I sat down with Teresa Tung to learn more about the changing nature of data and its value to AI strategy.

The success of AI depends on many factors, but the key to innovation is the quality and accessibility of an organization’s proprietary data.

I sat down with Teresa Tung to discuss the possibilities of proprietary data and why it’s so important to create value with AI. Tung is a researcher whose work involves breakthrough cloud technologies, including the convergence of AI, data and computing capabilities. It’s a prolific inventory, holding more than 225 patents and applications. And as Accenture’s global leader in data capabilities, Tung leads the vision and strategy to ensure the company is ready for the ever-changing advances in data.

We covered a number of topics, including six of Teresa’s insights.

Finally, we conclude with Teresa’s advice for business leaders using or interested in AI

Susan Etlinger (SE): In your recent article, “Essentials of the New Data,” you introduced the idea that proprietary data is an organization’s competitive advantage. Could you elaborate?

Theresy Teng (TT): So far, data has been treated as a project. When new statistics are needed, it can take months to acquire the data, access it, analyze it, and publish the statistics. If these findings prompt new questions, the process must be repeated. And if the data team has bandwidth or budget constraints, it needs even more time.

“Instead of treating it as a project – an afterthought – own data should be treated as a core competitive advantage.”

Generative AI models are pre-trained on an existing internet-scale corpus of data, making it easy to get started on day one. However, they don’t know your business, people, products or processes, and without this protected data, the models will give you the same results as your competitors.

Every day, companies invest in products based solely on their opportunity. We know the opportunity of data and artificial intelligence – better decision making, reduced risk, new avenues for monetization – so shouldn’t we think about investing in data similarly?

SE: Given that much of the company’s own knowledge lies in unstructured data, can you talk about its importance?

TT: Yes, most companies run on structured data – data in tabular form. But most data is unstructured. From voice messages to images to video, unstructured data is highly reliable. It captures the nuances. Here is an example: if a customer calls customer support and leaves a product review, this data can be extracted by its components and transferred to a table. But without nuanced inputs like the customer’s tone of voice or even swearing, there isn’t a complete and accurate picture of that transaction.

Unstructured data has historically been challenging to work with, but generative AI excels at it. That actually needs rich context of unstructured data to train on. In the age of generative artificial intelligence, this is so important.

SE: We hear a lot about synthetic data these days. what do you think?

TT: Synthetic data is necessary to fill in missing data. It allows companies to explore multiple scenarios without the extensive costs or risks associated with actual data collection.

For example, advertising agencies can run different campaign images to predict audience response. Pushing cars into dangerous situations is not an option for automakers training self-driving cars. The synthetic data teaches the AI – and therefore the car – what to do in extreme situations, including heavy rain or a surprising crosswalk.

Then there is the idea of knowledge distillation. If you use this technique to build data with a larger language model—say, a model with 13 billion parameters—that data can be used to fine-tune the smaller model, making the smaller model more efficient, more cost-effective, or deployed on a smaller device.

The AI is so hungry. To be relevant, it needs representative datasets of good scenarios, boundary conditions and everything in between. This is the potential of synthetic data.

SE: Unstructured data is generally data that is generated by human beings, so it is often case-specific. Can you share more about why context is so important?

TT: Context is key. We can capture this in the semantic layer or domain knowledge graph. It’s the meaning behind the data.

Think of every domain expert in the workplace. If a company runs a 360-degree assembly of customer data that spans domains or even systems, one domain expert analyzes it for leads, another for customer service and support, and another for customer billing. Each of these experts wants to see all the data, but for their own purpose. Knowledge of trends in customer support can influence, for example, the approach to marketing campaigns.

Words often have different meanings as well. When I say, “That’s hot for summer,” the context will determine whether I’m referring to a temperature or a trend.

Generative AI helps display the right information at the right time to the right domain expert.

SE: Given the speed and power of smart technologies, data management and security and artificial intelligence are at the fore. What trends are you observing or predicting?

TT: New opportunities come with new risks. Generative AI is so easy to use that it turns everyone into a data worker. This is an opportunity and a risk.

Because it’s easy, generative artificial intelligence built into apps can lead to unintended data leakage. For this reason, it is important to think through all the implications of generative AI applications to reduce the risk of them inadvertently exposing confidential information.

We need to rethink data management and security. Everyone in the organization needs to be aware of the risks and what they are doing. We also need to think about new tools like watermarking and confidential computing where generative AI algorithms can be run in a secure enclave.

SE: You’ve said that generative AI can jump-start data readiness. Can you elaborate?

TT: Sure. Generative AI needs your data, but it can also Help your data.

Applied to your existing data and processes, generative AI can build a more dynamic data supply chain, from capture and management to consumption. It can classify and tag metadata and can generate design documents and deployment scripts.

It can also support reverse engineering of the existing system before migration and upgrade. It is common to think that the data cannot be used because it is in an old system that is not yet cloud-enabled. But generative AI can jump-start the process; it can help you understand data, map relationships between data and concepts, and even write a program, including testing and documentation.

Generative AI is changing what we do with data. It can simplify and speed up the process by replacing one-time dashboards with interactivity such as a chat interface. We should spend less time throwing data into structured formats by doing more with unstructured data.

SE: In conclusion, what advice would you give to business and technology leaders looking to build a competitive advantage with data?

TT: Start now or be left behind.

We have realized the potential that AI can bring, but its potential can only be achieved with your organization’s proprietary data. Without this input, your result will be the same as everyone else’s, or worse, inaccurate.

I encourage organizations to focus on getting their digital AI core ready. The modern digital core is the technological ability to drive data in AI-led reinvention. It’s a combination of your organization’s cloud infrastructure, data and AI capabilities, and applications and platforms with security designed at every level. Your database – as part of your digital core – is essential to storing, cleaning and securing your data to ensure it’s high quality, managed and ready for AI.

Without a strong digital core, you have the proverbial eyes to see, brain to think, and hands to act.

Your data is your competitive differentiator in the era of generative artificial intelligence.

Teresa Tung, Ph.D. is the Global Data Capability Leader at Accenture. A rich inventory of more than 225 patents, Tung specializes in bridging business needs with breakthrough technologies.