Digital health strategy: from procurement to capability

Most hospital digital strategies are procurement plans wearing a strategy's clothes. The alternative is a capability: a small permanent team that owns software, data and clinical workflow as one system.

The procurement reflex

Over the last decade, European hospitals have spent extraordinary sums on electronic health records, patient portals, AI tools and integration platforms. The clinical and operational returns have been modest, uneven and in some cases negative. The conventional explanation, that change management was poor, describes the symptom rather than the cause.

The cause is that hospitals continue to commission digital work as procurement: write a specification, run a tender, accept a system, train the users, close the project. This model works for medical devices, where the requirement is stable and the object is finished. It does not work for software that touches clinical decision-making, because clinical decision-making evolves faster than any specification can be written. The specification is out of date before the contract is signed, and the institution has purchased a snapshot of its own past.

The costs of this reflex are not abstract. The most careful time-motion study of its era found that for every hour physicians spent with patients, nearly two additional hours went to the electronic record and desk work within the clinic day, with more following them home (Sinsky et al., 2016). That is what a procured system looks like when nobody owns the seam between the software and the work: the technology arrives, the workflow is bent around it, and the bending is paid for in clinical attention, the scarcest resource in the building. Nor is the failure rate a secret: an examination of large digital health programmes across North America, Europe and Australasia found implementation failure persistently high, attributed in the majority of cases to poor project management, with health services heavily reliant on external advisory companies and contractors for the judgement they do not hold in-house (Dendere, Janda and Sullivan, 2021). Reliance on outsiders for the institution's own digital judgement is the procurement reflex in its purest form.

The clearest natural experiment

The strongest evidence that capability, not procurement, is the variable comes from a pair of sepsis deployments we return to often, because the sector so rarely produces a controlled contrast this clean.

One widely purchased proprietary sepsis model, deployed at hundreds of American hospitals through the normal procurement channel, was externally validated and found to have missed 67% of sepsis cases while generating alerts on 18% of all hospitalisations (Wong et al., 2021). The institutions that bought it acquired an artefact. Meanwhile a sepsis system at Duke produced real clinical traction, and the published account of how is instructive: the institution built an internal team that designed a dedicated rapid-response nurse role around the model, calibrated the alert volume to the receiving team's capacity, and negotiated the authority boundaries between nurse and physician before go-live (Sandhu et al., 2020). Same clinical problem, same class of technology, opposite outcomes. The difference was never the algorithm. It was whether the institution possessed the internal capability to design the human-technology system around it.

That is the general lesson, and it will not stay confined to sepsis. Every AI tool now arriving in the sector poses the Duke question: who, inside this institution, owns the intersection of the software, the data and the clinical workflow? A hospital that cannot answer has, whatever its budget, no digital strategy. It has a shopping list.

The capability strategy

A capability strategy reframes the central question. Instead of asking which system to buy, it asks: what is the smallest internal team that can own the intersection of software, data and clinical workflow for our institution, and how do we build and keep it? Four moves follow.

Insource the integration layer. Your data is your strategy; the integration layer is where your data lives. Outsourcing it outsources the strategy itself, and rents back, at consulting rates, the institution's own understanding of its patients. Individual applications can be bought. The layer that connects them must be owned.

Pair clinicians with engineers, permanently. Not in steering committees, which meet monthly and decide nothing; in the same team, on the same roadmap, every day. The Duke account is precisely a description of what such a pairing produces: workflow, role design and thresholds engineered together with the model rather than after it (Sandhu et al., 2020). This is also where oversight by design becomes real rather than rhetorical.

Buy small, build the seams. Best-of-breed systems are fine; monolithic single-vendor estates trade capability for the comfort of one throat to choke. Value is created or destroyed in the seams between systems, and the seams are exactly what a tender cannot specify, because they are discovered in use. The team from the second move builds them.

Measure capability, not project completion. Go-lives measure procurement. The honest metrics are of a different kind: pathways instrumented end to end this quarter, time from a clinician's request to a shipped change, minutes of clinical attention returned per shift. The Sinsky finding gives the baseline for that last number, and it is damning enough to make recovering clinical time the central return on the whole strategy (Sinsky et al., 2016).

What this looks like in eighteen months

Concretely: an internal product team of six to ten people, half clinical and half engineering, owning a roadmap reviewed quarterly with the executive board. Two or three clinical pathways fully instrumented, with the seams between systems built and owned in-house. A small but real internal data platform. A vendor footprint that has been reduced and rationalised, because the institution now knows which twenty per cent of its licences do the work. And a standing answer to the Duke question, so that when the next AI tool arrives, the institution can design the system around it rather than merely receive it.

What this is not

This is not a call to build everything in-house; that error is as expensive as its opposite. It is a call to stop pretending that orchestrating twenty vendors is a strategy. Technology earns its place in a hospital the same way it earns its place in an encounter: by being designed into the work by people who understand both. Buying it is the easy part, and it was never the strategic part.

References

1. Dendere, R., Janda, M. and Sullivan, C. (2021) 'Are we doing it right? We need to evaluate the current approaches for implementation of digital health systems', Australian Health Review, 45(6), pp. 778-781. https://doi.org/10.1071/AH20289
2. Sandhu, S., Lin, A.L., Brajer, N., Sperling, J., Ratliff, W., Bedoya, A.D., Balu, S., O'Brien, C. and Sendak, M.P. (2020) 'Integrating a machine learning system into clinical workflows: qualitative study', Journal of Medical Internet Research, 22(11), e22421. https://doi.org/10.2196/22421
3. Sinsky, C., Colligan, L., Li, L., Prgomet, M., Reynolds, S., Goeders, L., Westbrook, J., Tutty, M. and Blike, G. (2016) 'Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties', Annals of Internal Medicine, 165(11), pp. 753-760. https://doi.org/10.7326/M16-0961
4. Wong, A., Otles, E., Donnelly, J.P., Krumm, A., McCullough, J., DeTroyer-Cooley, O., Pestrue, J., Phillips, M., Konye, J., Penoza, C., Ghous, M. and Singh, K. (2021) 'External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients', JAMA Internal Medicine, 181(8), pp. 1065-1070. https://doi.org/10.1001/jamainternmed.2021.2626