[Chat GPT August 2026]
##
British pathologist David Clark is an expert on digital pathology & real-world integrations. Below, Chat GPT writes a detailed commentary on his four recent articles on NHS and its digital pathology investment tactics.
David Clark - Nottingham - hematopathologist - Nottingham NHS Trust
https://www.linkedin.com/in/david-clark-61a79a23b/
He's written a series on faulty plans for NHS investments in digital pathology.
Entry point at Linked-In here:
https://www.linkedin.com/feed/update/urn:li:activity:7491025358016888833/
It matches to a four-part series he wrote at Pathology News;
https://www.pathologynews.com/the-masterbuilders-path-to-recovery/
##
Reviewing David Clark's Essays on the NHS Digital Pathology Investment
The NHS digital pathology experience is not a story of failed scanners. It is a story of incomplete system design. David Clark’s four-part critique shows how capital-first procurement, weak interoperability, hybrid workflows, training gaps, and unfunded recurring costs can convert promising technology into a megaproject problem. The remedy is architectural.
This essay was written by Chat GPT and should be read as an example of the current state of AI reading, organizing, and writing - not as ground truth.
Essay as 16pp white paper PDF - here.
The NHS Did Not Buy the Wrong Technology. It May Have Bought Technology in the Wrong Order.
Clark’s Part 1: The Strategy Gap
Part 2: The LIMS Trap — Where the Abstract Strategy Failure Became Concrete
The Hybrid Trap: When Digital Pathology Adds Work Instead of Removing It
Part 3: Technology Adoption Is a Workforce Phenomenon
The Sustainability Cliff: Capital Is Not a Business Model
Part 4; Clark's Criteria for Successes
“Design the Join”
A Better Investment Model: Seven Requirements for the Next Wave
An Important Irony: The NHS Is Still Investing
The Larger Lesson for AI in Pathology
Conclusion: From Buying Digital Pathology to Building Digital Pathology
##
EXECUTIVE SUMMARY
David Clark’s four Pathology News essays make an argument about the NHS digital pathology programme: its difficulties arose less from immature scanners or weak clinical rationale than from treating digital pathology as equipment procurement rather than redesign of a complex clinical operating system. Capital arrived before an end-state, interoperability architecture, workforce model, governance structure, or sustainable revenue model had been fully defined. The consequences included LIMS integration failures, middleware surprises, duplicated glass-and-digital workflows, inconsistent adoption, cross-Trust image-sharing barriers, and recurring storage and support costs that outlived grants.
The critique is strengthened by Bent Flyvbjerg’s work on major projects. His “Iron Law” describes projects that run over budget and time while underdelivering benefits; newer research finds IT projects unusually exposed to extreme tail risk. Digital pathology combines IT hazards—intangibility, goal ambiguity, stakeholder resistance, bespoke integration—with safety-critical clinical workflow.
The lesson is not to retreat from digital pathology. NHS experience includes successes, including highly digitised networks and routine AI use. The policy implication is to fund the next phase differently: define outcomes first, design network-level interfaces, make integration readiness a funding gateway, build recurring costs into business cases, standardise reusable components, align training and assessment, and learn from prior deployments.
###
Digital Pathology Is Not a Scanner Project
What the NHS Experience Teaches About Large-Scale Digital Transformation
1. The NHS Did Not Buy the Wrong Technology. It May Have Bought Technology in the Wrong Order.
Digital pathology has a compelling technological proposition. Glass slides can be converted into high-resolution whole-slide images; pathologists can report remotely; difficult cases can be routed to specialists without physically transporting slides; multiple clinicians can inspect exactly the same image; and, critically, the image becomes machine-readable. That final step creates the substrate for computational pathology and artificial intelligence.
The NHS therefore had good reasons to invest. What David Clark's remarkable four-part Pathology News series asks is a more uncomfortable question: What if the fundamental investment error was not choosing digital pathology, but treating digital pathology as a technology acquisition rather than as the redesign of a clinical production system?
Clark estimates that more than £100 million has flowed into NHS digital pathology systems and AI over roughly the past decade. His central contention is that investment in scanners and infrastructure preceded an adequately specified national operating model connecting funding, workflow, LIMS, interoperability, workforce, training, governance and long-term operating costs. In his formulation, the NHS funded pieces of the future before fully designing the future into which those pieces were supposed to fit. [1]
That distinction matters. A scanner can be successfully installed and still represent an unsuccessful transformation.
Clark's £100 million figure should be understood as an estimate rather than a figure that can readily be reconciled to one national digital-pathology appropriation. Government records do, however, confirm substantial overlapping investments. In 2018, UKRI and Innovate UK committed £50 million to five centres of excellence in digital pathology and medical imaging with AI. [5] A further £248 million diagnostic-digitalisation programme was announced in 2021; that pot was broader than pathology alone, but explicitly included technology allowing laboratories and hospitals to exchange tests, images and results. [6] NHS planning documents subsequently continued to direct systems to complete investments in digital pathology, LIMS and other digital diagnostics. [7] (GOV.UK)
Thus, the precise cumulative total attributable solely to digital histopathology may be debatable, but Clark's larger premise is not: this has been a substantial, multi-year public investment programme.
The more important question is what it bought.
2. Clark's Part 1: The Strategy Gap
The first Clark article identifies what might be called the capital-before-architecture problem. The NHS had a technology that clearly could transform pathology, but Clark argues that national investment was not preceded by a sufficiently rigorous determination of the desired end state.
Was the objective 50% digital reporting? Eighty percent? One hundred percent? Was the relevant unit of transformation an individual laboratory, an NHS Trust, a pathology network, or ultimately the NHS as a national system? Were expected benefits primarily reduced turnaround times, fewer courier movements, increased pathologist productivity, remote working, workforce resilience, regional subspecialisation, AI enablement—or some defined combination of these?
Without agreed answers, success becomes difficult to measure and system requirements difficult to specify. Clark argues that the result was a drift toward the easiest thing to specify and purchase: equipment. [1]
This is not merely retrospective criticism. The Royal College of Pathologists had understood quite early that digital pathology was more than image capture. Its 2019 strategy described digital pathology as a potentially transformational technology and explicitly called for leadership in implementation, education, training, assessment and professional practice. [8] Today, the College estimates that modern digital-pathology IT infrastructure for a large hospital can cost approximately £2–4 million, underscoring that scanners themselves are only part of the required environment. (Royal College of Pathologists)
Clark's argument is therefore subtler than saying "the NHS should have planned more." His concern is that the system boundary was wrong. A Trust can buy a scanner. But a Trust cannot by itself establish national interoperability standards, alter professional examinations, solve cross-Trust information governance, rationalise all regional LIMS architectures or establish a sustainable national funding mechanism for rapidly accumulating image archives.
Locally rational procurement can therefore produce collectively irrational architecture.
3. Part 2: The LIMS Trap — Where the Abstract Strategy Failure Became Concrete
Clark's strongest example is surprisingly mundane: the laboratory information management system.
A digital image is clinically useful only when it remains securely associated with the correct patient, specimen, block, slide, request, diagnosis and reporting workflow. In practical pathology operations, much of that linkage runs through the LIMS.
Yet many NHS laboratories operate mature, heavily customised LIMS platforms, sometimes decades old. Direct integration with newer digital pathology platforms proved difficult at multiple sites. Clark describes deployments discovering, after procurement, that middleware or specimen-tracking systems would be required to bridge the new digital pathology platform to the existing LIMS. Those additional layers had not always been identified, budgeted or sequenced in advance. [2]
This is where Clark's case becomes particularly instructive. An individual integration failure is an IT problem. The same integration failure appearing independently in multiple Trusts is a programme-design problem.
According to Clark, there was no sufficiently strong mechanism for an early Trust's LIMS experience to become a mandatory warning to later Trusts. Once integration emerged as a critical-path dependency, a mature national programme might have stopped and asked: Which other sites have the same architecture? What middleware is required? What will it cost? Should LIMS modernisation precede scanner deployment? Should funding be withheld until the integration pathway is demonstrated?
Instead, similar blockers were rediscovered locally. [2]
This failure of organisational learning may be more consequential than any single technical error. The NHS was, in effect, generating its own reference-class database but not systematically using it.
4. The Hybrid Trap: When Digital Pathology Adds Work Instead of Removing It
A particularly useful lesson for healthcare technology investment is that partial digitisation can be economically worse than either endpoint.
A traditional glass workflow has inefficiencies. A well-designed digital workflow promises to remove some of them. But a laboratory operating both workflows indefinitely may acquire the costs of both.
Clark describes laboratories scanning some cases while continuing microscopy for others, relying on manual processes to bridge systems, maintaining two modes of reporting and asking employees to remember which process applies to which case. Without an explicit end state, hybrid operation can cease being a transition phase and become the operating model. [2]
The available national data illustrate the unevenness of adoption. Material submitted during NICE's 2026 scoping work on AI-assisted prostate histopathology reported that 25 of England's 27 pathology networks had begun digital reporting and 106 of 132 acute and specialist Trusts were using digital images for primary diagnosis. But only 51 Trusts were reported as digitally reporting more than half of their cases. [9] (Nice)
Those figures are simultaneously evidence of impressive diffusion and evidence of the hybrid problem. Digital pathology has spread widely. Full transformation has spread much less widely.
This matters for return on investment. If fixed costs—scanners, image-management platforms, integration, validation and storage—are incurred while glass handling and microscopy infrastructure remain substantially intact, the expected efficiency dividend may never arrive.
5. Part 3: Technology Adoption Is a Workforce Phenomenon
Clark's third article moves from systems engineering to organisational behaviour.
A new diagnostic workflow is not adopted merely because management announces it. Pathologists observe one another. They hear whether a neighbouring institution's implementation is smooth or painful. They learn whether respected colleagues actually use the new workflow for routine reporting. In Clark's application of Everett Rogers' diffusion-of-innovation model, visible successful adoption creates social proof.
The reverse can also occur.
When clinicians hear repeatedly about difficult LIMS integrations, cumbersome hybrid workflows, repeated rescanning, workstation inconvenience or projects that never quite become routine, informal professional networks transmit a negative signal. Clark argues that the NHS programme may therefore have produced, in some places, social proof in reverse: early difficulties made later adoption harder. [3]
This interpretation avoids the easy explanation that reluctant pathologists simply resisted change. An individual who declines to abandon a reliable microscope for a slower, partially integrated digital workflow may not be technologically conservative at all. The person may be responding rationally to the system actually offered.
Training adds another structural mismatch. The Royal College of Pathologists' 2019 strategy explicitly anticipated digital pathology in training and examination. Its annual report that year stated an intention to put mechanisms in place for College examinations using digital pathology. [8,10] Yet Clark notes that the professional qualification system continued to be strongly anchored in glass microscopy years into the national deployment. (Royal College of Pathologists)
That produces an extraordinary policy signal. The healthcare system is investing to make one technology the future of diagnostic work while the professional credentialing system continues to tell trainees, through what it actually assesses, that competence is demonstrated in the old medium.
For major transformation, curriculum and examinations are infrastructure too.
6. The Sustainability Cliff: Capital Is Not a Business Model
Perhaps Clark's most important observation for policymakers concerns the distinction between capital funding and recurring operating expense.
A scanner can be purchased with a grant. A digital pathology service cannot.
After installation come maintenance contracts, image-management licences, network costs, middleware support, cybersecurity, hardware refreshes, validation work, additional scanner capacity and—above all—ever-expanding image storage. Whole-slide image files are large, and clinical archives do not politely disappear when the initial capital programme ends.
Clark argues that these recurring commitments frequently landed in pathology budgets constructed around the economics of glass-slide practice. [3]
This produces what he calls the sustainability cliff. A department may initially appear to have received a generous central investment. Several years later, the central grant is gone while the Trust must decide whether to fund storage, software licences, service contracts and hardware replacement from routine operating budgets.
The paradox is especially severe for an incomplete deployment. The best argument for continuing revenue funding would be documented productivity and clinical benefits. But an incomplete hybrid system may not yet have produced those benefits. The organisation therefore reaches the point at which continued funding must be justified precisely when its prior implementation has made the business case hardest to demonstrate. [3]
Clark cites a 2023 Royal College-led Downing Street meeting that estimated roughly £200–300 million over five years would be required to achieve full digitalisation. [3]
Whether that precise estimate proves correct is less important than its structure. Digital transformation requires a total-cost-of-ownership model, not merely an acquisition budget.
Sidebar: Bent Flyvbjerg — Why Digital Pathology Belongs to a Much Larger IT Risk Problem
Bent Flyvbjerg and Dan Gardner's 2023 book How Big Things Get Done distils decades of research on major projects into several memorable rules: understand the odds; "think slow, act fast"; begin with the goal and work backward; use repeatable modules rather than unnecessarily bespoke construction; and build a team capable of collective delivery. [11]
The connection to pathology became even sharper with Flyvbjerg and colleagues' recent analysis of 11,011 projects spanning 23 project categories, including 5,360 IT projects. The study found IT cost risk to have a markedly fatter tail than the other project types. IT was the sole category falling into the authors' most extreme statistical risk class. In ordinary language, the problem is not simply that the average IT project overruns: a minority of IT projects can go catastrophically wrong, making conventional averages dangerously reassuring. [12]
The researchers discuss four familiar contributors to IT risk: immaturity, intangibility, goal ambiguity and stakeholder resistance. They add two especially relevant factors: bespokeness and fast decision-making. Highly customised systems repeatedly create novel interfaces and novel failure modes. Modularity has the opposite effect: the same component or connector is used repeatedly, allowing learning to accumulate. [12]
Remarkably, the mean IT project in their dataset lasted only 3.2 years, versus 6.9 years for the other project categories. IT therefore did not become uniquely risky merely because IT projects lasted longer. They did not. [12]
This frames NHS digital pathology as a special case of a general phenomenon. Digital pathology contains almost every characteristic likely to magnify IT risk: ambiguous end points, intangible benefits, numerous stakeholders, institution-specific legacy systems, bespoke interfaces and organisational change. It then adds clinical regulation and patient-safety requirements.
The appropriate conclusion is not that large healthcare IT projects should be avoided. It is that they should be managed as fat-tail risks, where preventing a small number of disastrous pathways matters more than polishing the average forecast.
7. Clark's Part 4: Forward. Clark's Criteria for Successes
A simplistic account of digital pathology failure would be contradicted by the NHS itself.
Clark identifies major successes. The National Pathology Imaging Co-operative in West Yorkshire has achieved 100% digital scanning across its network. Nottingham University Hospitals has transitioned to 100% digital pathology reporting and has integrated AI into routine prostate-biopsy reporting. [4]
Other NHS programmes continue to demonstrate impressive adoption. NHS England's account of a North Central London implementation reports digital images supporting diagnosis and reporting in nearly 70% of histopathology cases at UCLH within weeks of go-live. (NHS England)
Those examples strengthen, rather than weaken, Clark's thesis. They demonstrate that the underlying technology works.
But they also expose the distinction between local digital success and network transformation. Clark describes Nottingham's digital system as still being constrained by Trust-level firewalls that prevent effortless importing of outside cases for specialist review. In some networks, a slide may be digitised at Site A, transported physically to Site B, and scanned again because the organisational and technical connection for transferring the already-existing image was never created.
The scanner has succeeded. The system has failed to make use of the fact that the slide was scanned.
That is a near-perfect illustration of the problem.
8. "Design the Join"
The most memorable proposal in Clark's final article is to design the join.
This is a direct application of Flyvbjerg's emphasis on modularity. Large systems become more manageable when components can be combined using stable, predictable interfaces. The brilliance of Lego is not the sophistication of any particular brick. It is that every brick knows how to connect to the next one.
The first NHS digital pathology wave, Clark argues, contained many capable components but too many bespoke joins. Different scanners, LIMS platforms, image-management systems, Trust security environments, archives and clinical applications were expected to communicate through locally engineered solutions.
For the next phase, Clark proposes national or common specifications for the interfaces themselves: digital-pathology interoperability requirements established before procurement, standard approaches to LIMS bridging, common whole-slide-image archive standards and reusable frameworks for assessing AI applications. [4]
This idea deserves to be placed at the centre of digital pathology policy.
Healthcare systems have traditionally specified the machine: scanner resolution, throughput, monitor characteristics, storage capacity, algorithm performance. In a networked digital health environment, it may be equally important to specify what every machine must be able to connect to, what information it must exchange, and under what governance rules that exchange occurs.
The product specification becomes less important than the ecosystem specification.
9. A Better Investment Model: Seven Requirements for the Next Wave
Clark's four essays, combined with the broader megaproject literature, suggest a substantially different investment model.
First, define the clinical destination before purchasing the technology. Funding requests should identify the intended operating state—not simply the quantity of hardware to be installed. The business case should state what fraction of applicable cases will be digital, which workflows will disappear, which network-level activities will become possible, and how those outcomes will be measured.
Second, make integration a gateway condition. Before scanner capital is released, the LIMS pathway, patient identity architecture, image-management interface, archive strategy, EHR/reporting connection and cross-site sharing pathway should be demonstrated or explicitly costed. Clark proposes precisely this type of phased release against readiness milestones. [4]
Third, fund the whole life cycle. Every business case should include five- and ten-year estimates for licences, maintenance, storage, network infrastructure, scanner replacement, cybersecurity, middleware and incremental support staff. Capital approval without credible recurring funding is not approval of a digital service; it is approval of its first few years.
Fourth, measure benefits at the level where the benefit occurs. If the justification for digital pathology includes regional subspecialty reporting, load balancing and expert consultation, performance cannot be evaluated solely by the percentage of cases scanned within one Trust. The measurement boundary must match the benefit boundary.
Fifth, standardise what can be standardised. Digital pathology should become progressively less bespoke. Interfaces, metadata, archive rules, AI validation procedures and LIMS connectors should increasingly resemble reusable infrastructure.
Sixth, align professional training with the desired future state. The curriculum, examinations and competency frameworks should teach and assess digital reporting, understanding of image systems and appropriate use and evaluation of computational pathology. Otherwise workforce policy works against capital policy.
Seventh, create a national learning system for implementation failure. A failed interface at one site should become information for every site. An unexpected storage expense should alter future cost models. A successful conversion strategy should become a replicable playbook. Clark notes that the NHS now possesses years of internal experience that could form precisely the reference class Flyvbjerg recommends. [4]
The national programme should not merely fund projects. It should learn across projects.
10. An Important Irony: The NHS Is Still Investing
This issue is not historical.
Current NHS England policy continues to call for digital diagnostics and pathology-network development. Its 2024/25 planning guidance specifically instructed systems to complete planned investments in digital pathology and LIMS, with the objective of improving network productivity. Current capital guidance for 2026/27–2029/30 likewise refers to regional digital roadmaps encompassing networked pathology systems and optimisation of histopathology workflows. (NHS England)
Meanwhile, the Royal College of Pathologists' recent submission to the NHS Productivity Commission makes a strikingly Clark-like diagnosis: digital pathology adoption remains fragmented across England's pathology networks; legacy systems and limited resources impede integration; and a national strategy should address infrastructure, interoperability with LIMS and EHRs, workforce development, governance and sustainable adoption. [13] (Royal College of Pathologists)
In other words, the policy window remains open.
The next tranche of money can either extend the architecture already created or be used to redesign the way the architecture itself is financed and governed.
11. The Larger Lesson for AI in Pathology
There is an especially urgent implication for artificial intelligence.
AI does not escape weak digital infrastructure. It inherits it.
If images cannot cross a Trust boundary, an AI service built on those images will confront the same boundary. If local governance differs across institutions, every AI deployment must repeatedly solve the governance problem. If scanners, LIMS and archives connect through bespoke interfaces, each new algorithm becomes one more item requiring bespoke integration and validation.
Clark puts the point forcefully: AI becomes a stress test of the architecture already built. Fragmented governance produces fragmented AI; fragmented image access produces fragmented AI. [4]
This reverses a common way of thinking about digital pathology investment. Whole-slide imaging is sometimes presented as the preliminary expense required before the exciting AI applications arrive. In reality, the value of AI may depend more on the quality of the unglamorous infrastructure underneath it than on the sophistication of the algorithm itself.
A £1 million AI programme placed on top of weak interoperability may create another custom IT project. The same AI placed on top of a standardised national image architecture becomes closer to a plug-in.
That difference is precisely Flyvbjerg's distinction between bespoke construction and modularity.
Conclusion: From Buying Digital Pathology to Building Digital Pathology
Clark's four articles should not be read as an indictment of digital pathology. They are more useful than that.
The NHS has demonstrated that digital pathology works. It has also produced something equally valuable: a large natural experiment in how difficult digital pathology is to implement across a fragmented, highly regulated healthcare system containing legacy IT, autonomous organisations, constrained capital budgets and an overstretched workforce.
The first wave bought scanners and built islands of digital capability. Some islands are excellent. But the greater promise of digital pathology lies in what happens between them: image exchange, regional workload balancing, remote specialist review, common archives, scalable AI, shared validation and a workforce trained from the outset to work digitally.
That requires a different definition of infrastructure.
Infrastructure is not only scanners, servers and storage. It is also interoperability standards, workflow design, governance, professional education, financial sustainability and the feedback machinery through which one organisation's expensive mistake becomes another organisation's inexpensive lesson.
Flyvbjerg's work supplies a useful general rule. Large IT projects are not merely ordinary projects with computers attached. Their risk distribution appears to be unusually unforgiving. The antidote is not more optimism or more detailed prediction of inherently uncertain projects. It is to reduce uniqueness: plan before committing, use the outside view, build modularly, standardise the joins and learn rapidly from repeated components.
For NHS digital pathology, perhaps the most important sentence in Clark's series is therefore not that the programme obeyed the Iron Law.
It is that the technology works. [4]
The remaining challenge is to design a healthcare system capable of receiving it.
References
1. Clark D. How NHS digital pathology obeyed the Iron Law of megaprojects – Part 1: The Strategy Gap and the Iron Law. Pathology News. July 29, 2026.
2. Clark D. How NHS digital pathology obeyed the Iron Law of megaprojects – Part 2: The Critical Path and the LIMS Trap. Pathology News. July 30, 2026.
3. Clark D. Part 3: Engagement, Training, and the Sustainability Cliff. Pathology News. August 4, 2026.
4. Clark D. Part 4: The Masterbuilder's Path to Recovery. Pathology News. August 5, 2026.
5. Innovate UK. New centres to speed up patient diagnosis: apply for funding. June 13, 2018. UK Government. Up to £50 million announced for centres of excellence in digital pathology and medical imaging. (GOV.UK)
6. Department of Health and Social Care. £250 million in NHS technology to modernise diagnostics. November 8, 2021. £248 million programme for digitisation of NHS diagnostics. (GOV.UK)
7. NHS England. 2024/25 priorities and operational planning guidance. Includes completion of planned digital pathology, LIMS and MRI-acceleration investments. (NHS England)
8. Royal College of Pathologists. Digital Pathology Strategy 2019. London: RCPath. (Royal College of Pathologists)
9. National Institute for Health and Care Excellence. Artificial intelligence technologies to assist histopathology for prostate cancer diagnosis [ID6684]: consultation comments on draft scope. April 2026. Includes reported national digital-pathology adoption figures. (Nice)
10. Royal College of Pathologists. Annual Report 2018–2019. Includes commitment to mechanisms enabling College examinations using digital pathology. (Royal College of Pathologists)
11. Flyvbjerg B, Gardner D. How Big Things Get Done: The Surprising Factors That Determine the Fate of Every Project, from Home Renovations to Space Exploration and Everything in Between. New York: Currency/Penguin Random House; 2023.
12. Flyvbjerg B, Budzier A, Aaen J, Keil M, Zottoli M. The uniqueness of IT cost risk: A cross-group comparison of 23 project types. Project Management Journal. 2026;57(1):14–43. DOI: 10.1177/87569728251340590.
13. Royal College of Pathologists. Response to the NHS Productivity Commission consultation. 2026. Proposal for a national digital pathology strategy, central funding, interoperability and workforce support. (Royal College of Pathologists)
14. NHS England. Capital guidance 2026/27 to 2029/30. Includes funding through diagnostic networks following regional digital roadmaps, including networked pathology systems and optimisation of histopathology workflows. (NHS England)
15. Department for Business, Energy & Industrial Strategy; UK Research and Innovation; Innovate UK. Artificial Intelligence to help save lives at five new technology centres. November 6, 2018. Government announcement of five AI centres for digital pathology and medical imaging. (GOV.UK)
