From Business Goals to AI Value: Why Every AI Initiative Should Start with Strategy, Not Technology

One of the most impactful concepts presented during the Harvard Data Science Initiative AI Leadership Program was not about large language models, machine learning, or the latest AI tools. It was a simple five-step framework.

At first glance, the framework appears straightforward. In practice, it explains why so many AI projects fail and why others are able to create measurable business value.

The message is clear: AI is not the starting point. Business strategy is.

The slide below illustrates the journey from Business Goal → AI Value. It reinforces a principle that has become central to my work in AI governance and healthcare compliance: successful AI adoption does not begin with technology. It begins with purpose.

Step 1: Start with the Business Goal

Every successful AI initiative begins by answering one simple question: What problem are we trying to solve?

Too many organizations begin by asking which AI platform they should buy, which chatbot they should deploy, or how they can automate a process. Those questions matter, but they come too early.

Organizations should first define the measurable business outcome they are trying to achieve. For a healthcare organization, that may mean improving patient access, reducing claim denials, improving documentation quality, increasing provider efficiency, strengthening regulatory compliance, reducing administrative burden, or improving patient satisfaction.

Without a clearly defined objective, AI becomes an expensive technology searching for a purpose.

Step 2: Identify the Critical Decisions

Once the goal is clear, the next question becomes: What decisions must people make to achieve that goal?

This is where AI should provide assistance, not replace judgment. For example, if the goal is reducing claim denials, the organization must understand the decisions that influence denial risk. Is the documentation sufficient? Does medical necessity support the service? Are modifiers appropriate? Does the coding accurately reflect the encounter? Is prior authorization required?

These are not simply technology questions. They are operational, compliance, coding, and clinical judgment questions. AI can help surface information, identify patterns, flag inconsistencies, and support decision-making, but it should not independently make decisions that require accountability, context, and human oversight.

This is why human oversight remains essential, particularly in regulated industries like healthcare.

Step 3: Determine the Information Needed

Before AI can assist with decisions, organizations must understand: What information is required?

This is often where AI projects begin to struggle. Healthcare organizations frequently discover that the information needed to support a decision exists in multiple locations, including the Electronic Health Record, practice management system, claims system, internal policies, regulatory guidance, clinical documentation, audit reports, contracts, and sometimes only in the experience of long-tenured staff.

If that information is incomplete, inconsistent, outdated, or poorly governed, AI cannot compensate. In fact, it may make the problem worse by producing outputs that appear confident but are based on unreliable or incomplete information.

As I often say: AI will amplify whatever information you give it, good or bad.

Step 4: Identify the Right Data

Only after understanding the required information should organizations ask: Where does that information come from?

This includes both structured and unstructured data. Structured data may include CPT® codes, ICD-10-CM codes, claim history, payment data, provider information, and quality metrics. Unstructured data may include clinical documentation, policies and procedures, medical records, emails, audit reports, meeting notes, and regulatory guidance.

This step is frequently underestimated. Organizations often assume they have good data because they have a lot of data. In reality, many healthcare organizations are working with duplicate information, conflicting versions of documents, outdated policies, missing metadata, poor document governance, inconsistent naming conventions, and knowledge scattered across departments.

This is exactly why data governance must precede AI implementation. If an organization cannot trust the information feeding the AI system, it cannot fully trust the output.

Step 5: Apply Technology and AI

Only now does AI enter the picture.

Notice something important: Technology is Step Five, not Step One.

Once business goals, decision points, information needs, and trusted data have been defined, AI can begin creating value. At this stage, organizations can more responsibly explore solutions such as intelligent document retrieval, knowledge assistants using Retrieval-Augmented Generation, clinical documentation support, coding assistance, compliance monitoring, risk prioritization, predictive analytics, workflow automation, and executive dashboards.

At this point, AI becomes an accelerator, not a guessing machine. It is no longer being used because it is new or impressive. It is being applied to a clearly defined problem, using known information, trusted data, defined workflows, and appropriate oversight.

Why This Matters for Healthcare

Healthcare organizations often approach AI backwards. Many begin with the question, Which AI platform should we purchase?

The better question is: What business problem are we trying to solve, and do we have the trusted information needed to solve it responsibly?

After more than four decades working in healthcare compliance, payment integrity, coding, auditing, and regulatory oversight, I have learned that successful technology initiatives rarely fail because of the technology itself. They fail because organizations lack clear governance, reliable information, defined workflows, accountability, human oversight, and standardized decision-making.

AI simply exposes those weaknesses faster.

This Framework Aligns with the Seven Pillars of AI Readiness™

This five-step model closely aligns with the philosophy behind the Seven Pillars of AI Readiness™, the framework I have been developing following my participation in the Harvard Data Science Initiative AI Leadership Program.

The vision is to help healthcare organizations prepare for AI responsibly by focusing first on governance, trusted data, knowledge management, and human accountability, not technology alone.

The Seven Pillars emphasize that organizations should strengthen strategy, governance, data integrity, knowledge management, people and culture, technology infrastructure, and continuous oversight. Only when these foundational elements are in place can AI consistently deliver trustworthy, explainable, and defensible outcomes.

Final Thoughts

Organizations do not become AI-ready by purchasing the latest technology. They become AI-ready by understanding their business objectives, identifying critical decisions, organizing trusted information, governing their data, and ensuring that human expertise remains at the center of every important decision.

Technology should always be the final step, not the first.

Because at the end of the day:

AI does not create business value. Better decisions do.