What you will learn: Why the biggest barrier to enterprise GenAI value is not the AI model itself, but the operating model around it—and how to align business priorities, workflows, data, governance, people and technology to deliver measurable results.
The enterprise AI conversation has changed remarkably quickly.
A few years ago, the question was whether organizations should experiment with generative AI. Today, most large organizations already are.
Early enterprise examples showed how quickly the use cases were expanding. GitHub launched Copilot to help developers write code; CarMax used GPT-3 to summarize customer reviews into research content; Mattel explored DALL-E 2 to inspire Hot Wheels product design; Coca-Cola used GPT-4 and DALL-E for its Create Real Magic creative experiment; and Morgan Stanley built an AI assistant to help advisors retrieve and interpret trusted internal knowledge.
What began as experimentation in content, creativity and productivity quickly expanded into enterprise knowledge, decision support and workflow automation.
For business users catching up to the terminology, GenAI is simply AI that creates things for you – such as text, images, summaries, recommendations or code – from the instructions and information you give it.
Many business users are already using GenAI in familiar ways: summarizing a long meeting, drafting an email, turning rough notes into a presentation, helping analyze a spreadsheet, helping research a market, creating campaign concepts, rewriting a proposal, assisting with code, or reviewing a large document and surfacing the key issues.
Individually, these may look like small productivity improvements. At enterprise scale, the opportunity becomes much larger when those capabilities are connected to trusted company data, business systems, workflows and measurable outcomes. RAG is one way to connect GenAI to trusted internal knowledge; APIs, data platforms and other integration patterns can connect it to live systems and actions.
MIT Technology Review Insights’ 2025 report, Building a High-Performance Data and AI Organization, 2nd Edition, found that roughly 65% of surveyed organizations had deployed generative AI, yet only 7% had deployed it widely. More strikingly, just 2% rated their organizations highly at delivering measurable business results from AI.
That value gap matters more than the adoption number
It is also a pattern I have seen well beyond AI. Organizations can invest six or seven figures in powerful enterprise platforms from Salesforce, Adobe, Oracle and others, then realize months later that adoption is low, workflows did not change, ownership is unclear, or the original business case never made it into day-to-day execution. Often, that is not a product failure. It is a strategy, operating-model and adoption gap.
Buying a high-performance car does not create a winning race team. A Formula 1 car still needs the strategy, driver, engineers, pit crew, data, decisions and operating rhythm working as one system. Enterprise technology is similar: the platform can be capable, but value only appears when the organization around it is designed to use that capability well.
AI is no different. The problem is increasingly not access to AI.
Leaders can choose from OpenAI, Microsoft, Google, Anthropic, AWS, Salesforce, Adobe, Snowflake, Databricks and dozens of specialized platforms. Models will continue improving, prices will change, and new capabilities will emerge.
The harder question is: How does the organization repeatedly turn AI into measurable business value?
From my experience leading digital growth and transformation, customer data, MarTech, automation, AI-enabled revenue programs and operating-model change, this challenge looks surprisingly familiar.
Most organizations don’t simply have an AI problem. They have a business operating-model problem.
Customer and operational data is fragmented. Technology is disconnected. Processes evolved around organizational silos. Accountability is unclear. Governance arrives too late.
Politics can slow momentum. Adoption is underestimated. And technology investment is often measured by deployment rather than commercial impact.
AI can accelerate a strong operating model. It can also expose – and amplify – a weak one.
So before leaders ask how much AI they can deploy, or add another shiny object to the portfolio, there is a better question: Where can AI create an economic advantage, and what must change across priorities, data, workflows, technology, people and governance to realize it?
That is the real GenAI opportunity.
First, What Exactly Is GenAI?
GenAI – generative AI – is AI designed to generate, interpret, summarize, transform or reason across information.
At an executive level, think of the capability spectrum this way:
Capability
The question it helps answer
Analytics
What happened?
Predictive AI / Machine
What is likely to happen?
Generative AI
What can we create, explain, summarize or recommend?
Agentic AI
What actions can AI plan and execute, using approved tools and boundaries, to achieve an objective?
A traditional dashboard might tell a sales leader that conversion declined.
A predictive model might identify which accounts are most likely to buy.
GenAI could summarize the relevant customer history and, when connected to approved behavioral or intent data, explain the signals, recommend a next-best action, draft an email for a sales rep, or help personalize the next digital interaction.
An AI agent could potentially go further: retrieve approved third-party information, update the CRM, launch a workflow, or initiate an action within defined permissions and controls.
That final step is important. The transition from AI that informs to AI that acts changes the value opportunity – and materially changes the governance, security and accountability requirements.
Architecture Should Follow the Business Problem
One typical pattern I see repeatedly in transformation work is that the technical conversation can take the front seat before the people accountable for the business outcome have translated their priorities into clear use cases.
Business leaders know the commercial problem but may not speak architecture. IT and data teams know the technology but may not own the P&L, customer outcome or workflow. Neither side should be expected to be the translator alone.
That translation layer is critical: business objective -> use case -> workflow -> data -> risk -> architecture -> technology. Without it, organizations can end up debating models and platforms before agreeing on what success actually looks like.
Enterprise GenAI conversations can quickly become dominated by terminology: Should we use RAG? Should we fine-tune a model? Should we build agents? Those are important questions, but they are not where strategy should start. They are architecture decisions.
A simplified view of common enterprise GenAI patterns is:
Approach
What it means
Where it can fit in
Agentic AI
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
There is no universal best architecture. RAG is useful when the AI needs current or proprietary organizational knowledge. Fine-tuning can improve specialized behaviour, but it is usually not the right answer for keeping facts current. APIs and tools connect AI to systems of record. Agents can orchestrate multiple steps.
The strategic sequence should remain: Business outcome -> workflow -> required data -> risk -> architecture -> technology. Not the reverse.
Start With the Money, Not the Model
One of the easiest ways to waste an AI investment is to begin with the technology.
“We purchased Copilot. Where should we use it?”
“We need an AI agent.”
“Our CDP needs an AI model.”
“We should build a RAG environment.”
“We need an enterprise AI platform.”
Those questions start too far downstream.
Leadership should first identify where AI can materially improve one or more of five outcomes:
Revenue. Operating cost. Customer experience. Productivity and speed. Risk.
Then identify the workflows that drive those outcomes.
This distinction matters because AI activity can create the illusion of transformation. Twenty pilots are not necessarily better than three. Thousands of prompts are not necessarily business value. An AI assistant saving an employee five minutes may be useful, but that alone does not establish a meaningful return on invested capital.
McKinsey’s recent research reinforces the point. Of 25 organizational attributes studied, workflow redesign had the strongest relationship with reported EBIT impact from generative AI. Only 21% of respondents using GenAI said their organizations had fundamentally redesigned at least some workflows, and more than 80% reported no tangible enterprise-level EBIT impact from GenAI.
AI that improves a task creates productivity. AI that redesigns a workflow can change the economics of the business.
Five Disciplines for Turning GenAI Into Business Value
The MIT research I referenced provided a strong diagnosis of enterprise challenges I have too have seen firsthand around strategy and priorities, data, talent, architecture and governance. But leadership still has to turn that diagnosis into an execution model.
I would add five disciplines.
1. Prioritize AI Use Cases Like Investments
Organizations rarely suffer from a shortage of AI ideas. They suffer from too many ideas competing for attention without a common method for determining which deserve investment.
One practical heuristic: Business Value x Feasibility x Data Readiness x Adoption Potential x Strategic Differentiation / (Cost + Risk)
Score each factor consistently – for example, 1 to 5 – and use the result as a forcing function, not a finance-grade valuation. A hypothetical contact-centre assistant scoring 4 for value, 5 for feasibility, 4 for data readiness, 4 for adoption and 2 for differentiation, divided by a combined cost-and-risk score of 4, produces a comparative score of 160. The number itself is not the decision; the discipline is forcing very different ideas to compete against the same business criteria.
A contact-centre assistant may have high productivity potential but limited differentiation unless lower cost-to-serve is measured alongside customer satisfaction, retention, sales conversion or lifetime value.
A proprietary recommendation capability using unique customer or equipment telematics data may be harder to build, but it can create a stronger competitive advantage when the resulting insights are activated and attributed to revenue outcomes.
A sophisticated agent might be technically impressive but unnecessary if basic workflow automation solves the problem at a fraction of the cost.
Executives should be able to identify the three to five operational workflows with the largest realistic pools of economic value before approving a broad AI portfolio, while still allowing safe employee experimentation within clear guardrails.
2. Redesign the Workflow, Not Just the Task
his may be the most important step.
Map the current process from trigger to outcome. Where does the data live, and is it accessible? Where is information searched for manually? Where do employees wait for another team? Where is data re-entered? Which decisions are repetitive? Which interactions require judgment or trust? Where does the customer experience unnecessary friction? And where could a redesigned experience create a genuine moment of delight?
Then decide what the future state should look like. Some activities may remain human-led. Some should become AI-assisted – the important word is assisted. Some may become automated. Some may eventually become autonomous within clearly defined limits.
This is not fundamentally different from other large transformation programs.
I recall a time when I led a global mandate to modernize marketing and its operating model across six countries, it become apparent very quickly that the value did not come simply from installing technology or centralizing ownership. It came from connecting technology, data, workflows, people, governance, measurement and regional execution into a more unified operating model that improved the capability of the whole system, and all combined applying it to rise the tide of all regional teams.
AI requires the same discipline. It simply increases the speed – and the consequences.
3. Make Data Readiness Use-Case-Led
There is a dangerous interpretation of the phrase “AI-ready data.” It sounds like: clean the entire enterprise data estate first, then start AI.
For many large organizations, that could become a multi-year transformation with no near-term business case.
A more pragmatic approach: start with the highest-value use case. Determine what data it requires. Fix what prevents that data from being usable, trusted and governed. Prove the outcome. Then expand the reusable foundation.
This requires a cross-functional partnership, not a program run in isolation by IT or Data Engineering. Sales, Marketing, Finance, Operations, Product and other business functions need to be at the table as the customer of the capability and the owner of the business outcome.
Ask whether the required data is accurate enough, current enough, accessible, understood, traceable, legally permitted and governed appropriately for the decision being made.
The objective is not universally perfect data. It is fit-for-purpose trusted data connected to a measurable business outcome.
I have applied this principle directly.
In one enterprise customer-data program I led, we did not begin with “we need a CDP.” We began with a commercial opportunity the business could not pursue effectively because customer, account, equipment, telematics, service, transactional, digital-behaviour, marketing and consent data was fragmented across multiple systems and six countries. Sales and regional teams did not have the capacity to manually pursue the full opportunity, and adding headcount was not the answer.
We built the business case around revenue growth, lower cost-to-serve and a more relevant customer experience. We established a reusable Customer 360 and identity foundation, connected a hybrid predictive and rules-based sales-opportunity engine to marketing and sales activation workflows, and moved from proof-of-concept to MVP before deciding what larger packaged, composable or hybrid platform investment – if any – was justified.
The first question was whether we could connect and govern first and third-party data into a usable customer profile across a fragmented global ecosystem.
The second was whether we could activate that foundation at scale and prove commercial value without adding regional or sales headcount.
The result was approximately $10 million in incremental first-year revenue, more than $50 million in influenced CRM pipeline, 100% customer coverage with personalization within nine months, and zero additional regional headcount.
Start with the economic opportunity. Build only enough foundation to prove it. Then reuse and scale what creates value.
The same principle should guide enterprise AI.
4. Design Governance Into the Workflow
Governance should not be a committee AI teams discover two weeks before launch. It should be designed into the operating model from the beginning.
The required controls should increase with the consequence and autonomy of the AI.
AI Authority
Example
Governance posture
This should cover more than model risk. Governance increasingly needs to address data permissions and compliance, privacy, IP, cybersecurity, identity, model selection, output quality, agent authority, auditability, monitoring, incident response, regulatory obligations and financial controls.
NIST’s AI Risk Management Framework reinforces this lifecycle approach through Govern, Map, Measure and Manage, and its GenAI profile specifically highlights governance, pre-deployment testing, content provenance and incident disclosure.
Centralize standards and guardrails. Federate innovation and business ownership.
At enterprise scale, I would use a cross-functional AI Value & Governance Council to set portfolio standards and risk thresholds, with business-unit owners accountable for outcomes and Technology, Data, Security, Legal, Privacy/Risk and Finance accountable for the controls, economics and conditions required to operate safely.
The business should own the business outcome. The enabling functions should create the conditions under which it can be achieved safely and repeatedly.
5. Know When to Rent, Buy or Build
Every AI strategy eventually becomes an investment decision.
Do we know the point at which renting a capability stops making economic sense compared with buying or building it – and what return on invested capital each path needs to clear?
RENT
Use general-purpose AI through subscriptions, platforms or APIs. This can make sense when capability is becoming commoditized, speed matters and proprietary differentiation is limited.
BUY
Adopt a specialized AI-enabled SaaS or industry platform. This can work when the workflow is relatively common and a vendor can deliver the capability more efficiently than your organization.
BUILD
Develop proprietary applications, RAG capabilities, decisioning, workflows or agents. This deserves consideration when proprietary data, unique workflows, intellectual property, regulation or scale creates a genuine strategic advantage.
But calculate the economics correctly. The comparison is not simply subscription fee versus development cost.
True total cost of ownership includes implementation, integration, data engineering, cloud and inference, security, people, governance, training, monitoring, maintenance, change management and switching costs. There is also an opportunity-cost question.
Building proprietary technology around a capability that becomes commoditized can destroy returns. Renting a strategically differentiating capability forever may be equally expensive.
The goal is to make your data, customer knowledge, workflows, business rules and operating model more durable than whichever model or technology vendor happens to lead today. That is strategic optionality.
Understand the Blockers Before Funding the Roadmap
Before launching, leadership should explicitly assess eight recurring blockers:
- Unclear business ownership and accountability for the outcome.
- Fragmented, inaccessible or low-trust data.
- Architecture and integration debt.
- Security, privacy, IP and regulatory exposure.
- Hallucination, unreliable output or weak evaluation discipline.
- Poor workflow adoption and change management.
- Vendor lock-in and uncontrolled AI economics.
- Excessive agent permissions or autonomy.
The important point is that most of these are not model-performance problems. They are operating-model problems, which is why simply waiting for a better model rarely solves them.
8 Questions Every Executive Sponsor Should Answer
- Where is the economic value? Which three to five use cases and workflows offer the strongest realistic opportunity to increase revenue, reduce cost, improve customer experience, increase productivity or reduce risk?
- What data does AI actually require? Where does it live, who owns it, can we legally use it, can we trust it, and what must be remediated before the use case works?
- Does this problem actually require AI? Could workflow redesign, traditional automation, analytics, rules or an existing capability solve it faster, cheaper or with less risk?
- How much authority should AI have? Should the workflow remain human-led, become AI-assisted, AI-recommended, AI-executed with approval, or autonomous within defined boundaries? Who remains accountable when it fails?
- How will we know the system is good enough? What thresholds will we use for accuracy, reliability, hallucination, security, latency, cost and compliance? How will we measure customer impact and business performance? What is the human override or kill mechanism?
- Should we Rent, Buy or Build? What is the real three-to-five-year TCO and investment return, and where does proprietary capability genuinely create strategic differentiation?
- How trapped could we become? Can we change models, platforms, vendors or clouds without rebuilding the data foundation and business workflow?
- What proves we should scale – or stop? Which metrics determine whether an initiative receives more capital, requires redesign or gets killed? What measurable evidence 12 months from now would demonstrate that the AI strategy is working?
Those questions move the conversation from “How are we using AI?” to “How are we creating and protecting enterprise value with AI?”
A 120-Day Plan: From AI Opportunity to Measured Value
A company does not need another three-year AI strategy before doing anything. But it does need enough structure to avoid scaling the wrong things.
Stage
Executive question
Key work
Output
What should we pursue and how should it work
Days 91-120: Industrialize
By Day 120, leadership should be able to clearly articulate where the highest-value opportunities are, why they are worth pursuing, what data they require, how they should be delivered, who owns them, what AI is permitted to do, what the economics look like, what value has been demonstrated and what qualifies for further investment.
If those answers remain unclear, buying more AI technology probably isn’t the next move.
From AI-Ready to Value-Ready
The winners of the next phase of enterprise AI will not necessarily be the organizations with the most copilots, models, agents, platforms or pilots – or the biggest headline about headcount reduction.
They will be the organizations that become better at repeatedly connecting technology to economics.
They will know how to identify where AI creates an advantage, connect the right data to the right decisions and outcomes, redesign workflows instead of automating inefficiency, balance automation with human judgment, govern risk without suffocating innovation, and stop funding initiatives that survive on novelty or sponsorship rather than evidence of value.
They will also recognize that AI cannot remain an isolated technology agenda.
Whatever title or acronym you wear today – CIO, CDO, CMO, CRO, COO, CFO, Risk leader or business-unit executive – the organization increasingly has to operate from shared outcomes.
That is why I would challenge leadership teams to think beyond the idea of becoming simply “AI-ready.”
The objective isn’t readiness for a technology. It is readiness to create value as technology continues changing.
Don’t build an ‘AI-ready data organization.’ Build a business-value operating model in which data + AI + people + workflows + governance work as one system.
That is how GenAI moves from experimentation to enterprise value.