In today's AI-first world, most organizations are racing to implement large language models and expand their compute capacity. Yet the fundamental question leaders should ask themselves and their teams is: If you could restart your AI journey with everything you know now, what would you do differently?

At Human Managed, our path began with data and analytics before evolving into an intelligence platform. This evolution revealed a consistent truth: AI's greatest challenge isn't technical complexity—it's organizational clarity.

The Intelligence Architecture Framework

According to the Artificial Intelligence Index Report 2025 by AWS, while AI investment rebounded and AI is becoming a central driver of business value, most companies are still early in their AI journeys. About half of organizations using AI in key functions report cost savings or revenue gains, but these benefits are often modest—typically less than 10% cost savings or under 5% revenue increase. This indicates that although AI is delivering financial impact, the scale of value is still generally low for most companies.

I believe that the gap exists because most are investing in compute and models without restructuring how decisions flow through their organization. Here are four principles to transform your approach:

1. Prioritize Decisions Over Models

Most AI initiatives start with technology selection rather than decision mapping. The Gartner Hype Cycle 2023 notes: "Analytics demand is shifting from predictive to prescriptive capabilities... AI systems need to act autonomously by understanding what impact actions will have." This shift reinforces that decision-focused AI delivers greater value than model sophistication alone.

Begin by cataloging which decisions in your enterprise are:

• Frequent enough to benefit from automation
• Impactful enough to justify investment
• Structured enough for AI augmentation

Only after this clarity should you select your technical approach.

2. Connect Signals In Context

While the sheer volume of data available to enterprises has grown exponentially—especially with the rise of generative AI—McKinsey highlights that data quality, contextual integration and relevance to specific business use cases are far more critical to unlocking AI’s true value than simply accumulating more data. For example, McKinsey notes that generative AI’s power lies in its ability to leverage unstructured data (such as text, images, and videos), which comprises about 90% of all data, but this data must be carefully cleansed, tagged and integrated with structured sources to be useful.

The transformative question isn't "How much data do we have?" but rather "Have we structured our signals to reveal meaningful patterns?" For example, at Human Managed, data-integrated systems that connect user behavior, access patterns and asset criticality detect threats 60% faster than those analyzing each data stream in isolation.

3. Build Day-2 Operations From Day One

Studies show that up to 80% of AI projects fail, often not during initial development but when integrating AI into operational workflows and scaling across the enterprise. Research from Stanford, Harvard, and RAND highlights that many failures stem from misalignment with business needs, insufficient infrastructure and lack of ongoing model maintenance. While precise resource allocations vary, successful AI implementations increasingly recognize the critical role of AIOps—frameworks for continuous model monitoring, feedback and improvement—in sustaining AI value over time.

Intelligence at scale requires more than data scientists; it demands operational maturity to manage the complete lifecycle of your AI systems.

4. Encode Institutional Knowledge

The secret weapon in intelligence architecture is systematically capturing your organization's existing expertise. Reputable research from institutions like the MIT Media Lab and Stanford highlights that AI systems incorporating codified business logic and domain-specific frameworks tend to outperform generic large language models on enterprise-specific tasks. By embedding contextual knowledge and aligning AI capabilities with domain requirements, these systems achieve greater accuracy, interpretability and operational effectiveness—key factors for delivering meaningful business value.

Consider how you're transforming tacit knowledge—threshold values, escalation criteria, compliance requirements—into structured formats your systems can learn from.

From AI Consumers To Intelligence Architects

The most forward-thinking executives are no longer just purchasing AI capabilities; they're redesigning their organizations to become intelligence-native. Deloitte’s 2024 Tech Trends report highlights that organizations viewing AI as an architectural challenge—integrating AI deeply into their core systems, processes, and talent strategies—are better positioned to scale AI adoption enterprise-wide. While exact figures vary, Deloitte stresses that modern, scalable architecture and solid fundamentals are critical enablers of AI’s transformative potential.

This mindset shift requires leaders to ask different questions:

• Which decisions truly drive organizational value?
• What signals must be connected to inform these decisions?
• How can we encode our best thinking into systems that learn?

By focusing on these fundamentals rather than chasing the latest model capabilities, you build intelligence that compounds over time rather than degrades.

If you want to explore this approach further, I've developed a practical handbook on decision intelligence architecture available for anyone.

－

This article was originally published on Forbes Technology Council on June 26, 2025