Two years ago this would have been a six-figure project

Breast Cancer Now's Service Pledge programme works with NHS hospitals to improve breast cancer services. Thousands of survey responses from patients and clinical staff are the raw material for conversations that drive change in patient care. Breast Cancer Now's Service Pledge 2025/2026 has been jointly sponsored by Eli Lilly and Company Limited, Novartis Pharmaceuticals UK Limited, and Roche Products Ltd. Lilly, Novartis and Roche have not had any control or involvement in this programme.

In phase one of our work together we focused on speeding up the process of getting data from paper surveys into the system. The AI provides reliable transcription across varied form structures, without asking hospitals to change how they collected feedback or forcing patients to abandon paper. The transcription bottleneck that had capped the programme's growth was gone.

What the team didn't yet have was an easy way to turn that data into findings they could share with stakeholders or take to an NHS Trust meeting. They'd still need to export to spreadsheets and compose summaries by hand. This phase extended AIDA, the system we have been building with the team, from getting the data in to helping them decide what to do with it.

The ability to compare

Primary patients, secondary patients, and clinical staff each complete different surveys. One of the things the Service Pledge team had never been able to do easily was compare results across different patient groups. The questions for patients overlap but aren't identical. Answering something like "do secondary patients feel less informed than primary patients?" meant exporting data and doing the work by hand.

We built two ways to compare data. The first works at an audience level: the team can create cohorts within a single survey to see who's had better experiences within the NHS. The second works at a population level: comparing patients in the same hospital but at different stages of treatment, patients in different hospitals, and staff in different hospitals.

We had assumed cross-survey comparison would require significant normalisation work. In practice it didn't: the data that should have been identical already was. We just needed a small mapping layer between the different patient groups, who are asked comparable but not identical questions. The data architecture from phase one had absorbed a major new requirement without a rebuild.

Findings that travel

The full analysis view in AIDA shows every question. It's what you use when you're investigating something in depth. It's not what you open when you have 20 minutes before an NHS Trust meeting.

Key Insights provide that snapshot. The team selects which findings to feature, sets labels, chooses a patient quote to include, and exports the whole thing as a branded PDF or Word document. The layout uses a prominence hierarchy: the most important finding gets the largest treatment, others sit below it in decreasing weight. The document communicates a point of view, not just data.

The quote selection is worth mentioning separately. An earlier version filtered automatically for English-language responses. We replaced it with a dropdown that lets the team browse all quotes and choose. The people who run Service Pledge know which quote will land in a conversation with a particular hospital. The software shouldn't make that call.

Why this is viable now

A project like AIDA wouldn't have been commissioned two years ago. Building bespoke survey processing, a comparison engine, and a configurable reporting tool to Breast Cancer Now's exact specifications would have run to a six-figure budget, and most charities couldn't justify that against off-the-shelf alternatives that almost-but-not-quite fit.

AI coding has changed the economics. The same project, built collaboratively with AI, sits within reach of a charity budget. That doesn't mean cutting corners - the system handles real patient feedback data, achieves high accuracy, and integrates with how NHS trusts actually work. It means the cost of building software that fits an organisation has dropped to the point where fitting the organisation is the right answer, rather than asking the organisation to fit the software.

For the Service Pledge programme, that's the difference between living with a transcription bottleneck and removing it. The category of "things a charity can have built" is larger than it was, and a lot of the workflow problems that have been tolerated for years because the fix was too expensive are now in scope.

What staged development buys you

AIDA is now in its fourth phase. Each phase has been scoped to work on its own, validated before the next was started, and built on the assumption that the team needed to trust the system before they'd depend on it.

That's slower than commissioning everything at once but it's also a safer approach. The scanner problem from phase one, where we'd removed a transcription bottleneck only to reveal a scanning bottleneck upstream, is a good example of why this matters. You find out what's actually in the way before you've spent your budget on assumptions.

For charities thinking about AI, staging development is a worthwhile approach. The value of a system isn't fully visible until it's in use, and the gaps between what was built and what was needed only show up when real people try to do real work with it. Building incrementally is how you find out what you're actually building.