AT&T Logs 5X Cash ROI on AI as Agentic Workflows Top 1,000

AT&T is a Dallas-based telecommunications company whose chief data and AI officer, Andy Markus, said it achieved a five-times cash return on its AI investment last year as it scaled more than 1,000 agentic workflows.

AT&T, the Dallas-based telecommunications carrier, said it achieved a five-times cash return on its AI investment last year, according to its chief data and AI officer, Andy Markus. The disclosure is one of the few public, cash-based ROI figures from a large enterprise AI program. It comes as AT&T reports more than 1,000 agentic workflows in production or being built and processes 45 billion tokens daily.

New York Life Group Benefit Solutions, the workplace benefits unit of New York Life, has produced its own in-year results with a different architecture. CIO Matt Marze said the unit built a full-stack AI capability on AWS using Anthropic models. Its Service Sage tool reduced call hold time by approximately 19 percent and answered around 110,000 questions in its first seven months, according to an AWS case study.

The two programs show that enterprise AI can generate measurable returns within a year when it is tied to workflow automation, data modernization, and explicit governance. For C-suite readers, the relevant comparison is not which vendor won, but how each company built the platform, measured the outcome, and scaled from pilot to production.

What did AT&T actually deploy?

AT&T's AI program spans network operations, customer service, internal productivity, and back-office processes. Markus said the company has more than 1,000 agentic workflows in production or being built, and the systems process 45 billion tokens daily. The scale reflects a shift from generative chatbots to agentic AI, in which models are authorized to take actions inside the company's systems rather than only generating text for humans to review.

The company's approach is model-agnostic. That means AT&T does not rely on a single foundation model; instead, it routes tasks to the model that performs best for each use case. The platform layer handles orchestration, observability, and guardrails. Markus said the design is intended to avoid vendor lock-in and to keep unit costs low as token volume grows.

Markus described the ROI result as a flywheel effect. “Last year we had five-X cash ROI on our AI investment. We’ve hit the flywheel,” he said in an interview published by CIO. The phrase suggests that returns are compounding as more workflows are added and as the platform is reused across business units.

The 45 billion daily token figure is a measure of how deeply AI has been embedded. For context, a single large enterprise processing 45 billion tokens per day is consuming AI at a scale that was rare outside AI labs just a few years ago. The number also implies a significant infrastructure footprint, with cost controls becoming a board-level concern.

How did AT&T measure five-times cash ROI?

AT&T has not published the absolute dollar amounts behind its five-times cash ROI figure. The public evidence is Markus's statement in the CIO feature. He framed the metric as cash ROI, meaning cash returned by AI initiatives relative to cash invested. That formulation is more conservative than soft productivity estimates, which is why it carries weight with finance executives.

The ROI number likely aggregates savings and revenue contributions across the 1,000-plus agentic workflows. If each workflow returns even a modest cash benefit, the portfolio effect can be substantial. The token volume suggests that many of these workflows are running at production scale, not in pilot mode.

For peer executives, the lesson is that AT&T treats AI ROI as a portfolio metric. The flywheel comment implies that the value of the platform grows as more workflows are deployed on it. That is a different mental model from evaluating a single chatbot or copilot deployment.

What is New York Life Group Benefit Solutions doing with AI?

New York Life Group Benefit Solutions is the workplace benefits unit of New York Life. Its CIO, Matt Marze, has described the AI work as part of a broader technology modernization rather than a standalone project. The unit consolidated business applications, modernized data management, and built an AI operating platform.

“We’ve done a lot of tech modernization to drive our cloud adoption and strategic data management, and we’ve consolidated all our business applications on a set of platforms, both commercial and our own. We’ve also assembled an AI operating platform or system to give us full-stack AI capability,” Marze said in the CIO feature.

The full-stack capability includes data infrastructure on AWS, generative AI tools built with Amazon Bedrock, and Anthropic models as the primary model family. Enterprise editions of the models run in a private cloud on AWS. This architecture gives the business unit control over data residency, security, and governance while still using managed cloud services.

What has New York Life's Service Sage achieved?

Service Sage is the unit's generative AI tool for service operations. According to an AWS case study, the tool reduced call hold time by approximately 19 percent and helped answer about 110,000 questions in the first seven months. The tool gives service representatives immediate answers to complex questions about benefits, claims, and policy details.

A 19 percent reduction in call hold time is a meaningful operational metric because hold time is directly linked to customer experience and call center labor costs. The 110,000 questions answered suggests the tool moved quickly from pilot to production. It also indicates that the AI system is trusted by the workforce, since representatives will only use a tool that gives reliable answers.

The AWS case study frames Service Sage as one outcome of a broader data and AI strategy. New York Life modernized its data platforms on AWS, then used Amazon Bedrock to build generative applications. The case study says the approach gives the company a foundation for AI at scale rather than a series of disconnected pilots.

Why did New York Life choose Anthropic models on AWS?

Marze said the AI models used at New York Life Group Benefit Solutions are primarily from Anthropic, deployed as enterprise editions in a private cloud on AWS. “Currently, the AI models used are primarily from Anthropic, and enterprise editions are deployed in a private cloud on AWS, with all the governance and security layered around it because, ultimately, the business is responsible for what the AI does,” he said.

The choice of Anthropic reflects the unit's need for models with strong safety and alignment properties. In a regulated financial services environment, the cost of a model error is not just a bad answer; it can be a compliance issue. Anthropic's enterprise edition, deployed in a private cloud, gives New York Life a controlled environment for sensitive data.

The governance principle in Marze's quote is important: the business is responsible for what the AI does. That means the technology organization builds the platform, but the business unit owns the outcome. This ownership structure is a prerequisite for agentic AI, where models are granted permissions to execute actions that have real business consequences.

How do the two approaches compare?

Dimension	AT&T	New York Life Group Benefit Solutions
AI ROI	5X cash ROI last year	Not disclosed
Agentic workflows	1,000+ in production or being built	Not disclosed
Daily token volume	45 billion	Not disclosed
Primary models	Model-agnostic platform	Primarily Anthropic
Cloud	Multi-cloud, model-agnostic	AWS with private cloud for enterprise editions
Named AI tool	Not specified	Service Sage
Call hold time impact	Not disclosed	Reduced approximately 19%
Questions answered	Not disclosed	Approximately 110,000 in first 7 months

The table shows two viable architectures for enterprise AI. AT&T is pursuing a model-agnostic platform to maximize flexibility across a large, diverse operation. New York Life is pursuing a focused stack with a single primary model provider on a single cloud, which simplifies governance in a regulated business. Both have produced measurable, attributable results within a year.

What can other enterprises learn from these deployments?

1. Measure ROI in cash terms, not just in productivity estimates. AT&T's 5X figure is a cash ROI number, which is more credible to finance executives than softer metrics.
2. Build an AI platform before scaling use cases. Both AT&T and New York Life invested in platforms and data modernization before deploying AI at scale.
3. Put governance around model access. New York Life's private cloud deployment and business-ownership principle show how to manage risk.
4. Move from generative to agentic workflows. AT&T's 1,000-plus agentic workflows suggest the next stage of enterprise AI is action, not just text generation.
5. Expect results within a year. AT&T's five-times cash ROI and New York Life's seven-month Service Sage results are both in-year outcomes.

The common thread is that neither company treated AI as a bolt-on feature. Both invested in data infrastructure, platform tooling, and governance before scaling use cases. The result is that AI is embedded in operations and measured against cash or operational metrics.

What do the executives say about the agentic AI transition?

The CIO feature that reported these results frames them as evidence of an industry transition from generative AI to agentic AI. AT&T's Markus tied the ROI to the flywheel effect of scaling. New York Life's Marze tied the AI work to a broader modernization program.

“Last year we had five-X cash ROI on our AI investment. We’ve hit the flywheel.”

Marze's comments add a second layer to the picture. His emphasis on full-stack capability and business ownership of AI outcomes reflects the governance demands of financial services. Together, the two executives represent different routes to the same destination: AI systems that are embedded in operations, measured in cash, and governed by the business.

What are the implications for the enterprise AI market?

The AT&T and New York Life stories have market implications. AT&T's model-agnostic platform suggests that large enterprises are pushing back on vendor lock-in. The 45 billion daily token volume shows that AI usage in a single enterprise can reach levels that previously seemed like the domain of AI labs.

New York Life's choice of Anthropic on AWS is a signal for the model and cloud market. A major insurer selecting Anthropic as its primary model provider, with AWS as infrastructure, is exactly the kind of reference win that both companies want. For competitors, it shows that safety-focused models can win regulated enterprise accounts.

The 19 percent hold-time reduction at New York Life is a concrete example of AI value in customer service. When a generative AI tool can answer 110,000 questions in seven months, it is not just deflecting calls; it is changing how service representatives spend their time and how quickly customers get answers.

For enterprise buyers, the takeaway is that AI procurement is shifting from point solutions to platform decisions. The choices that AT&T and New York Life made about models, clouds, and governance will shape their AI capabilities for years.

What should CIOs and CFOs do next?

The first action is to define ROI in cash terms. AT&T's five-times cash ROI is a benchmark, but the methodology matters more than the number. CFOs should ask how AI investments compare to other capital allocation options and demand cash-based metrics rather than soft productivity estimates.

The second action is to invest in data modernization. New York Life's full-stack capability was built on top of cloud adoption and strategic data management. Without that foundation, generative and agentic AI tools have nothing reliable to act on.

The third action is to establish governance before deployment. New York Life's private cloud and business-ownership principle are a template for regulated industries. AT&T's model-agnostic platform is a template for large enterprises that want flexibility without vendor lock-in.

The fourth action is to plan for agentic workflows. AT&T's 1,000-plus agentic workflows and New York Life's full-stack platform both point in the same direction: the next stage of enterprise AI is action, not just text generation.

What's next for AT&T and New York Life?

AT&T's focus is on expanding its agentic workflows beyond the 1,000 mark and continuing to scale token volume. Markus's flywheel comment suggests the company expects returns to compound as the platform is reused across more business processes.

New York Life Group Benefit Solutions is likely to extend Service Sage to more service areas and to expand its generative AI tools into sales growth. The company's full-stack capability gives it a foundation for agentic workflows in the future, should it choose to deploy them.

For the broader market, the next phase of enterprise AI is agentic. Both AT&T and New York Life are showing that agentic AI, when built on a governed platform and measured in cash, can produce in-year returns. The two cases also show that there is no single winning architecture; the right choice depends on scale, regulation, and business context.