The Companies That Win with AI Won't Necessarily Have Better AI
Why trustworthy business data, not better AI models, is how to compete with AI
Everyone seems to be asking the same question:
Which AI platform should we choose? GPT? Claude? Gemini? Open-source? It's an understandable question—but increasingly the wrong one.
The capabilities of leading foundation models continue to improve while becoming broadly available to organizations of every size. Over time, most companies will have access to remarkably similar AI technology.
What they will not have is the same customer relationships, operational history, proprietary workflows, institutional knowledge, and business data.
As AI models become commodities, trustworthy proprietary data becomes strategic infrastructure.
For many software companies—particularly small and mid-sized businesses—that shift may become the most important competitive change of the decade.
In an earlier article, Is Data Really the New Gold?, I argued that organizations should treat data as a strategic asset capable of improving operations, strengthening products, and even becoming a product itself.
The next question is more important:
Is that data actually worthy of AI?
AI Doesn't Fix Bad Data—It Scales It
Artificial intelligence doesn't magically transform poor information into good decisions.
It magnifies whatever already exists.
If the underlying data is inaccurate, incomplete, inconsistent, outdated, or lacking business context, AI simply makes those mistakes faster—and often with greater confidence.
A misleading dashboard may influence one manager. But a poorly grounded AI agent can influence thousands of customer interactions, purchasing decisions, pricing recommendations, or operational actions every day.
That raises the standard for enterprise data.
Organizations deploying AI should evaluate six characteristics:
Accuracy: Is the information correct?
Completeness: Are critical fields, documents or histories missing?
Consistency: Do systems define customers, products and revenue the same way?
Timeliness: Is the information current enough for the decision?
Governance: Are ownership, permitted use, lineage and access understood?
Context: Does the AI understand what the data means in this business?
Context is often the most overlooked.
A language model may understand English perfectly while having no understanding of your company's pricing exceptions, implementation practices, customer relationships, contract terms, or product architecture. Without that context, AI can generate answers that sound persuasive while being operationally wrong.
Competitive Advantage Is Moving Up the Stack
For years, companies competed by digitizing business processes. Then they competed by analyzing data. Increasingly they will compete by enabling AI to make better decisions.
That changes what creates differentiation. Most organizations can purchase similar AI models. They cannot purchase your customer history. They cannot purchase your implementation experience. They cannot purchase years of support interactions, product usage patterns, engineering decisions, pricing knowledge, or operational expertise.
Those assets become valuable only when they are trustworthy enough for AI to use confidently.
When Trustworthy Data Creates Value
John Deere's See & Spray platform illustrates this principle. Its competitive advantage is not simply computer vision.
The system combines agronomic knowledge, field imagery, equipment telemetry, application history, and outcome data to determine exactly where herbicide should be applied.
Because the underlying data is reliable and continuously refined, the AI doesn't merely generate insights—it performs precise physical actions.
Deere reported customers used the technology on more than five million acres during the 2025 growing season while reducing herbicide use by nearly 50 percent on average.
The lesson extends far beyond agriculture. AI becomes valuable when proprietary operational data allows it to make decisions competitors cannot easily replicate.
The Same Principle Applies to Software Companies
Consider a growing B2B SaaS company implementing an AI customer-support assistant.
The foundation model is excellent.
Unfortunately, customer contracts reside in one system, implementation notes in another, support tickets contain inconsistent terminology, and product documentation hasn't been updated in months. The AI confidently delivers different answers depending on which source it happens to retrieve. Customer confidence erodes almost immediately.
Now imagine the same company after reconciling customer records, standardizing terminology, improving documentation, and connecting implementation history with support interactions.
Nothing about the underlying AI model changed. Only the quality of the business data improved. Yet the AI suddenly becomes faster, more accurate, more consistent, and significantly more valuable.
The competitive advantage came from preparing better information—not buying a better model.
Sometimes the Data Isn't the Problem
Zillow Offers provides a different lesson.
The company possessed enormous amounts of housing data and sophisticated pricing models. Its challenge wasn't poor data quality. It was asking AI to predict a future that became fundamentally unpredictable while committing billions of dollars of capital.
Historical patterns could not adequately forecast a rapidly changing housing market.
The lesson is equally important. AI readiness isn't simply about clean data. It is also about knowing where prediction uncertainty becomes too great to automate high-risk decisions.
Sometimes the question—not the data—is wrong.
Most Companies Are Closer Than They Think
Many executives assume becoming "AI-ready" requires a massive data-transformation initiative.
Usually it doesn't. Most small and mid-sized software companies already possess the proprietary data needed to create meaningful competitive advantage. What they often lack is consistency.
Duplicate customer records.
Conflicting product definitions.
Incomplete documentation.
Business knowledge trapped in email or inside experienced employees' heads.
Disconnected systems.
Rather than attempting to clean every record in the organization, companies should begin with a handful of high-value AI use cases.
Identify the information required for those use cases. Improve that information.
Measure the business impact.
Then repeat.
Data readiness should become an ongoing management discipline—not a one-time IT project.
Beyond Cleaning Data
Traditional ETL tools were designed primarily to move and reshape information for reporting. AI raises the standard.
Organizations increasingly need continuous monitoring of data quality, business context, governance, and model performance. The objective is no longer simply moving data. It is ensuring AI continues making trustworthy decisions as the business changes.
(The technical architecture for building this capability is the subject of a forthcoming Marlborough Street Partners paper, Beyond ETL: Building an AI-Ready Data Quality Operating System.)
Five Questions Every Leadership Team Should Ask
Do we know which data matters most to our highest-value AI initiatives?
Can we trace that data to reliable sources and accountable owners?
Is it accurate, complete, and timely enough for AI-assisted decisions?
Does it include the business context necessary for AI to interpret it correctly?
Can we demonstrate that improving data quality improves business outcomes?
If leadership cannot confidently answer those questions, the organization should be cautious about moving from AI assistance to AI autonomy.
The greater the financial, regulatory, or customer consequence of a decision, the higher the standard should be for trustworthy data.
The Real Competitive Advantage
Every company will soon have access to powerful AI.
Not every company will possess trustworthy proprietary information that allows AI to make better decisions than competitors.
Twenty years ago, organizations competed by digitizing business processes. Ten years ago, they competed by analyzing data.
Over the next decade, they will compete by enabling AI to act on trustworthy proprietary knowledge.
The winners will not necessarily own the smartest AI.
They will own the data that makes AI the smartest.
By Duane Kotsen, Partner
Example Sources:
Zillow Group, Q3 2021 results and plan to wind down Zillow Offers, Nov. 2, 2021.