Future-proofing Geospatial Intelligence: How Ordnance Survey is Embedding AI Responsibly

Artificial intelligence is changing how organisations collect, interpret and distribute data. It presents the most powerful and practical tool available to enhance our mapping capabilities, enables us to reach more users and deliver faster, smarter insights to customers and partners. Paper cartography continues to have its place, but as the world modernises, the need for real-time geospatial intelligence shines through. To do this at a national scale, AI is essential. But so too is a need to deliver data responsibly.

AI can speed up insight and extend reach, but it can also produce outputs that look convincing while carrying uncertainty. Small classification errors can be harmless in isolation, but if they are repeatedly written into trusted datasets, the consequence becomes distorted across planning, infrastructure and public services. Our task is to use AI to amplify capability, while preventing small errors from becoming systemic. But so too is a need to deliver data responsibly, with transparency and provenance.

Meeting the Challenge of Scale

The landscape we map shifts constantly. The OS National Geographic Database holds over 600 million features, updated 30,000 times every day. At a foundational level, we’re leveraging the power of AI to detect changes and update our maps rapidly, and at scale. This allows for near real-time updates that identify shifts in land use, infrastructure and buildings – all essential data captured and made available to councils, businesses and utilities providers to name a few examples.

Manual methods cannot keep pace with that volume. Machine learning and computer vision let us extract features from aerial and street-level imagery, flag changes and prioritise where human review is required. Automation gives us the coverage and speed that modern decision-making demands, while allowing human specialists to concentrate on the complex cases where judgement matters most.

Guardrails for Reliable Data

AI errors behave differently from human errors. A single human misclassification is limited in scope and consequence, but an algorithm can repeat the same mistake across thousands of features. To prevent that, we build layers of verification into our processes to ensure trustworthy data.

Data validation combines automated checks with human review to catch anomalies before they enter authoritative national datasets. At OS, we apply confidence scoring to AI-derived data to communicate the probability behind the AI classification. We document provenance and make any uncertainty visible, so customers understand what the data implies and where caution is needed. This approach reflects our Responsible AI Charter and ensures that model behaviour is well understood and communicated. In practice, this means our users can understand not just what the AI says, but how certain it is. These checks are a critical part of making the data useful and dependable, enabling us to scale without sacrificing the reliability our partners expect.

From Automation to Autonomy

Today we publish many theme-specific datasets, covering transport networks, the built environment, waterways and the natural environment. The richness of those layers is powerful, but it can be difficult for non-specialists to see how everything connects.

Generative and agentic AI are beginning to bridge that gap by translating complex geospatial data into plain language and actionable insight. This brings with it an opportunity to democratise access to our data, enable natural-language requests, and empower users to build exactly what they need, without the requirement for specialist geospatial background.

A neighbourhood planner, for example, could ask a simple question about suitable sites for a new school, and receive an answer that combines catchment analysis, transport access and land availability. Another practical application is roof analysis. Using computer vision we can classify roof material, measure aspect and orientation, and estimate solar potential for individual properties. This capability supports green-energy planning and grant programmes by turning imagery into quantified, decision-ready information. These examples reflect how we are productionising ML workflows to deliver national-scale datasets with embedded quality assurance.

Building AI Confidence Through People

None of this works without people who can interpret AI outputs and question them when appropriate. AI is a force multiplier for human expertise. It should reduce repetitive work and surface tasks that require human resolution.

While efficiency is a key driver behind AI use, it’s essential that we do not lose the human equalities that bring depth, creativity and value to our work and relationships with clients and partners. Preserving the culture of craftsmanship and judgement that has defined OS for generations is central to how we adopt new tools. Our internal AI working group and change management team play a key role in upskilling staff and embedding AI responsibly.

We are building an AI Community to share updates and best practices, developing an AI Guild to democratise opportunity, and identifying AI Champions across business functions to support peer learning and responsible adoption. An AI Accelerator is also being established to drive experimentation and capability-building, alongside curated learning pathways tailored to different role families.

Looking ahead, OSs strategic workforce plan includes expanding AI capability over the next four years. This includes an academy model combining graduates and apprentices, with rotations across functions to develop well-rounded individuals. We are also exploring early career pathways that span product, marketing, consultancy and customer disciplines, while continuing to invest in our core data science and engineering capabilities.

A commitment to training is essential so that we can ensure that we understand model behaviour, interrogate results and apply domain knowledge with strong focus on quality, risk management and IP protection.

Building a Resilient Future

Embracing AI at OS is not about adopting technology for its own sake. It is about making sure Britain’s geospatial intelligence keeps pace with a changing world while maintaining its accuracy and authority.

That means building resilience into both our technology and our culture. Technical guardrails protect the quality of our data, while cultural transformation ensures our people can work confidently alongside AI. And by integrating emerging tools like agentic AI, we are preparing our organisation and our customers for the next wave of innovation, where geospatial intelligence is conversational and embedded into everyday decisions.

We are mapping more than today’s landscape. We are shaping how Britain navigates the future, with AI used to extend capability while protecting the reliability our partners and the public need.