An AI support agent reduced required resources by 92%.

A customer support operation reported a 92% reduction in required resources after Outerscope implemented a company-aware AI agent accessible through Gmail, MCP and Second Brain.

The support function served 35,000 clients across eight services. The system combined company context, tools for taking action, and human approval within the tools used to handle support.

What we identified

Supporting a large client population across eight services required employees to understand both the customer request and the relevant company context. Resolving a request could also involve an operational action, such as requesting and issuing a refund from a 3rd party, rerouting a replacement for a faulty product from an alternative provider, updating authorized records across internal platforms, and more.

The challenge therefore included understanding, response preparation, and execution. A useful system needed access to the context behind each request and the tools required to carry out the proposed resolution.

The organization also needed people to retain approval over consequential actions. The implementation had to fit the support workflow while making the work easier to handle.

What we built

A company-aware agent across Gmail, MCP and Second Brain

Outerscope made the agent accessible through Gmail, MCP and Second Brain, connecting the support workflow to the company information it needed. The system could interpret requests in context and prepare work for review.

These entry points made the assistant available from support conversations and the company’s shared working context.

Tools that could carry out a resolution

The agent was equipped with tools to make operational changes, including issuing refunds and reordering products. These capabilities connected the response to the action required to resolve the request.

Human-in-the-loop approval remained part of the workflow. Staff could review proposed work and approve consequential actions before execution.

Improvement through reviewed work

The agent was designed to improve recursively through the support process. Reviewed interactions and feedback helped refine how it handled future requests, connecting daily use with continued improvement.

How the support process changed

The system brought company context and action-taking capabilities into one support workflow. Employees could review prepared responses and proposed actions, using their judgment at the points where approval was required.

This reduced the resources needed to operate support across the client base. The implementation addressed the work of understanding the request, preparing a resolution, and carrying out the approved action.

Human review was part of the operating design. The agent supported the team across the process while keeping consequential decisions with people.

What the engagement achieved

The reported resources required to run customer support fell by 92%. The system supported an operation serving 35,000 clients across eight services, with access through Gmail, MCP and Second Brain.

This measures support resource requirements; it does not measure a change in employee headcount. The implementation combined contextual understanding, operational tools, and employee approval around the task of resolving customer requests.

For service organizations, this approach provides a practical way to improve support where responses depend on company knowledge and resolution requires action across connected systems.