How one risk manager is using AI to change the face of horizon scanning

Imagine it’s 1976. The semiconductor industry is just beginning to gather steam.

A young engineer at Texas Instruments, Morris Chang, notices something the industry giants seem to overlook: the future belongs to companies that focus solely on chip manufacturing and do not compete with their clients.

This insight – rooted in a deep understanding of market risks and opportunities – leads to the creation of TSMC, a company that now controls 64% of the global semiconductor manufacturing market.

It outpaced classic leaders like Intel and Samsung, turning the entire industry upside down.

TSMC’s story is a vivid example of how timely detection and proper interpretation of signals can not only mitigate risks but also pave the way to global leadership.

Let’s come back to our current reality. Of course, we’d love to have the same sharp “instinct” that Morris Chang had, to see fundamental market shifts in advance.

But today’s business landscape is evolving at a pace that may be even faster than the semiconductor industry once did.

Technological breakthroughs, geopolitical changes, new twit from Trump or Musk, the rise of environmental and ESG considerations, supply chain transformations – risks take on countless forms, while traditional assessment methods often seem cumbersome and struggle to keep up with reality.

Now, imagine you’re a consultant heading to a meeting with a major client or a risk manager preparing for an executive committee meeting. You need more than just yesterday’s data; you need to quickly and thoroughly analyze market signals to pinpoint emerging threats and opportunities.

That’s precisely what Horizon Scanning helps you to do. It is a methodology for early detection and analysis of emerging risks and opportunities across various areas of your market environment.

Do it yourself or use a multi-agent system?

Running a horizon scan requires a risk professional to first map out the target company business model: where the company operates, which products sells and markets services, the production capacity and location, competitors, financing, regulatory environment and so on.

The risk manager can then apply any or all of the common techniques like Porter’s 5P, PESTLE, and scenario analysis to identify factors that may disrupt key decisions or present new opportunities.

By integrating data from reputable sources—such as market intelligence, regulatory updates, emerging technology trends, economic forecasts, and competitor announcements—the risk manager can build an informed view of how shifts in the external environment might affect the company’s assumptions.

The aim is to maintain a clear line of sight between emerging signals on the horizon and the tactical or strategic moves that leadership can make to adapt quickly, protect core advantages, and seize new growth avenues.

We’ve streamlined this process by using GenAI. We developed a platform that leverages multiple agents, each focusing on its own area of expertise – from geopolitical disruptions to competitive moves, from new technological breakthroughs to changes in consumer behavior.

Then, they bring it all together to generate a comprehensive risk / opportunity list.

A look at our technical workflow

To orchestrate these agents, we rely on a workflow designed which strings together several steps.

When the user enters a query in the personal web dashboard, the system initiates a search for data in publicly available sources and news aggregator APIs.

At the same time, it retrieves relevant information from a large repository of analytic reports, trend forecasts, and future megatrends (RAG analysis).

One of the intermediate versions of the workflow.

Once these data sets are enriched and pre-processed, specialised agents each delve into their area of focus. Each specialised agent focuses on its own domain to unearth relevant signals.

Agents are set up with specific prompts to avoid missing critical points. These insights then flow back to a central “conductor” agent.

This conductor compiles the results into an integrated report, which you can explore in your personal account. You can explore identified risks, compare sessions, and export the results to Excel or PDF for further analysis or board presentations.

Lots of back testing to guide our agents

A critical element of making this system work is the design of our multi-phase prompt. While we’re not releasing all of our proprietary details, we can explain the broader structure.

First comes the input and context, which includes specifics about the company or product to analyse, along with relevant market conditions and the user’s objectives—say, identifying potential new markets or alternative technologies.

Next, the main goal of analysis is clearly stated, so the system knows exactly what success looks like (for instance, “Find regulatory red flags that could undermine EV technology in the next two years”).

Admin management panel for configuring and editing AI agent prompts.

The prompt also orchestrates a step-by-step logic for each agent. During Phase 1, agents focus on spotting direct competitors, substitutes, possible regulatory threats, and so on.

In Phase 2, each agent formulates and collates key risk factors, opportunities, and emerging signals into a consolidated list of risks and opportunities.

Then, in Phase 3, everything is formatted into a structured JSON output—keeping the data precise and machine-readable without extraneous commentary.

This process is intentionally strict to maintain accuracy. Agents that drift off-topic or insert unverified data are reined in by additional guardrails that enforce both factual consistency and formatting standards.

It’s a carefully tuned balance: we want enough flexibility for agents to capture late-breaking trends, but enough structure to filter out “hallucinations”.

Also, in the process of finding the perfect prompt, we once again saw that small changes in wording can lead to completely different results. We tried several versions of the prompt architecture in different OpenAI and Anthropic LLMs to see which one gives the most accurate analysis.

Real-world spotlight: Tesla

To understand how the system functions, imagine how Horizon Scanning can help identify emerging risks using Tesla – a company widely regarded as a trendsetter in the electric vehicle (EV) industry – as an example.

Here’s just 1% of what the Horizon Scanning system has identified.

User dashboard displaying the history of risk assessments and identified risks.

Today, speed and accuracy of analysis determine who will be tomorrow’s leader. Morris Chang spotted a market shift in time, and TSMC has dominated the industry for years since.

What emerging trend can you spot today? What unexpected threat might turn into your best opportunity?

Horizon Scanning is your chance to protect yourself against risks no one has noticed yet – and respond before the competition does. Just as Morris Chang once spied a critical market shift that shaped TSMC, Horizon Scanning helps you catch those shifts happening right now. Which new trend can you uncover? Which potential risk might become your biggest advantage?

Give it a try at radar.riskacademy.ai. Explore how quickly it collects and synthesises relevant data, then generate a comprehensive risk report of your own. You can also contact us for a one-on-one demo if you want to know how we built the platform.