Case study: Claude inside a national transport body

The figures

What the engagement delivered, in numbers.

~3 hours

Saved per person, per working day, across the trial group.

30+ hours

Of interview material reviewed in hours rather than days.

~3×

Reduction in AI running cost through tuning how the assistant is used.

Day one

Reviewable by Internal Audit from the start, inside the organisation’s existing controls.

How the engagement ran

Where they started

A critical review, and no tool that would clear the bar.

A senior leader at a national transport body faced an organisationally critical governance review: more than 30 hours of interviews with staff and outside comparators, and reporting into the executive team and the Board. The qualitative work was the constraint. They had compared the office-suite assistant and Claude side by side on the real material and found Claude materially stronger for large-volume qualitative analysis and senior-level writing, but an individual subscription had no organisational controls, no retention rules and no audit trail. Consumer AI tools were ruled out on exactly those grounds: anything used had to sit inside the same controls as the rest of the organisation.

Claude, inside their own environment.

Claude, the most capable Anthropic model, running inside the organisation’s own Microsoft Azure tenant, in a dedicated environment with no dependency on production, reached through existing staff logins with multi-factor authentication and conditional access already applied. Private networking between every component, per-user private storage for uploaded material, and every component logging to one place, with alerts for failed sign-ins, unusual volume and cost thresholds. Storage, control plane and audit trail kept in-country.

Usage logging, retention rules and information-owner sign-off were built in from the start, not added afterwards. Scout drafted the acceptable-use addendum and supplied the technical inputs to the organisation’s own data protection assessment, which stayed with its information governance team. Reviewable by Internal Audit on day one, and repeatable for every directorate that comes after. No separate platform, procurement or shadow accounts.

Time back, every working day.

More than 30 hours of interview material processed at speed: transcript review, synthesis and structured analysis that used to take full days landed in hours. Board-ready reporting on a sharper cadence, produced inside the organisation’s assurance boundary rather than outside it.

Across the trial group, around 3 hours a day back per person, roughly half a senior working day. The patterns spread: soon after go-live the IT function asked to extend the platform to its own engineers for coding and infrastructure work. Because the controls were already in place, adding them was a matter of access, not a new project.

Kept current as the models change.

Scout runs the platform on the organisation’s behalf. New model versions are assessed and rolled out deliberately; new features are reviewed against the organisation’s real workloads and information governance before they land; training continues for new joiners and new use cases. A steering group across the directorates using it meets on a fixed cadence, with usage and cost reported monthly.

Configuration is tuned in flight as the team’s needs sharpen, and by tuning how the assistant is used, AI running cost has come down about 3× from the pilot’s starting point. When staff wanted to dictate rather than type, an in-tenant speech-to-text service was added rather than sending audio to a third party, keeping it inside the same boundary as everything else.

What this shows

AI can sit inside a public body’s existing controls, its own environment, staff logins and audit trail, rather than beside them.

The gain was in the material reviewed and the time returned: thirty-plus hours of interview material dealt with in hours, and around three hours a day given back to each person.

A managed service keeps it sharp as the models change, with new versions and features reviewed before they land, running cost tuned down over time, and new teams added without a new project.

“We needed the most capable AI available, inside our own controls, for a review the Board was watching closely. Scout had Claude running in our own environment and reviewable by Internal Audit from day one, and it gave the team back around three hours a day. It has kept pace as the models have changed without us having to think about it.”

Chief Officer

Keep reading

Activate AIThe most capable AI, inside your own environment, run as a managed service.Read

OptimiseKeep what you have running well, and running for less.Read

HousingFrom assessment to a funded programme for a housing association.Read

Talk to us about your organisation.

Tell us where your organisation is starting from.

Talk to us

Or email contact@joinscout.org

The most capable AI available, put to work inside its own assurance framework.

What the engagement delivered, in numbers.

A critical review, and no tool that would clear the bar.

What this shows

Talk to us about your organisation.