Aug 31, 2026
Michael Becker, CEO, GeoSure
Corporate travel risk management has operated at country and large-city resolution for decades. That was not an accurate reflection of where risk lives, but of the cost of human analysis, the time required, and the precision of the information available at the time. The problem was not treated as a solvable one.
AI Agents diminish those constraints. This article looks at what happens when the unit of assessment moves from large ones — countries and large cities — down to smaller ones: neighborhoods and blocks. It covers what that addresses, including cost and the quality of decisions, and what it does not.
A country or mega-city rating is a compression: it takes a distribution of conditions across millions of square kilometers and returns a single ordinal value.
What the compression discards is the variance that matters most. Criminological research on crime concentration has found consistently, across many cities and several decades, that a small share of street segments accounts for roughly half of a city's recorded crime, and that the concentration is mostly stable over time. While there may be structural reasons behind these inequities, the purpose behind this brief is not to discuss or attempt to fix those inequities, but rather to identify pattern recognition scientifically. The same pattern can hold for health infrastructure, civil unrest exposure, and transportation safety.
A large-geography rating was never designed to represent that structure; it was designed around what a staffed analyst desk could reach and maintain — a couple hundred country, capital-city, and port-of-entry profiles, revised on a periodic cycle and delivered as narrative documents, perhaps supplemented with westernized news media reporting or governmental warnings.
The granularity of the product matched the capacity of the organization producing it, rather than the geography of risk distribution or holistic view of localized insights.
Country-level intelligence supports specific types of decisions:
It does not support decisions made once travel is approved:
Country-level ratings are not wrong — they are useful for comparing countries. Limitations come from applying one measure to a decision it cannot inform, which produces uninformed choices in both directions: it restricts travel that presents no elevated exposure, and approves travel into specific districts that do.
Cost per assessment. A briefing that took several analyst-hours is now generated on demand, and when marginal cost approaches zero, assessment stops being rationed to mainly high-risk destinations and becomes available for every trip.
Update latency. Periodic revision cycles give way to continuous ingestion: systems built for tens of thousands of sources across 28 global regions in 90+ languages, surfacing thousands of relevant items across different threat categories each day. Users can then work this information across several dimensions: contextual queries, scores, score changes, and daily briefings/alerts.
Interface. A fixed report answers the questions its author anticipated, while an agent answers the question asked at point of need — the difference between receiving a Colombia country profile and asking whether a specific district in Bogotá suits an evening meeting planned for that night.
Analyst leverage. Higher resolution lets analysts probe deeper and ask their own questions, which produces faster and more timely decisions.
The result is not a better report but a change in the addressable unit of decision.
Country ratings produce binary outcomes: a destination is approved or prohibited/restricted.
Exposure becomes measurable. When assessment exists for every itinerary rather than only higher-risk itineraries, an organization can compare exposure across trips, business units, and personnel, which makes it a management input.
Coverage extends to ordinary places. Travel volume concentrates in countries rated as unremarkable, so a program that attends only to high-risk destinations leaves its largest volume of traveler-days unmonitored. Resolution is a question of coverage in ordinary places, not only precision in dangerous ones.
Duty-of-care obligations in most jurisdictions are measured against a reasonableness standard: what a prudent employer could reasonably have done with the means available.
Reasonableness is not fixed; it moves with what is available and affordable. When neighborhood-level assessment could not be obtained at scale, no organization could be faulted for relying on country-level intelligence, but as that capability becomes obtainable at commodity cost, the reference point shifts with it.
ISO 31030:2021, the international guidance standard for travel risk management, does not certify organizations and does not prescribe a resolution. It asks that risk be treated proportionate to context, and proportionality implies granularity appropriate to the decision — so where lodging decisions matter, a country rating is not proportionate.
The general finding in risk communication research is that vague, indistinguishable, repeated warnings produce habit, while guidance that is specific, personally relevant, and paired with a clear action is more likely to be followed.
Resolution supplies that specificity: a traveler told that a route carries elevated theft exposure after dark, with an alternative supplied, has an action, while a traveler told the country is at Level 2 does not.
Who is traveling matters as well, since a city can score well on physical safety and poorly on women's safety, or well on both and poorly on LGBTQ+ safety. GeoSure scores six dimensions independently — physical safety, theft, women's safety, LGBTQ+ safety, health, and political freedom — because the guidance depends on the traveler.
Data availability. Sub-city assessment depends on what is reported locally, and where reporting is thin it is partly inferential — a precise output is not the same thing as a precise measurement.
Media and OSINT coverage bias. News volume tracks media presence rather than incident rate, so any system reading news must correct for coverage density or it will overweight well-covered cities.
Response infrastructure. Intelligence is not extraction, medical assistance, or a phone line that answers at 3am, and better data improves prevention rather than response.
Accountable judgment. Agents produce assessments, but decisions about executive travel, site security, and crisis posture remain the responsibility of humans.
Model error. Any system summarizing at volume produces errors, so what matters is whether each claim traces to a dated source a human can check.
The last one is where the shift becomes concrete: reported enterprise contracts with incumbent providers fall between $100,000 and $200,000 a year for country and city-level intelligence delivered on a periodic cycle, a price that reflects the cost of maintaining analyst capacity rather than the cost of producing an assessment once compilation is automated at scale.
GeoSure scores six independent risk dimensions across approximately 200+ countries, 1M cities, neighborhoods, and micro-hoods, updated continuously, with deep briefings generated on demand for any of them. Scoring draws on authoritative government and institutional data — national police statistics, statistical agencies, health ministries, international conflict and crisis databases, and satellite-based environmental monitoring — developed over more than a dozen years of methodological work.
It reaches you two ways: the Model Context Protocol, plugging directly into your existing agent infrastructure, or through GeoSure's own agents, which reason across the full intelligence picture and produce briefings on demand.
To see a briefing built for your organization's travel footprint — contact info@geosure.ai.
