Scaling Enterprise AI: How Leaders Prove Value and Control Cost

Time savings are the easiest AI benefit to measure, but the leaders on a panel at the ALIGN AI Executive Summit Atlanta 2026 argued that real value comes from what an organization does with that time, and from the engineering discipline needed to scale.

The panel, Scaling AI That Works: Proving Value, Managing Cost, and Driving Impact, was sponsored by Mistral AI and held on May 20. Dr. Frances Boykin, Director, Advanced Analytics at AT&T, moderated. The panelists were Vida Williams, Chief Data Officer at Stride Inc.; Shashank Kadetotad, Senior Director of Enterprise Data Science and AI at Mars; Ian Mathew, Solutions Architect and North American Lead at Mistral AI; and Nisha Jaisinghani, Director IT, Data & Analytics, Corporate Capabilities at Georgia-Pacific.

How do you measure AI value?

Jaisinghani said time savings are where her team sees value first, especially in software development, where AI-enabled tools speed up coding. Ticket resolution agents in tools like ServiceNow now surface past fixes for similar issues. "That's where we are seeing about 40% being claimed as savings," she said.

Mathew, a former value engineer at Adobe, warned against reading time savings as headcount cuts. The better question, he said, is what engineers do if they can now complete a year of work in a month. In customer service, freeing agents from routine questions can cut wait times and help retention.

Williams took what she called a controversial view: Stride's biggest return has come from advancing data governance and data maturity, which increased self-service and sped up decisions. Kadetotad said Mars also measures what it calls decision complexity, such as the time and effort behind pricing decisions, though the value depends on how often those decisions are made.

How do enterprises move AI from pilot to production?

Kadetotad noted a reversal. In traditional AI, most of the effort goes into making a pilot work, and scaling is comparatively easy. With large language models, "pilots are pretty easy to do," but problems surface at scale because the models are probabilistic. Mars insists on a crisp, business-approved problem statement and invests in shared infrastructure for agents.

Williams said Stride separated the development of AI services from the products that use them. That modular design lets the team observe the whole system and decommission a misbehaving model without interrupting the platform. She warned against cutting engineering teams in favor of quick builds: "You will have a security risk within two days."

Mathew's advice was to remember that AI is software. "It's not magic," he said, and it needs good data governance, testing and security like anything else.

How should leaders prioritize AI use cases?

Mars ranks use cases by the availability and cleanliness of the data they need and the time to build a minimum viable product, Kadetotad said. Jaisinghani described a quick win at Georgia-Pacific: an agent that links parts in engineering drawings to master data items, with a human validating the results. Doing it manually, she said, would take about 250 people a year.

Mathew added that quick wins should also scale, because the setup cost of procuring and securing a model should pay off beyond one team. Williams said Stride weighs the total cost of AI against people and what she called collective cognitive offloading. If AI replaces the junior work that trains future experts, she asked, who will handle the expensive cases in five years?

How do you control AI costs?

Georgia-Pacific started with unlimited tokens for developers and then set limits as adoption grew, Jaisinghani said. Mathew said some engineers now spend more on tokens than their salary, and recommended routing simple tasks to small, cheaper models. Kadetotad said Mars routes prompts to the right model, uses caching and keeps a central registry of the agents employees build.

Williams folded cost monitoring into Stride's governance program and asks her engineers a blunt question: "Is the thing that you wrote today worth your drinking water?" She said framing cost that way built a culture of cost consciousness. Her team also uses Atlan to see query costs.

What makes AI adoption stick?

Jaisinghani said a central AI team can't solve every problem. She prefers small pods that pair an AI specialist with a business expert and a product mindset. Mathew starts every engagement with an evaluation framework to learn exactly where and how often a model fails.

Kadetotad compared adoption to the last 10 meters of a 100-meter race. Mars now budgets for training, talks directly with the people who will use a tool, and defines in advance when to abandon a project and document what it learned.

ALIGN returns to Atlanta in 2027

The ALIGN AI Executive Summits bring data and AI executives together for invitation-only discussions in eight U.S. cities in 2027: Atlanta, New York City, Dallas, Chicago, San Francisco, Boston, DC and Los Angeles. Registration is coming soon. Details are on the ALIGN AI Executive Summits page.

Frequently asked questions

How do companies measure the value of AI?

Panelists start with time savings but look at what teams do with that time, along with decision speed, data maturity and the cost and frequency of decisions AI supports.

Why is scaling generative AI harder than piloting it?

Mars' Shashank Kadetotad said LLM pilots are easy, but probabilistic behavior causes problems at scale, so production needs clear problem statements and shared infrastructure.

How can enterprises control AI token costs?

Set token limits as adoption grows, route simple tasks to smaller models, cache results, monitor agents centrally and include cost in governance reviews.

How should AI use cases be prioritized?

By data availability and cleanliness, time to a minimum viable product, total cost versus people, and long-term effects on workforce skills.

This article is based on remarks by Dr. Frances Boykin, Director, Advanced Analytics, AT&T (moderator); Vida Williams, Chief Data Officer, Stride Inc.; Shashank Kadetotad, Sr. Director of Enterprise Data Science and AI, Mars; Ian Mathew, Solutions Architect, North American Lead, Mistral AI; Nisha Jaisinghani, Director IT (Data & Analytics - Corporate Capabilities), Georgia-Pacific at ALIGN AI Executive Summit Atlanta 2026 on May 20, 2026. More coverage: ALIGN AI Executive Summits.

Filed under: AI Strategy & ROI, Agentic AI, AI Governance & Trust