Put Claude to work inside the controls you already run.
AIM designs, builds and operates enterprise systems on Anthropic’s Claude models, deployed in your own cloud tenancy, under your identity provider, your data boundaries and your approval rules.
Claude in Microsoft Foundry · Enterprise retrieval · Governed agents · Managed operations
Six layers between a model and a business outcome.
Select a layer to see who owns it.
Choosing the model is the easy part.
Frontier model access is a procurement decision now. Claude runs in Microsoft Foundry on Azure, on AWS Bedrock, on Google Cloud Vertex AI and through Anthropic’s own API, and you can be calling one this afternoon. Almost nothing about that decision determines whether your AI programme produces anything.
What determines it is the four layers nobody demonstrates: the context the model is allowed to retrieve, the tools it may act through, the approvals it cannot pass on its own, and the operations that keep it accurate six months after the launch announcement. That is the work AIM does.
- Choosing the workload, and declining the ones that will not work
- Enterprise retrieval over your content, with permission fidelity tested
- Tool integration into Microsoft 365, Fabric, Dynamics 365 and line-of-business systems
- Identity and entitlement design in Microsoft Entra
- Information protection and labelling through Microsoft Purview
- Approval gates at the actions that carry consequence
- Evaluation harnesses, held-out sets and regression testing
- An agent register with named business owners and change control
- Adoption, training and the operating model around the tool
- Managed operations, with security operations delivered by Arancia
- Consumption forecasting and cost attribution
- Commercial structuring through CSP and Azure agreements
- Claude model tiers, from Haiku through Sonnet and Opus to Fable
- Long context windows and extended reasoning on the larger tiers
- Deployment through Microsoft Foundry on Azure, AWS Bedrock, Google Cloud Vertex AI, or the Anthropic API
- The Model Context Protocol, a standard for connecting tools and data sources
- Claude Code and the Claude Agent SDK for building agentic systems
- Vision and document understanding alongside text
- Zero data retention options for sensitive workloads
- Enterprise administration and controls in the Claude console
AIM is a Microsoft Solutions Partner and a Cloud Solution Provider. Where Claude is deployed through Microsoft Foundry, it sits inside the same Azure tenancy, identity model and commercial agreement as the rest of your estate, which is usually the shortest path through your own procurement and security review.
One capability. Four ways in.
Deploy
A landing zone in your own tenancy: identity, data boundaries, retention settings, billing and network path agreed with your security team before the first workload.
Explore Deploy Pilots impress and then stallBuild
Enterprise context, tool access, orchestration and an evaluation harness, so the thing that worked in the demo still works on the hundredth real case.
Explore Build Nobody can list the agents runningGovern
An agent inventory, permission design in Entra and Purview, approval gates, evaluation evidence and an audit trail that answers the question an auditor actually asks.
Explore Govern It is live and nobody owns itOperate
One accountable team for prompts, retrieval, tools, evaluations, model versions, cost and the security operations around them.
Explore OperateStart with how the software itself gets built.
Before we talk about what your business could do with Claude, here is what changed in how we build. This is not a forecast. It is the difference between how enterprise software was delivered a few years ago and how AIM’s engineering teams work on client programmes today, across Dynamics 365 implementations, application modernization and our own products.
What does not change
- An engineer owns every line that ships. Review is not optional and not delegated
- Architecture decisions stay with people who will still be there in year three
- Security testing and code scanning run exactly as they did before
- The client approves scope, design and release. None of that moves
- Anything produced that nobody on the team can explain does not reach production
- Accountability for the delivery date stays with AIM, where it belongs
This is not a description of what could be done. AIM is implementing its own Dynamics 365 estate this way now, across Customer Insights, Sales, Finance and Project Operations, together with a custom Customer Licensing Portal that is built software rather than configuration. The work is supported by Dexter, our internal AI-assisted delivery platform, which is part of how we deliver rather than something you licence. We made the change to our own business before recommending it to yours.
Two questions. One recommended engagement.
What is in place today?
What is the binding constraint?
Answer both questions to see a recommended engagement.
We do not just build it. We run it.
Accuracy
Retrieval quality, prompt and model version changes, regression tests against a held-out set, drift reviewed on a cadence rather than after a complaint.
Cost
Token spend attributed to the business process that caused it, model tier routing reviewed against measured quality, not against a vendor’s benchmark.
Security
Identity, access review and monitoring, with SOC and NOC operations delivered by Arancia, our sister company and a dedicated security practice.
Two things we have actually built.
Not a capability statement. One client engagement and one on our own business, both built with Claude, both describable in enough detail that you can judge whether we know what we are doing.
Auditing fleet mileage against the route that was actually driven
Vehicles on the road generate GPS route data. The organization needed to know whether the mileage being claimed against those routes matched the distance the route should have required. Done by hand it is a reconciliation nobody can complete at volume, so in practice it gets sampled, which is another way of saying it mostly does not get checked.
AIM built the tool that does it. Recorded routes are compared against distances calculated through Azure Maps, and the discrepancies are surfaced for review rather than buried in a report. The application itself was developed with Claude, which is the point worth noticing: this is not an assistant bolted onto an existing system, it is working software that exists because building it stopped being prohibitively expensive.
- The problem
- Claimed mileage cannot be verified at volume
- What we built
- A route and distance reconciliation tool
- Built with
- Claude, with Azure Maps for route distance
- Pattern
- The evidence exists, the checking is the expensive part
A full Dynamics 365 implementation, delivered this way
AIM is implementing its own Dynamics 365 estate with Claude in the delivery loop: Customer Insights, Sales, Finance and Project Operations, together with a custom Customer Licensing Portal. The portal is the part that matters most here. It is built software rather than configuration, which is the harder half of any Dynamics programme and the half where delivery risk actually lives.
We chose to do this on ourselves first. A partner recommending a change to how software gets delivered should be able to show the change inside their own business before invoicing anyone for it.
- Scope
- Customer Insights, Sales, Finance, Project Operations
- Custom build
- Customer Licensing Portal
- Delivered by
- AIM’s own engineering and Dynamics teams
- Supported by
- Dexter, AIM’s internal AI-assisted delivery platform
What else stands behind this
Start with the workload, not the licence.
A two-hour session with the people who own the process, the data and the risk. You leave with a shortlist of candidate workloads, the constraints that rule some of them out, and an honest view of which one is worth a pilot.