Four principles behind every engagement

Technology alone does not create lasting public value. Whether an AI system delivers and endures depends on how it is designed, implemented and measured within the realities of an institution. Four principles guide our work, each addressing a common challenge in taking AI from a pilot to sustained institutional use.

We work inside the institution

We work alongside the teams responsible for implementation, aligning with delivery objectives and staying engaged through deployment and beyond. This helps ensure that the system reflects existing workflows, operational requirements and the realities of the institution in which it will be used.

Effective implementation depends on this context. Solutions designed without an understanding of how work is actually carried out can struggle with adoption. Working within the institution allows those constraints to inform design from the outset.

Working alongside institutional implementation teams

We hand the system over

Capability transfer is built into every engagement. We leave trained teams, documentation and operating routines that enable the institution to manage the system after implementation.

From the outset, we consider who will operate the system, how it will be supported and maintained, and what capabilities are required to manage it effectively. The objective is durable institutional ownership, not continued dependence on an external provider.

Capability transfer and workforce training

We decide from evidence

Evidence informs decisions throughout the engagement, from selecting and prioritising use cases to assessing readiness for scale. We combine field data, user experience and measurement to understand how a system performs in practice and what it changes for the people and services it is intended to support.

This includes looking beyond aggregate performance to identify gaps in coverage, adoption and outcomes. Decisions to improve, continue or scale are based on evidence from implementation rather than demonstration alone.

Evidence-driven decision making and measurement

We evaluate independently

We only evaluate systems that no one in our group built, and we do not bid to build systems that we have evaluated. Evaluation is kept structurally separate from delivery.

Evaluation is most useful when its findings are not shaped by an interest in a particular outcome. Keeping assurance separate from delivery allows institutions to assess system performance, safety, fairness and impact with greater independence.

This approach also enables us to evaluate systems developed by other providers without creating a conflict between evaluation and implementation. The principle applies equally to our own work: systems we build are not evaluated by us.

Independent algorithmic evaluation and auditing
Explore Capabilities

Explore our capabilities

Discover our applied AI capabilities across government, nonprofits, and global philanthropy.

Explore our capabilities →