In July - August 2026, reporting by the Financial Times examined links between former senior NHS figures, Palantir, and the procurement of the NHS Federated Data Platform. The reporting did not establish misconduct. It did, however, renew questions about how public bodies demonstrate that major technology decisions are independent, fairly run, and clearly in the public interest.
Public institutions should not need a scandal before they explain how conflicts were declared, decisions were made, and value was tested.
That distinction matters.
Pub
lic debate about national data infrastructure should not depend on proving misconduct before asking whether the system is sufficiently transparent. Nor should concerns about supplier relationships be dismissed simply because a procurement was formally competitive. In programmes involving sensitive data, large public expenditure, and long-term technical dependence, legitimacy depends on more than procedural compliance.
The issue is therefore broader than one company or one contract. It concerns how public institutions select technology partners, how expertise moves between buyers and suppliers, how citizens can understand those relationships, and how organisations demonstrate that a major data or AI programme is delivering value.
Public-sector AI is entering a more demanding phase.
The question is no longer only whether a system works technically. Buyers, citizens, clinicians, regulators, and delivery partners increasingly expect to understand how a system was selected, who can access its data, how decisions are audited, and what tangible benefit it produces.
This is particularly visible in health, where national data platforms and AI-enabled services sit at the intersection of operational resilience, research capability, privacy, clinical confidence, and public trust. The infrastructure may be digital, but the legitimacy required to sustain it is institutional.
For AI providers, this changes the definition of a strong product. Performance, security, and integration remain essential. They are no longer sufficient on their own.
The Procurement Challenge
Large technology procurements inevitably involve close contact between public institutions, suppliers, advisers, technical specialists, and former public-sector leaders. That proximity is not inherently improper. Complex programmes require expertise, and people move between sectors.
The challenge is that perceived influence can become as damaging as actual influence when decision-making is opaque. If stakeholders cannot see how requirements were set, how conflicts were managed, or why one supplier was selected, they may reasonably question whether competition was fair and whether the public received value for money.
A robust procurement process therefore needs more than a compliant tender. It needs a defensible record of how decisions were made.
That includes:
Clear declarations of interests and transparent recusal processes
Separation between market engagement, solution design, and formal evaluation
Evaluation criteria that can be explained after the award
Strong evidence that a programme delivers measurable outcomes
Technical choices that preserve interoperability and customer control
These are not administrative burdens. They are mechanisms for preserving confidence in the system.
Value Must Be Visible
The scale of public spending on data and AI platforms has sharpened a straightforward question: what outcomes are being purchased?
A credible answer should move beyond generic claims of transformation. It should specify the operational baseline, the intended improvement, the measurement method, and the timeframe. In health, that might include reduced administrative burden, improved bed utilisation, faster research recruitment, more reliable forecasting, reduced elective-care delays, or better-quality data available to frontline teams.
This is where many AI programmes struggle. They can demonstrate technical activity ie models deployed, dashboards built, data sources connected—without demonstrating whether the work changed an important operational outcome.
Each commitment should name the baseline, the target, the measurement owner, the review point, and the consequence if the promised outcome does not materialize. A better model is to define a small number of measurable commitments at the outset:
When outcomes are explicit, technology procurement becomes easier to defend—and easier to improve.
Interoperability Is Strategic
Public institutions should be wary of confusing a coherent platform with a closed ecosystem. A platform can provide useful common services while still allowing specialist tools, regional innovation, and local workflows to connect through well-governed interfaces.
For suppliers, this creates a more constructive competitive position. The goal is not necessarily to replace a national platform. It may be to provide a differentiated capability that works with existing data environments while preserving customer choice.
This is especially relevant for emerging AI systems, edge intelligence, privacy-preserving analytics, and domain-specific decision support. Their value often comes from operating close to the workflow, where latency, context, resilience, and human oversight matter.
The strongest proposition is therefore not “replace everything.” It is: “deliver a measurable capability that interoperates cleanly, leaves the customer in control, and can be evaluated transparently.”
Governance by Design
[ Federated Data Platform (FDP): The specific solution design where individual organizations (such as separate NHS trusts or Integrated Care Boards) maintain sovereign, localized control of their own data instances (”tenants”) rather than dumping everything into a single massive central data pool.
Palantir Foundry: A proprietary enterprise data platform and ontology-powered operating system built by Palantir Technologies to integrate siloed data, manage pipelines, and run AI-driven workflows.]
Trust cannot be added at the end of a deployment. It has to be designed into the product and commercial model.
For founders, this means treating governance as a core product capability:
Traceability: Users should be able to understand the origin of data, the logic of automated outputs, and the history of material decisions.
Human accountability: AI should support accountable professionals rather than obscure responsibility.
Data minimization: Systems should use only the data required for a defined outcome and apply appropriate controls throughout its lifecycle.
Independent evaluation: Claims about performance and value should be open to testing by customers and, where appropriate, third parties.
Interoperability: Customers should retain practical freedom to integrate, extend, and change components over time.
Commercial clarity: The relationship between pilot scope, success criteria, pricing, and scale-up should be understandable before deployment begins.
These principles are not only good governance. They are a route to adoption in markets where reputational risk and institutional confidence are central to the buying decision.
A New Standard for AI
The next generation of public-sector AI will not be judged only by model accuracy or technical sophistication. It will be judged by whether it solves a real problem, produces evidence of value, preserves accountability, and leaves public institutions in control.
In critical public services, trust is not a communications strategy. It is infrastructure.