Building AI that adapts to mission-critical complexity
The team at Breast Cancer Now were sceptical about AI's ability to accurately transcribe complex patient surveys across 20+ different NHS form structures. Would it be accurate enough for the researchers to feel confident?
When Breast Cancer Now approached us about their Service Pledge Programme, they faced a constraint that threatened to limit their impact across the NHS: the manual transcription bottleneck. Their small team of four senior officers ran 18-month projects with NHS hospitals, generating enormous amounts of patient and staff feedback through paper surveys. Processing was very time-consuming, preventing both the strategic conversations that drove real change and any expansion of the programme to additional hospitals.
The team wanted to reach far more NHS trusts and therefore more patients, but the transcription workload made this impossible without hiring additional staff - something charity budgets rarely allow. They were fairly sceptical about AI's ability to handle this work accurately. They'd seen AI tools struggle with simpler tasks, and their surveys involved complex layouts, varied question formats, and nuanced patient responses. The risk of errors was unacceptable - these were insights that directly influenced patient care.
The technical challenge was to create AI that could achieve the transcription accuracy needed to overcome this well-founded scepticism, whilst enabling the team to increase their reach without increasing headcount.
When simplification means mission compromise
Each NHS trust had evolved slightly different survey formats. Patient advocates had input into question wording and layout. Each form contained approximately 400 data elements across different question types, often spanning multiple pages.
We knew we couldn't take the conventional approach of standardising forms or forcing digital collection. The patient cohort, often older and dealing with serious illness, consistently preferred paper to digital. Standardisation would have disrupted collaborative relationships with NHS trusts and digital surveys would have excluded or burdened patients who preferred paper-based feedback. Simply hiring more transcription staff wasn't viable within charity resource constraints.
Instead, we built AI sophisticated enough to handle their reality - processing over 20 different form structures across multiple institutions and patient populations.
Engineering for organisational reality
Each NHS trust uses slightly different question wording or sequencing to meet their specific needs. We created comprehensive mapping files so the system could recognise equivalent data points and ensure consistent categorisation.
We built dual-schema systems - per-page schemas to process individual form pages, and per-question schemas to reconstruct complete responses across multi-page questions. Extensive mock generation capabilities enabled rigorous testing without compromising sensitive healthcare information. The result was a processor maintaining high levels of accuracy whilst preserving the nuance of Breast Cancer Now's collaborative approach. The team's initial scepticism was overcome through demonstrable accuracy on their actual data.
From operational efficiency to strategic capability
The system has addressed some of the transcription bottlenecks that had constrained the programme's growth. There is still work to be done to address some of the limitations of the non-digital infrastructure and build confidence in the reliability of the AI.
The hope is that the same four-person team will be able to work with considerably more hospitals without compromising the quality of their engagement, meaning more patients benefit from the Service Pledge Programme's improvements to breast cancer care. But it's not just about time saved - with data now in a structured format, the team can spot patterns across trusts that weren't visible before, enabling comparisons and identifying factors that correlate with better patient outcomes.
What mattered to Breast Cancer Now - the collaborative approach, the trust they'd built, patients being able to use paper if they preferred - none of that needed to change. We just removed the manual transcription work that was eating up their time.the team both the capacity to reach more hospitals and patients, and new analytical capabilities
When real-world infrastructure meets AI ambition
The AI can handle complex survey forms accurately and quickly. But the existing scanners can only process a handful of forms at a time - so we'd solved one bottleneck and created another upstream which is a bit of a reality check.
This is why proof of concept work matters. You find out what actually gets in the way before you've committed significant budget. The AI worked as promised, but getting the full benefit means the rest of the process needs to adapt. For any charity exploring AI, it's worth looking at the whole pipeline, not just whether the AI can do the job. Sometimes the real cost includes upgrading a scanner.
Technological change continues to accelerate but only a quarter of charities say they feel prepared to respond to the opportunities and challenges. Let's close the opportunity gap together.
