The UK motor insurer reports processing around 600,000 quotes a day and estimates that an internal writing assistant saves 6,500 working hours annually. Those figures describe different workloads and different kinds of evidence. Neither establishes that an AI agent can already resolve an insurance claim accurately, independently, or more cheaply.
For organisations considering a similar move, 1st Central offers a useful case study in connecting cloud investment to specific business tasks. It also illustrates why IT teams need to distinguish what is running today from what is still being designed.
1st Central’s Azure migration puts operational systems behind the AI story
The most substantial deployment described in Microsoft UK Stories is the migration of both 1st Central’s data platform and its core insurance systems to Azure. The insurer uses the cloud to run systems handling policies, claims, and customer interactions. This is an operational deployment, extending beyond cloud storage or an isolated AI experiment.
That scope matters to interpreting the company’s strategy. Its stated objective is to analyse detailed information quickly enough to influence a customer’s journey while it is happening. A data platform used only for retrospective reporting would serve a different purpose from models involved in producing an insurance quote before the customer leaves the transaction.
According to Microsoft’s account, 1st Central processes approximately 600,000 insurance quotes each day, analyses about 5,500 data attributes per quote, and responds in roughly two-and-a-half seconds. Some of that processing identifies possible inaccuracies in customer information before a policy is purchased. The figures describe a time-sensitive operational workload, although they do not identify the individual Azure services supporting it.
These are company-reported operating figures, not an independently reproduced Azure benchmark. Microsoft’s story does not establish whether the response time is an average, a percentile, or a target, nor does it provide peak-load measurements. An IT team can use the account to understand the workload’s broad requirements, but it cannot use those numbers to size its own deployment or compare Azure with another platform.
The investment predates the current interest in generative AI. In a July 2025 company announcement, chief executive Michael Lee said 1st Central had invested £100 million over the preceding decade in proprietary technology and data platforms. That is the insurer’s own account of its spending, rather than an independent assessment of the return, but it supports the narrower conclusion that this is a long-running engineering programme.
Microsoft’s September account traces that programme to the company’s founding in 2008, when expanding price-comparison websites created an opportunity to analyse customer information more effectively. CIO John Davison describes the systems then available as difficult to integrate and poorly suited to large datasets. That history explains 1st Central’s preference for technology it controls; it does not establish that every insurer should replace packaged software with an internally developed platform.
Microsoft 365 Copilot addresses a different workload from insurance pricing
The employee-facing part of the deployment is more familiar. Microsoft UK Stories says 1st Central now uses Microsoft 365 Copilot and Microsoft 365 E7, describing the latter as bringing together AI productivity capabilities with identity, security, and agent-management capabilities. The account does not give a licence count or enough commercial detail to calculate the insurer’s cost.
Employees use Copilot to find information, summarise documents, connect material held in different parts of the organisation, accelerate research, prepare presentations, and improve communications. The emphasis is on helping someone begin work with a stronger first draft. These are assisted knowledge-work tasks, distinct from the real-time modelling used to produce insurance quotes.
That separation prevents an easy misreading of the case study. The reported quote volume and response time do not measure Microsoft 365 Copilot’s performance. Equally, the existence of an Azure-hosted insurance platform does not tell readers which models or development tools power the insurer’s internal assistants.
The clearest productivity claim concerns an internally developed assistant that helps employees write in 1st Central’s preferred tone of voice. The insurer estimates that it saves around 6,500 hours annually. Microsoft’s account does not identify the assistant’s implementation or explain the calculation, so the saving should remain attached to that particular tool and to the company’s estimate.
For an IT purchasing decision, hours saved and financial return answer different questions. The estimate does not establish a reduction in payroll, a corresponding increase in completed work, or a net saving after licensing and operation. It is a useful indication of where the insurer sees value, rather than a benchmark another organisation can apply to its own employees.
The practical inference is to make a productivity trial equally specific. A tone-of-voice assistant has a recognisable task and an output that employees can review. Comparing time spent producing and correcting that output would be more informative than treating every interaction with Copilot as a unit of productivity. That is a measurement approach suggested by the case, not a method 1st Central says it used.
Microsoft 365 agent governance makes the oversight claim actionable
1st Central says it has developed an oversight framework to monitor agents while allowing innovation to continue. Microsoft’s customer account does not describe the framework’s controls or approval process. Calling it an oversight framework therefore tells readers about the insurer’s stated approach, but not how to reproduce it.
Microsoft’s separate administrator documentation supplies a more concrete starting point for other organisations. Its Agents admin guide for Microsoft 365 describes tenant policies for agent access, sharing, and publishing, together with controls to approve, deploy, remove, and block agents. These are available management concepts, not evidence that 1st Central has configured any particular control.
The distinction between agent types also affects administration. Microsoft describes declarative agents as customisations that use Copilot’s models and orchestration and rely on user-initiated interactions. Custom engine agents offer greater flexibility, including automatic triggers without direct user input. Consequently, the word “agent” alone does not establish whether software simply answers an employee’s question or initiates an action.
For Microsoft 365 administrators, the documented controls suggest three practical areas of work.
Microsoft 365 agent access needs an owner and a deployment boundary
Microsoft identifies AI Admin and Global Admin as roles able to configure and manage agents in the relevant Microsoft 365 controls, while Global Reader provides view-only access. Its guidance recommends using the least-privileged role necessary rather than assigning Global Admin simply because it can perform the task.
An administrator can inspect their assigned role in the Microsoft 365 admin center by selecting Users > Active users, opening their user entry, and checking Roles. That is a prerequisite check for this organisational administration workflow, not a personal Microsoft-account setting.
Microsoft also notes that agents created with Copilot Studio have additional controls in the Power Platform admin center. A Microsoft 365 agent policy should therefore not be assumed to represent every development and governance setting for a Copilot Studio deployment. The relevant management boundary depends on how the agent was built and where it operates.
Copilot data preparation starts with existing access
Microsoft’s Copilot controls security and governance guidance recommends assessing oversharing through SharePoint Advanced Management and Microsoft Purview, subject to licensing. Its documented actions include running SharePoint data access governance reports, sending site access reviews to owners, and removing organisation-wide site access where appropriate.
This is particularly relevant to 1st Central’s stated employee use case of connecting information across different parts of the business. Easier discovery makes it important to establish whether the underlying access is appropriate. The recommendation is to review permissions, not to assume that broad access becomes safe because an AI assistant is the interface.
Microsoft also distinguishes restricting content discovery from restricting access. Its guidance lists SharePoint restricted content discovery and restricted access control as options while oversharing risks are remediated. Administrators should choose the control that matches their objective rather than treat a change in discoverability as interchangeable with a change in authorisation.
Microsoft Purview supplies evidence of Copilot activity
Microsoft documents Purview Audit for auditing Copilot and agent interactions, alongside eDiscovery capabilities for searching prompts and responses. It also describes retention and deletion policies for interactions through Purview Data Lifecycle Management.
Those capabilities give substance to the otherwise broad instruction to “monitor AI.” They allow organisations to plan what interaction records they need to retain and how authorised staff will investigate an issue. They do not, by themselves, establish whether an insurance decision was correct or whether a customer received suitable service.
Licensing remains part of the implementation decision. Microsoft’s security guidance separates foundational controls associated with A3/E3/G3 licensing from optimised controls associated with A5/E5/G5 licensing and relevant products. The presence of “Copilot” in a deployment description is not enough to establish entitlement to every monitoring or data-protection feature.
These are documented options for Microsoft 365 environments. They should not be presented as a complete control framework for 1st Central’s Azure-hosted insurance systems, or as proof of the insurer’s regulatory compliance.
1st Central’s customer-service agents remain a design direction
The next phase described by Davison is applying AI to repeatable, rules-based policy, claims, and customer-service processes. In this context, an agent would work towards a defined goal and potentially take actions, rather than only return text for an employee to review.
The customer story does not announce a named production service, a customer rollout date, or a particular insurance decision already being made autonomously. Its strongest supported claim is that 1st Central is designing services able to recognise when human help is needed and transfer the conversation to an adviser.
Davison’s description of insurance processes as rules-based explains why he sees an opportunity for automation. It does not establish that every customer situation can be handled by the same automated path. His own account retains people for supervision and for complex or sensitive interactions requiring judgement.
That makes the human handoff part of the proposed service design. The company says escalation to a person must remain available, while AI could make straightforward transactions faster and accessible at any time. The promise of round-the-clock automated handling should not be read as confirmation of round-the-clock human staffing; Microsoft’s account does not establish that operating arrangement.
For enterprise readers, the supported implication is that an agent proposal needs both a defined task and a defined point of escalation. A drafting assistant leaves an employee reviewing an output. A customer-facing service that takes action also needs a clear boundary around what it can complete and what it must transfer.
The same discipline applies to the claimed customer benefits. Better consistency, faster resolution, and more personalised insurance are outcomes the insurer expects. Microsoft’s story provides no customer-service measurements showing those outcomes from the planned agentic services. The present deployment and the future proposition should remain separate when assessing the investment.
What this means for your Azure and Copilot plans
Treat 1st Central as a reason to examine specific workloads, rather than as a purchasing template. Its combination of operational cloud systems, employee assistance, and planned customer-service automation offers several decisions an IT team can make now without assuming that the insurer has proved every part of its strategy.
For an Azure project, the immediate question is which operational task needs the data and how quickly it must respond. For a Microsoft 365 Copilot project, the nearer-term work is identifying a useful employee task, checking data access, and deciding how agents will be approved and monitored. A customer-facing agent adds a further requirement: defining the action it may take and the service available when it cannot finish safely.
The concrete takeaways are:
- Separate the Azure quote-processing workload, Microsoft 365 Copilot use, and proposed customer-service agents when building your own business case; the evidence for one does not validate the others.
- Treat 1st Central’s 600,000 daily quotes and roughly 2.5-second responses as reported operational context, not as an independently verified performance benchmark or capacity-planning specification.
- Use the 6,500-hour writing-assistant estimate as a prompt to measure a comparable task locally, including review and correction time, rather than forecast an equivalent saving.
- Check Microsoft 365 administrative roles, agent access and publishing policies, and any additional Copilot Studio controls before broadening an organisational agent deployment.
- Assess SharePoint oversharing and the available Purview auditing and retention capabilities against your actual licences and information-governance requirements.
- Keep automated service availability separate from human-support availability, and define the handoff before treating an agent as a replacement for an existing customer-service route.
1st Central’s strongest evidence concerns the foundations already in use: Azure-hosted insurance systems and employee tools aimed at identifiable tasks. Its customer-service ambitions extend that work into a more consequential setting, where an agent’s authority and access to human help become part of the service itself. For IT teams, the useful direction is clear: build the operational platform, measure assistance at the task level, and establish the boundaries before allowing software to act on a customer’s behalf.