Assistants
Claude alongside your people on the work they already do, in the tools they already use. Two to four uses live during set-up.
The question has moved
From whether to use AI, to how to deploy it under a regulator.
For a regulated organisation the question is now how: which controls, which approvals, what evidence for regulator, auditor and board.
Neither a licence nor a pilot is a capability. A capability is Claude doing work you can defend.
Why Claude
Four reasons regulated organisations default to Claude.
We recommend Claude, deploy it inside public bodies, and are building the partnership.
Safety-led by design.
How it is builtAnthropic builds Claude with safety as the design goal, publishes its Responsible Scaling Policy, and has models tested by independent bodies, including government AI safety institutes, before release.
Proven in public service.
PrecedentGovernments, the UK’s among them, have signed agreements with Anthropic and put Claude to work in public-facing services.
It acts, as well as answers.
What it doesClaude reads a case file, checks it against your policy, drafts the response, flags risks and stops for a person to approve anything consequential.
Deployable inside your controls.
Where it runsThrough the major cloud providers or the API, on the logins, access rules and data residency you already have. No new security boundary or supplier.
What activation puts in place
01
Your own environment
Where your systems already run.
Your staff’s existing logins and access rules; data where your regulator requires. Every question and answer recorded and kept as your policy says.
What we set up for you
The set-up is the investment. It’s what makes Claude yours.
A licence gives everyone the same blank assistant; what we build around it is yours.
Your organisation’s context.
ContextYour policies, data definitions, processes, tone and what must never happen, so Claude works like a good new starter.
Guardrails.
LimitsWhat Claude may do alone, what stops for a person, what data it sees and where it can send it, enforced and logged.
Reusable prompts and skills.
AssetsInstructions that make a task come out right every time, tested, versioned and shared.
Agents and evaluations.
AssuranceAgents for specific processes, each with an agreed standard and a test set, re-run on every new model version.
Connections to your systems.
IntegrationThrough the Model Context Protocol, the open standard for governed, auditable access to tools and data, every call logged.
From assistants to agent teams
Start with assistants, grow into agents.
Whatever the shape, the same approval boundary, logging and evidence.
Agents
An agent picks up a case, gathers what it needs, acts within limits you set, and hands anything consequential to a person. Same logins, data boundary, evidence.
Agent teams
Specialised agents split a process under a supervisor, with one audit trail. We start where value is clear and risk understood.
Proof
A national transport body, inside its own assurance framework.
More than 30 hours of interview material and Board-level reporting, Claude inside the organisation’s own environment, now a Scout managed service.
~3 hours
Saved per person, per working day across the trial group.
30+ hours
Of interview material reviewed in hours, not days.
~3×
Reduction in AI running cost.
Day one
Reviewable by Internal Audit, inside existing controls.
See Claude working in your own environment.
Bring a real task your teams do every week. We work it live, inside your controls, and you see what it produces and what it costs.
Book a Priorities Session
No cost, no commitment. Every enquiry is read by Stuart Arthur.
Who it’s for
Same approach, different uses.
Same set-up in every sector; first uses come from the first session.
Housing associations
Regulatory returns from repairs and complaints data; resident contact reviewed for vulnerability or safeguarding; leases and service charges reconciled.
Local authorities
Casework prioritised, a person reviewing every statutory decision; Freedom of Information and subject access responses drafted; committee reports from source documents.
Transport and public bodies
Passenger correspondence answered consistently in every language you serve; board and ministerial reporting assembled; safety matters stop for a person.
Utilities
Regulatory evidence from asset, customer and environmental data; vulnerable customers identified more reliably; submissions reference-checked before sign-off.
Health
Non-clinical paperwork lifted off clinicians: referral letters, discharge summaries, Freedom of Information requests and complaints. Clinical decisions stay with clinicians.
Arm’s-length bodies, police and justice, universities
Sponsor reporting, disclosure review, student services, with the legal basis agreed before anything goes live.
Governance foundation
How every use is governed.
- 01
UK GDPR and the Data Protection Act 2018, or the law that applies where you operate, with a lawful basis and impact assessment for each use
- 02
The UK government’s Scan, Pilot, Scale approach to AI
- 03
The GDS Service Standard
- 04
The NCSC Cyber Assessment Framework
- 05
Cyber Essentials Plus
- 06
WCAG 2.2 AA accessibility, built in
- 07
AI assurance to government guidance, including the Algorithmic Transparency Recording Standard where it applies
- 08
Your sector regulator’s evidence expectations, mapped in
Commercials
Two fixed fees, no day rates.
Set-up
Claude working inside your systems, typically six to twelve weeks depending on scope. Fee agreed before work begins, sized to your systems, data, regulatory duties and first uses.
- A map of your systems and data
- Claude in your environment, on your existing staff logins
- Data protection impact assessments for the first uses
- Connections to the systems the first uses need
- Human-approval points designed and enforced
- Two to four uses live, in daily use
- Testing, record-keeping and assurance in place
- Your team trained to run and extend it
Ongoing
Fixed at contract, reviewed annually, not rising with usage. Claude usage billed at your cloud provider’s rates, no mark-up.
- New uses added on an agreed plan
- Continuous checking of quality and accuracy
- Newer versions of Claude adopted before older ones are withdrawn, re-tested
- Connections updated as your systems change
- Governance evidence kept current
- Ongoing training
- A named senior engineer and an agreed response time
Model versions, plainly
Anthropic releases new versions regularly and retires older ones with notice; changing version is a configuration choice, not a rebuild.
The monthly fee covers the move: re-testing against your standards, refreshed data protection assessments where needed, a controlled changeover you can reverse.
Other models
Claude first, not Claude only.
Our way of governing Claude works for other models too; some organisations have good reasons to run more than one.
Microsoft 365 Copilot
Many organisations already pay for Copilot in Microsoft 365; we govern and deploy it like Claude.
OpenAI and Codex
Some tasks and engineering teams are better served by OpenAI’s models or Codex. Same approval boundary, logging and evidence; we say when Claude is the better choice.
Open-source models
Where nothing can leave your infrastructure, or cost at scale decides it: open-source models inside your environment, same standards.
Whichever models you run: one way of governing them, one set of evidence.
Claude and the Academy
Adoption needs the capability and the people.
Buy either alone; together is how adoption sticks.
How we begin
A working session before any commitment.
Your digital or data lead, head of security and information governance lead, working on a live problem with Claude, leaving with a fixed set-up fee, indicative monthly fee, governance outline and a clear first use. If we’re not the right partner for it, we’ll say so.
More in Activate AI
Book a Priorities Session.
Your real activation challenge, worked live.
Or email contact@joinscout.org