Technology
conclusion was that consumer health tools frequently sit outside the protections users and clinicians assume, since frameworks such as HIPAA cover defined healthcare entities rather than the consumer apps most people actually use. For women’s health data, that gap is not a technicality, and Dr. Abu-Salma closed with valuable perspectives on how that gap could be closed.
Implementation is the real bottleneck
If the first session explained why this field is hard, the second, chaired by Prof. Richard Dobson, Professor of Health and Medical Informatics at KCL, explained why deployment of AI in live clinical workflows is harder still. Dijana Vilic, who heads clinical scientific computing at Guy’s and St Thomas’, walked the room through her “You should have spoken to me first” pathway; illustrating what it actually takes to move an AI tool from a promising model into a live clinical pathway at one of the country’s largest AI mature NHS Trusts. Her account was a corrective to anyone who imagines deployment as a procurement decision. Every tool must pass through a chain of information governance, clinical safety cases, technical integration, monitoring and staff training, and the honest answer to most founders arriving with a finished product is that they should have started the conversation far earlier. Her rule-of-thumb timelines landed hard: bringing a commercial AI product into the Trust typically takes one to three years, with over 90% failing at the internal validation stage, while a well-scoped in-house build can reach the clinic in six to eight months. That asymmetry surprised many in the room, and it carries a practical implication for service leaders weighing build against buy: the commercial route buys maturity and support but pays for it in integration and governance time, while the in-house route moves faster precisely because the governance relationships already exist. And every proposal, however clever, faces the same blunt test from
Professor Asma Khalil, Professor of Maternal Fetal Medicine at St. George’s Hospital, London.
operational leaders: does it save money, or does it save time? If the answer to both is no, it will not be adopted, whatever the model’s accuracy. Her advice to innovators doubles as advice to
Trusts: the conversation with clinical scientific computing, information governance and clinical safety teams should begin early.
Spreading what works: the innovation network route Ruth Bradbury of the Health Innovation Network South London described the complementary problem: getting a tool to work not just in one Trust, but across a whole system. Programmes such as Accelerating FemTech exist precisely to compress the learning curve, pairing women’s health innovators with NHS sites and expertise, so that early-stage companies get structured access to clinical environments, while services get a filtered pipeline rather than a wall of cold pitches. Her broader point was about the machinery of spread and scale: an innovation proven at one site does not travel on its own merits, and the health innovation networks exist to do the unglamorous work of adaptation, evaluation and adoption support that turns a local success into a system change. Alongside her, two midwives, Natasha Bell of
King’s College Hospital and Hebe Davies-Colley of the HIN, showed an example of an AI-driven maternity education tool, with a role-play training application commissioned for maternity staff and piloted at King’s. It was a reminder that some of the most immediately deployable AI in women’s health is not diagnostic at all; it is workforce development.
Dr. Elsa Zekeng, founder of SökerData.
The global south contrast: simple tools, hard environments Dr. Katy Kuhrt, a clinical academic at King’s
48
www.clinicalservicesjournal.com I November 2026
College London, widened the lens to low- and middle-income settings, where the majority of maternal deaths occur. Her work on the CRADLE vital signs device, in settings including Sierra Leone, showed a different design philosophy: robust, low-cost measurement combined with simple risk signals such as the shock index, built for environments where connectivity, staffing and infrastructure cannot be assumed. The contrast was instructive. In the NHS, the constraint is integration and governance; in global settings, it is infrastructure and workforce. In both, the technology may be the easy part.
What AI can do today, and the many ways it goes wrong The third session paired a live demonstration of current capability with a systematic tour of failure modes. Dragutin Oreski, founding engineer at MEGI Health, demonstrated what is now achievable with modest resources: conversational health applications, voice interfaces that hold a natural clinical-style dialogue, and language models running entirely on a local machine with no data leaving the device. That last demonstration drew particular attention from the clinicians in the room, because it addresses the objection they hear most often. If a capable model can run locally, inside a Trust’s own infrastructure, then a whole class of data-sharing concerns changes shape, and the question shifts from whether patient data must leave the building to whether the locally run model is good enough for the task. The point was not any single product but the
direction of travel. Capabilities that required a research lab five years ago are now available to any competent clinical team, which is exactly why the governance question has become
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71