Activating Data to Close the AI ROI Gap: 4 Steps to Realizing Business Value Through Agentic AI

As we enter the fourth consecutive “year of AI,” many organizational leaders are considering two seemingly opposite questions. Firstly, is generative AI the most transformative technology of the 21st Century? And secondly, is AI overhyped? I would paradoxically answer yes to both questions. But I think those are the wrong questions. Instead, I believe these leaders should be asking themselves a different question: how can my organization get business value from AI right now?

The reality is that many companies have spent the past three years investing in new AI technologies and experimenting with new AI tools, but they have

yet to reap the envisioned rewards

. In spite of CEO pressure to “AI all the things,” organizations are not seeing the return on investment they would like. This shouldn’t be surprising. History teaches us that the most profound technical innovations take time before they pay off. There is a lag between technical invention and business innovation.

Thomas Edison showcased the power of electricity in Manhattan in 1882, but it wasn’t until Ford unveiled the electrically-charged assembly line in 1913 that electricity completely overtook steam power in manufacturing. Can you imagine a business leader in 1885 imploring their factory workers to start experimenting with electrical power? Still, electrical power prevailed and paved the way for many of the revolutionary innovations of the 20th century, from radio transmissions to digital computing.

As a more recent example, the World Wide Web went mainstream in the early nineties. Consumer usage exploded immediately, but business adoption lagged. It took half a decade before most established enterprises started benefitting from the web through ecommerce. Still, the web paved the way for social media, mobile engagement, cloud computing, and ultimately AI. Business value is generated incrementally from new technologies.

If the electrical age of business started with the assembly line, and the web age started with ecommerce, what will be the killer app for the age of AI business? The launch of ChatGPT in late 2022 introduced the power of large language models to the general public. Due to its popularity, the “chatbot that understands me and sounds human” became the archetype for how AI could be applied. As a result, many businesses started with AI by introducing similar assistants tuned to be their company’s version of ChatGPT. In many cases, results have been well-received by users, but business returns on productivity are hard to measure.

One of the most well-developed applications of LLM’s for business is in the area of coding assistants. Claude Code, Cursor, and other tools have gained widespread popularity, showing almost magical results. However,

studies

indicate that the productivity gains of individual developers have not yet translated into overall organization productivity. On top of that, speeding up development

does not help an organization’s business performance

if what’s being produced does not itself deliver business value. Coding assistants will help scale AI adoption over time, but they are not the killer app.

To find the most impactful application of AI, organizations must focus on the gears that drive their own business models. In our book


Unbundling the Enterprise


, Stephen Fishman and I examine the concept of “value dynamics,” a method for breaking down business models into a set of interconnected value exchanges. Value exchanges involve multiple “currencies,” including fees, time savings, reach, and improved quality. The most unique currency is data. In the book, we show how companies like Google and Meta parlayed data accumulation into digital dominance. Their success came by providing real-time, automated links in their value exchanges. They linked data collection to revenue generation in a virtuous circle. Both companies contextualized customer data in the form of ad targeting, then used this to drive their core revenue and collect even more data through user engagement.

While many organizations have spent the past couple of decades collecting and refining data, they have yet to fully capitalize on data’s potential yield through such a flywheel. At its core, a large language model is simply applied data. It has the potential to be the engine that drives such a flywheel of value for organizations, but that engine needs fuel in the form of contextualized data, and needs to be attached to the gears of the organization’s business model. This “data activation” process makes data trustable and available at scale, setting the foundation for more dynamic automation in the enterprise, and ultimately uncovering the killer AI app for such organizations.

What will organizations look like that have activated their data for the AI age? Consider the following the following scenarios:

• 
  A pharmaceutical company that currently has to make multi-million dollar, years long bets on new drugs to a more nimble company with shorter, parallelized clinical trial cycles enabled through dynamic, AI-infused automation
  


• 
  A retail bank who currently sends out “hope and pray” product offers to all of their customers with little uptake and manual downstream fulfillment to personalized offers with streamlined credit origination, leading to higher adoption of profitable lending products
  


• 
  A retailer whose current inventory management system is fraught with both overstocked and oversold items to a company that understands its inventory position in real time, thanks to direct outlet, warehouse, and supplier connections analyzed through always-on AI agents
  

The road to ROI outlined in these scenarios follows this new form of dynamic automation, and it is driven by data activation.

So how can organizations start this journey? Here are four steps to get going…

Step 1: Understand the Value Dynamics of Your Organization

Breaking down an organization’s business model into its underlying value exchanges is invaluable for several reasons. The resulting value exchange map shows what capabilities drive the business, what business functions are most critical, and how each element in an organization contributes to the creation, capture, and distribution of value. For our purposes, the value exchange map can be used to visualize the core business processes that will be candidates for dynamic automation. As a next layer down, you can map each value exchange and component to how they are operationalized within the organization. This could be in the form of software applications, data stores, or even employee tasks. Automation opportunities can then be weighed by impact and implementation complexity in order to zero in on the best place to apply AI and data activation.

Step 2: Propagate Optionality Through a Data Activation Layer

An organization’s ability to activate data depends on the optionality of its digital landscape. Optionality abounds when digital assets–software functions, data sources, third party services–are accessible in real time. In an AI context, this means two things. First, an organization must be able to synthesize data from disparate sources in order to provide precise context to LLM’s that lead to accurate reasoning and avoid hallucination. Secondly, software components that execute core business functions–such as a bank’s loan adjudication service or a retailer’s live inventory system–must be callable by LLM-based applications in order to fulfill the automation. In both cases, APIs are the best mechanism for making data and functions appropriately accessible. The Model Context Protocol (MCP) is gaining traction as the API protocol of choice for data activation. This set of accessible capabilities can be made into a contextual platform for your organization. Transforming your digital landscape from a set of siloed applications and data to a layer of business-aligned APIs is crucial to achieving ROI through data activation.

Step 3: Embrace the Agentic Paradigm of Digital Solutions

The prevailing software architecture of the AI age is emerging. Optimized software solutions require a balance of AI-infused and non-AI components. AI Agents–the AI-infused components in this emerging architecture–use LLM-based reasoning grounded in contextual awareness to execute tasks through tools at their disposal. They are the instruments of data activation and dynamic automation. A digital landscape optionalized through APIs (including MCP tools) is the most fertile ground for such agents to thrive. Agentic architecture allows for numerous patterns that combine the deterministic software components in the existing infrastructure with such AI Agents. These patterns range from simple chatbots and worker agents to agentic workflows right up to autonomous multi-agent systems. Organizations that adopt this architectural approach will be able to harvest the most value out of their existing digital assets while adopting AI at a pace that allows them to manage the increasing complexity of solutions that bring more and more value.

Step 4: Use AI as a Productivity Tool to Build Agentic Automations

Applying AI for worker productivity may not yield the highest returns for a business on its own. Utilizing AI productivity gains in the service of activating an organization’s data and providing agentic automation, however, can accelerate real returns. This doesn’t just mean using AI to accelerate the work of developers. Even before the AI explosion, one of the biggest barriers to delivery was the organizational gap between business domain experts who understand the application of technology, and the IT teams that build solutions. Organizational trends like DevOps have helped bridge that gap, but AI can help in even more tangible ways. As a language-based technology, LLMs are able to translate between requirements and solutions in an unprecedented way. Multimodal AI allows for the capture of business sketches that can generate usable artifacts for downstream development. Transcripts can be turned into prototypes. This is a new type of data activation: turning business domain knowledge into solution scaffolding in real time.

Following these four steps, organizations can activate their data and start seeing returns on their AI investments. Furthermore, they will be better prepared for the new ecosystems, jobs, and opportunities created by the AI economy. By understanding the value dynamics of your business, turning your digital assets into exercisable options, and orienting around agentic architecture, you will prepare your organization for the AI future by inventing it yourself.