Working tool · v0.1 · sample data · built 3 Oct 2026

ATLAS

The critical-case tracker I built at Google, rebuilt for the browser. It watches a case queue for six kinds of trouble, writes the morning brief, and triggers the one email I never automated: asking the customer, in my own words, whether the fix actually worked.

Load your own

bettinajerome.tech/atlas · sample data

Morning brief

Quarterback today

SLA due in the next 24 hours

Who holds the baton open cases

Queue

Case	Account	Pri	Status	Age	SLA	Customer	Vendor	Flags

Confirmation email ·

A draft, in my voice. Read it, change it, then send it yourself. This tool never sends anything.

Rules

plain rules you can see and change · the Google version added a reasoning layer on top

Risk order in the brief: P1, then overdue, then escalated, then how long the case has been silent. Change a threshold and the brief, the counts and the flags recompute.

Load your own

a CSV export from your case system · parsed in your browser · nothing is uploaded

Columns

id, account, subject, priority (P1–P3), status (New · In progress · Awaiting customer · Awaiting vendor · Resolved), opened, sla_due, last_customer, last_vendor, escalated (yes/no), owner, resolved_at, customer_confirmed (yes/no), contact

Dates as ISO (2026-10-02T14:30) or any form your browser can read. Leave a cell blank if you don't have it; the rule that needs it is skipped for that case. Extra columns are ignored.

The original · Google, 2022 to 2026

Built because nobody should learn about a Sev-1 from the customer.

At Google I ran enterprise accounts where the queue was the risk. ATLAS, the Autonomous Triage and Live Alert System, streamed Salesforce case data into BigQuery, fired on SLA thresholds, used Vertex AI to work out why a case was stalled and who held the next move, and pushed a morning brief and live alerts to Slack. It ended manual queue-checking, and it stopped us relying on customers to tell us something was wrong.

One flag was never automated on purpose. When support marked a case resolved, I emailed the customer myself to ask whether it was actually fixed, in my own words. The tool's job was to make sure I never forgot to.

ResultMore than 30% fewer formal customer escalations, sustained. My number, from the account data at the time.

Ran onSalesforce · BigQuery · Vertex AI · Looker · Slack

StatusStill running at Google as of my last day. This page is a rebuild on sample data, with plain rules instead of the reasoning layer.

Who it servedTAMs first. CSMs caught churn signals early; engineering saw stalled cases; leaders saw risk, not a historical dashboard.

About this tool

What it won't do

It never emails anyone. It never changes a case. It draws no conclusions a rule can't explain. The confirmation email is a draft you read and send yourself, because that message only builds trust if it is actually from you.

Privacy

Your CSV is parsed in the page and kept in memory until you close the tab. Nothing is uploaded, stored or sent. The sample queue is six fictional accounts; any resemblance to a real company is accidental.

Read it with the others

ATLAS catches trouble at the escalation stage. The Readiness Scorer and the Adoption Tracker are the same instinct pushed earlier, before a rollout stalls.