NexOps Operator: an AI assistant for client emails

The work behind each reply

A team of 15 operators, consultants and account leads handled dozens of active client conversations each day. Before replying, staff had to reread the thread, work out the context, choose an approach and check that their response did not promise something the business had not agreed to.

Working inside the existing inbox

We built NexOps Operator as a browser side panel for Gmail and Outlook 365. It works in Chrome, Microsoft Edge and Brave, so the team can keep using its existing inbox.

Operator reads the open conversation and prepares a complete reply. Staff can add a short instruction, review the draft and edit it before approval.

Checking the response

The assistant considers the conversation and the sender’s position, then gives a short explanation of its recommended approach. A separate review checks the draft against the company profile, looks for unsupported claims and flags commitments that have not been authorised.

Operator also records important promises, deadlines and actions from approved drafts against the relevant contact. Every response remains subject to human approval.

Estimated time saving

The calculation in this project used approximately six minutes for reading, understanding and drafting a substantive email manually, compared with about 90 seconds for analysis, review and approval with Operator.

Using a conservative saving of four minutes across 20 replies per person per day and 15 users gives approximately 20 hours of team capacity per day. Over a year, that is around 4,400 hours, or 2.4 full-time roles based on 1,800 productive hours per role.

At an estimated employment cost of £40,000 per role, the model puts the annual productivity value at roughly £90,000-£100,000. These are calculated estimates based on the stated workload and timing assumptions.

Where the email data goes

NexOps does not collect or store mailbox content, draft history, company profile data or usage telemetry on a central NexOps platform. The API key is stored locally on the user’s machine.

In the standard setup, relevant email content is sent to a commercial model provider to generate the response. For organisations that require processing within their own environment, the same workflow can run on a local model using client-controlled hardware.

The choice depends on the organisation’s data requirements and permitted use of external AI providers.

Discuss an email workflow

Tell us which parts of your email work take the most time and what needs human review.