AI-Ready Data: Why AI Fails on Foundations, Not Algorithms

Watch the full session · 59 min

In a Data Science Connect webinar, field CTOs from Fivetran and Snowflake and an Accenture managing director explained what AI-ready data means. They argued that AI fails on weak foundations rather than algorithms, and described how quality, access, governance and flexible infrastructure separate leading teams.

AI projects rarely fail because of the model, according to a Data Science Connect expert panel webinar sponsored by Fivetran, Accenture and Snowflake. "AI doesn't fail because of algorithms. AI actually fails because the foundation is not right," said Ruchi Soni, Managing Director, Data & AI at Accenture.

Soni was joined by Mark Van de Wiel, Field CTO at Fivetran, and Venkat Suru, Field CTO at Snowflake, in a discussion moderated by Brian Mink, co-founder and president of Data Science Connect. The panel covered what it means to be AI ready and what leading teams are doing differently.

What does it mean for data to be AI ready?

For Soni, AI-ready data is "reliable. It is connected and it is well governed." Many organizations underestimate how much foundational work is needed before AI creates business value, she said. If teams spend more time finding and cleaning data than using it, they aren't there yet.

Van de Wiel said operational data sits in silos across SaaS applications, databases, streams and files. Consolidating it onto a platform that can scale to AI workloads is the starting point. "There is no AI strategy without data strategy," Suru said, adding that platforms now serve business users, application teams and MLOps and LLMOps teams, not just data professionals.

What data challenges hold organizations back from AI?

The biggest barriers are foundational, not technical, Soni said. She listed fragmented and error-filled data, silos, unclear ownership and missing metadata, which leave teams chasing data instead of using it. Skills gaps are also stalling initiatives, she said, even when the data is in good shape.

Suru urged organizations to treat even internal data as a product. "Start to think of data as your product," he said. Once teams do that, process, priority and executive attention follow. He compared the shift to how IT departments moved from being seen as a cost burden to being measured on business value.

Van de Wiel said moving and storing data doesn't differentiate a company. What it does with the data does, so infrastructure should be as easy to use as possible.

How do data quality and access affect AI projects?

Simple issues such as blank values, nulls and date and time conversions can feed garbage into AI, Suru said. In regulated industries, quality checks also have to meet standards such as HIPAA and PCI DSS.

Van de Wiel warned about polished generative AI answers built on bad inputs. "Forget about hallucinations. We're just using the wrong foundation to generate a response, but the response will be well written," he said. Under pressure to be productive, people are tempted to use such answers.

Soni said every client conversation includes a data quality framework and access controls. A banking client, for example, won't give everyone access to all its data. Getting access controls wrong, she said, slows development, raises costs and creates frustration.

What role do governance, security and compliance play?

Van de Wiel described governance as a single source of truth on a scalable platform, then clear definitions and a semantic model so everyone understands what they're looking at, then access control. He recalled the era of personal spreadsheets that became one person's version of the truth.

Soni defined the three pillars: governance as clear ownership, definitions and quality; security as protecting sensitive data, such as masking customer IDs at a finance company; and compliance as meeting regulatory and ethical standards. Organizations should "treat governance, security and compliance as enablers rather than obstacles," she said.

Suru said security is the top worry for customers moving to cloud platforms, and that as business users gain access to language models, governance must cover who can use which model on which data. He argued for running AI close to the data under a single governance model.

How can leaders tell if their data infrastructure is ready for AI?

Soni said leaders should look beyond the technology stack to data quality and consistency, integration and access, governance and ownership, and culture and skills, which she said many people forget.

Suru said platforms need flexibility, including support for open table formats such as Apache Iceberg, for both traditional machine learning and language models across platforms, and a choice of CPUs and GPUs to manage cost. Van de Wiel encouraged organizations to consider cloud managed services, since they can't predict growth, acquisitions or mergers, and running their own infrastructure requires scarce expertise.

What are AI leaders doing differently?

Leading organizations modernize from self-hosted systems to cloud while keeping data flowing, Van de Wiel said, and "leave no data behind." Soni said they treat data as a strategic asset, build cross-functional teams that combine domain expertise, data science and engineering, and embed AI into processes rather than treating it as a separate project.

Soni contrasted a retail client that standardized sales and inventory data before deploying recommendation and demand forecasting models with organizations that skipped that work. In those cases, she said, models gave wrong insights, teams lost trust and initiatives stalled. Van de Wiel cited consumer goods distribution and predictive maintenance in manufacturing as classic machine learning uses that depend on centralized data.

Catch the next expert panel live

Data Science Connect has 24 free live webinars on the 2027 schedule across four series: the AI × Enterprise Panel Series, AI Demo Days, Agentic AI Showcases and the Plug & Play Series. Each brings enterprise and vendor experts together to answer audience questions live. See the full schedule at webinars.

Watch: AI Starts with Data : What Leading Teams are Doing Differently on YouTube
Event: Webinars · More coverage: Webinars insights