Automation case study

Managing Email Overload: AI-Assisted Triage, Focused Digests and Controlled Cleanup

A personal workflow for daily priorities, focused technical reading, and controlled mailbox cleanup.

Email combines several different jobs in one place: requests that need action, information worth reading, and routine material that eventually loses its value. Treating all three alike makes an inbox difficult to manage.

For my own mailbox, I developed an AI-assisted workflow around three functions: general triage, a Cybersecurity / AI digest, and rule-based cleanup. Each has a distinct purpose and a defined boundary on what it can do.

01 · Three functions

Three functions, three decisions

FunctionDecision it supportsMailbox action
General triageWhat needs my attention?Read messages and send a prioritized digest
Cybersecurity / AI digestWhat deserves focused technical reading or follow-up?Review relevant material and send a subject-specific digest
Rule-based cleanupWhich routine messages meet the agreed cleanup criteria?Apply review labels or move eligible messages to Trash

These are recurring workflows using an existing AI assistant and its connected Gmail tools. This case study focuses on that setup, which followed an earlier locally hosted agent.

02 · General triage

General triage: making the next action visible

The general digest organizes incoming messages into urgent items, actions for today, actions for later, waiting items, and background information. Actionable entries explain why the message matters, suggest a next step, and link back to the original email.

Related deliveries can be consolidated into one item. Routine newsletters and promotions are grouped so that they remain visible without dominating the report. The digest includes a scan window and counts, helping me understand what the review covered.

One inspected digest recorded 34 deliveries consolidated into 33 distinct items, with three identified as actionable. Those figures illustrate a particular run; they are not a measurement of classification accuracy.

The triage workflow can suggest cleanup candidates, but those suggestions do not themselves authorize changes to the mailbox. Its task is to help me decide what to do next.

03 · Cybersecurity / AI digest

From inbox priorities to focused technical reading

The Cybersecurity / AI digest serves a different reading need. It brings relevant technical developments, security notices, research, and professional learning opportunities into a focused report.

The reviewed example included links to vendor advisories and research, identified outstanding items as carryovers, and distinguished conditional checks from confirmed applicability. A product appearing in a security notice was a reason to investigate the environment, rather than proof that a particular system was affected.

That distinction also applies to AI research. Vendor-published findings need to be recognized as such, with room for further evaluation before drawing operational conclusions.

The general digest supports daily attention; the specialist digest supports deeper reading and follow-up. Both preserve the need to consult the original material and exercise technical judgment.

04 · Controlled cleanup

Cleanup: translating preferences into explicit rules

Cleanup uses defined policies for selected categories of recurring mail. The reviewed reports distinguish short-lived feeds, promotions, newsletters, automated reports, and other approved categories. Some messages enter a review stage before becoming eligible for later cleanup.

Exceptions are as important as inclusion criteria. One recorded run explicitly preserved security alerts and messages outside the intended digest category. It also left reports untouched when they did not satisfy a rule's exact sender pairing. These examples show why a broad sender name or a general impression of low importance is insufficient as a cleanup policy.

This is rule-based work performed through an AI-assisted workflow. The existence of explicit rules does not establish error-free execution, so the reports remain part of the review process.

The ten recent cleanup summaries reviewed for this case reported between 12 and 25 messages moved to Trash per run, with additional messages placed under a review label. They reported no direct permanent deletion. These are reported actions, not an independent audit of every affected message.

Moving a message to Trash provides a limited recovery opportunity. Gmail normally permanently deletes it after 30 days, and manually emptying Trash can shorten that opportunity. Review therefore needs to happen within that window. Gmail Help

05 · Reliability

Reliability: checking the work and recognizing gaps

The inspected triage email contained both HTML and plain-text versions. It also documented a recovery decision: when no successful morning digest was found in Sent Mail, the afternoon scan resumed from the previous afternoon's verified checkpoint. That made the extended coverage window explicit.

A separate monitoring task checks for the expected triage digest. Its history records a warning when a digest was missing, as well as a problem with the monitoring task itself. Monitoring adds visibility, but a check running within the same platform cannot provide independent protection against every platform failure.

Finding a digest in Sent Mail confirms that a sent copy exists. It does not prove that I read it, that every relevant message was included, or that every recommendation was correct.

06 · Results and lessons

Results, contribution and lessons

The result is an operating personal workflow with distinct outputs: a prioritized action digest, a focused technical digest, and cleanup reports grouped by rule. The reviewed records show repeated operation, a documented recovery after a missed run, and explicit handling of exceptions. They do not establish a measured time saving, a guaranteed detection rate, or uninterrupted service.

I defined the priorities and cleanup preferences, reviewed the outputs, and refined the workflows with AI assistance. The underlying assistant and Gmail integration are existing tools; my contribution is the workflow design, operational requirements, and ongoing evaluation.

The most transferable lesson is to match the automation to the decision. Interpretation benefits from context and judgment. Mailbox changes need explicit criteria, limited authority, and reviewable results. Making those boundaries visible is central to building automation that remains useful in daily work.