Advances in the Practice of Horizon Scanning

midjourney_v8.2

There has been tangible progress in the practice of horizon scanning, particularly in automated scanning and exploration of the ways in which AI can potentially support the organisation and analysis of signals of change. Emerging thinking in this area suggests that progress depends as much on robust and auditable interpretation, traceability and usefulness, as it does on finding more signals. Whilst AI makes (some) signal collection, analysis, clustering and synthesis quicker, it raises many questions about what gets overlooked, how the unfamiliar is interpreted and the extent to which shiny AI output conceals weak evidence, conflates key low signal insights, flattens unresolved tensions or critical uncertainties.

There are a few reports which detail recent developments in horizon scanning practice that are worth a look - I’ve detailed five below and will keep adding to this list as I find more examples.

For those who want to jump to the punch, these were my two key thoughtstarters:

Key Process Takeaway

A credible AI-assisted scan should show what the system saw, what it changed or excluded, and how those choices were checked.

Key Scanning Question

Does AI help us to notice and examine what we would otherwise miss? or does it simply help us to efficiently organise what is already visible?

1. The 2026 OECD Report

Building Capacity in Technology Scanning Report 🔗

The report explores AI’s increasing role in helping analysts identify weak signals, detect emerging trends, rapidly scale analysis and provide policymakers with timely insights. In notes that using AI to identify the frequency of topics across the sample used enables a small team to interrogate an existing body of knowledge with their own strategic questions. However, frequency across reports shows us what is receiving attention without necessarily establishing strategic significance.

2. The UK Govt Approach

Weak signals and trend analysis: horizon scanning 🔗

This example linked 3 existing products - the ongoing horizon scanning tool, emerging trends report and the global megatrends report.

Two interesting pointers from this report:

• Scanners deliberately explored fringe areas and subcultures across global regions - conducting intentionally diverse discovery that was then linked back to whether the scan hits offered information, potential risks or research priorities, thus linking the scan hits to relevant decision making rather than just “here’s what’s happening”.

• The team also mapped emerging signals against their mega trends to ask whether they might accelerate or constrain the trends which reveals a useful question - does this signal of change reinforce a longstanding direction or suggest the ways in which it might change?

3. 2024 Paper: Enhancing the horizon scanning utility of futures-oriented systematic and scoping reviews 🔗

Eray Arda Akartuna, Shane D. Johnson, Amy Thornton, Futures, Volume 158, 2024, 103340, ISSN 0016-3287, https://doi.org/10.1016/j.futures.2024.103340.

Here the authors propose the NERC framework - neutrality, evidence, relevance and clarity with each dimension scored to enable comparison and mapping. The team also identified and extracted enablers, circumstances, stakeholders and risk characteristics which helped them shift from “this is happening” toward “this signal may enable this consequence for these actors”.

However this study used a limited set of sources and the authors noted that a key challenge lies in the approach’s accuracy for low-visibility signals (signals identified by a low number of publications). The proposed NERC framework favours (and assigns better scores) to signals of change or insights with existing evidence - often the opposite of (or atleast a critical part of), what we’re looking for in a horizon scan.

4. Cambridge University paper

Reporting the use of AI in horizon scanning: a brief communication on methodological standards 🔗

This paper explores the ways in which we can make AI-assisted scanning more auditable, focusing on reporting disclosures such as:

• Where AI influenced the scan

• Which exclusions were baked into the process and where they are visible

• The AI configuration (and chain of reasoning) not just the product

• The validation and human review in detail

• The benefits against previous processes.

Leaving us with an interesting and useful take-away:

A credible AI-assisted scan should show what the system saw, what it changed or excluded, and how those choices were checked.

5. The 2025 OECD-WEF Report

AI in Strategic Foresight: Reshaping Anticipatory Governance 🔗

This report distinguishes three patterns:

• AI for analysis augmentation

• AI as creative sparring partner

• AI integrated and customised into workflow

The report recommends:

Strategic foresight practitioners should invest in developing workflows that leverage AI to handle the “heavy lifting” of data processing and initial drafts, thereby freeing up time for higher-level analysis, interpretation and critical thinking.

Some key insights from the report:

1. Recognising existing patterns can crowd out genuinely unfamiliar developments.

1. Practitioners note AI’s limited ability to reason inductively and provide forward-looking analysis.

2. AI’s dependence on existing knowledge can make identifying unknown or low-probability signals difficult.

3. AI can help us identify and organise signals but it’s interpretation or analysis can pull us back into familiar narratives.

Fluid outputs can be shallow, derivative or wrong.

1. Output quality and trustworthiness are the most frequently cited challenges.

2. A polished AI output can potentially reach neat and tidy conclusions or skim over key horizon scanning questions, disguising the jump from an observed event to a speculative implication.

The information available to AI can potentially create systematic blind spots.

1. Report respondents raise english-language, western focus and historical data as challenges, along with the inaccessibility of private or emerging information.

Verification can consumer the time the automation saves.

1. The report notes that practitioners struggle to establish sources and cite the reasoning behind the outputs, leading to validation and fact-checking as an increased demand where audit work becomes the time-consuming human part of the equation.

Effective use requires methodological skill.

1. Automation doesn’t remove the need for the scanner to decide:

1. What counts as a signal for this scan?

2. Which sources and perspectives should be included?

3. What should be retained despite weak evidence?

4. Which analytical questions should guide interpretation?

Poor framing (just as in non-AI work), can be efficiently reproduced across a much larger scan collection.

Whilst AI can make scanning faster and broader, this does not necessarily make scanning more perceptive. The report is particularly cautious about AI’s ability to identify unfamiliar, low-probability developments and interpret potential disruptions which is the very aspect that makes horizon scanning (as opposed to trend summaries) useful.

While AI can enhance productivity, there is a risk of “deskilling” or over-reliance on the tools with significant limitations. AI’s ability to provide existing knowledge rather than forward looking perspectives means that human foresight expertise may be more critical than ever to identify unknown, low-probability signals and potential disruptions.

The key question that surfaced for me in reading this report was:

Does AI help us to notice and examine what we would otherwise miss? or does it simply help us to efficiently organise what is already visible?