Our work

A probabilistic prior authorization agent for a US health insurer

A regional US health insurer processing 1.4 million prior authorization requests per year was losing time, money, and clinical goodwill to a broken triage process. Average decisions took 6.2 business days, 41% of denials were overturned on appeal, and 62% of ultimately approved cases were being reviewed by clinicians who never needed to see them. We built Sophia, a probabilistic decision-support agent that reads the clinical package, builds a calibrated probability distribution over likely outcomes, and routes cases to the right reviewer at the right time with the right information.

Read case study

When doctors can’t read doctors’ handwriting: machine learning in medical indemnity

One of the UK's largest medical defense organizations was bleeding money from its claims reserving process. Its claims handlers, clinicians with legal training, were systematically over-reserving on new notifications. Each individual estimate looked prudent. In aggregate, around £38 million sat in claims reserves earning base rate instead of the 5–7% it could have returned in long-term portfolios. We built a settlement estimation tool trained on 50 years of medico-legal case history that combined a custom system for reading old scanned documents, a prediction model that learned settlement patterns from past cases, and a privacy layer that stripped patient-identifiable information before anything else touched the data.

Read case study

Working within the UK government to replace 750 websites with one

Between 2010 and 2012, the UK government consolidated more than 750 separate departmental websites, each with its own design language, content governance, and stakeholder politics, into a single domain: Gov.uk. Iain was a senior project manager on the program, working across what became the Government Digital Service to define and implement the ten design principles that would govern the new platform. The work involved translating user research into project-level decisions across dozens of service areas, brokering compromises between legal accuracy and readability, and pushing back against architectures that optimized for internal convenience over user experience. The results speak for themselves: task completion rates jumped from 46% to 61%, the average task time dropped by a full minute, and Gov.uk is now the second most-used government website on the planet, serving 23 ministerial departments, 20 non-ministerial departments, and over 410 agencies and public bodies.

Read case study