Beyond the Tip of the ICEberg: Revealing Hidden AI Value

At NuroShift, we work at the frontier of Human–AI collaboration. One of the biggest misconceptions we encounter is the belief that AI can magically extract meaning from any dataset with a single prompt.

In truth, AI is only as insightful as the context you provide. Data, like an iceberg, conceals most of its story beneath the surface. The visible indicators, such as risk scores, incident logs, vulnerability counts-are merely the tip. The true insights lie below:

• Misaligned controls
• Incomplete governance processes
• Underreported human behaviours
• Root-cause dependencies within systems and policies

To help AI see beneath the surface, we developed a prompt engineering methodology called ICE — Investigate Connect Explain. This lightweight methodology helps surface hidden context and guide AI reasoning more effectively.

Prompt Engineering with ICE

I – Investigate: Unearth the backstory behind the data.

Don't just prompt AI to 'assess risk exposure'-ask what systemic or contextual factors might have shaped the reported metrics. Was there a recent cloud migration? A shift in authentication protocols? A policy that wasn't enforced?

• Prompt Example: "Summarise the potential impact of our multi-factor authentication rollout in March 2024 on user behaviour, audit logs, and phishing resilience."

• Tip: Reference SIEM data, policy timelines, change management logs, or staff training schedules to provide contextual input.

C – Connect: Identify relationships between data points.

Has a rise in vulnerability reports corresponded with a new CI/CD pipeline? Are third-party

assessments driving internal control updates? AI can detect latent patterns-but only if you explicitly connect the dots.

• Prompt Example: "Evaluate whether the spike in medium-severity vulnerabilities in Q2 aligns with our third-party vendor onboarding and asset registry expansion."

• Tip: Use knowledge from your threat model or asset inventory to expose risk interdependencies.

E – Explain: Translate your insights into contextual scaffolding for AI.

Avoid vague prompts like "Analyse this vulnerability report." Instead, explain what the AI should consider-risk appetite, recent remediation activity or residual risks?

• Prompt Example: "Analyse this risk report considering that three of the flagged systems were already scheduled for decommissioning, and that compensating controls were put in place for the remaining exposures."

Exploratory Prompting: When You Don’t Know What You Don’t Know

Sometimes you sense risk misalignment, but lack precise insight to construct a detailed prompt example or tip. In these cases, ask AI to hypothesise based on patterns.

• Prompt Template: "What unmonitored system behaviours or misconfigurations might explain recurring lateral movement attempts across segmented networks?"

This elevates AI from a reporting tool to a cyber research partner-surfacing unseen risk factors and sparking human-led investigation.

Why ICE Matters: From Analyst to Amplifier

When combined with techniques like Chain-of-Thought prompting—which guides the model through multi-step reasoning—or Tree-of-Thought strategies for branching logic, ICE turns AI into a genuine insight amplifier.

An ongoing conversation with an AI isn’t flat — it’s more like a 3D presentation, where your iterative prompts, corrections, and clarifications build on each other. The chat history becomes layered context that the model remembers and applies. ICE helps you shape that context with intention.

But prompt engineering is never fire-and-forget. It’s a craft. The best results come through iteration, reflection, and ethical vigilance.

A Word on Ethics

Always question what your prompts imply. Does your framing risk reinforcing bias? Are you assuming causation where only correlation exists? Use context to inform—not manipulate—the model’s perspective. Responsible prompt engineering means steering AI to clarity, not conjecture.

Closing Thought

Next time you engage AI, don’t just fire a prompt at the data tip. Dive beneath the surface:

• Investigate
• Connect
• Explain

ICE it - That’s how you unlock the full potential of AI.

We’d love to hear from fellow explorers: How do you build context in your prompts? Let’s share, refine, and level up—together.

‍