Incident management for e-commerce
E-commerce incidents cluster around busy periods and recent releases, and a broken checkout is felt within minutes. Noci AI triages every alert as it arrives, its agents investigate with your telemetry and recent code changes, and you choose, step by step, whether an agent asks before it acts.
What makes an e-commerce incident different?
An online shop has a visible spine: browse, search, basket, checkout, confirmation. A fault anywhere along it is a fault a shopper meets directly, and many shoppers meet it at once. The trouble is rarely a whole site going dark. It is more often one link in that chain: a search that returns nothing, a promotion code that stops working, a payment step that hangs, a delivery option that disappears.
Three things shape the work of whoever is on call:
- Load is uneven. A campaign, a sale or a holiday sends traffic up fast, and small weaknesses become incidents.
- Releases are frequent and timed. Teams ship changes ahead of busy periods, so a recent change is often the first suspect.
- Many parts belong to others. Payment, tax, delivery and fraud checks usually come from outside providers, and the cause may sit with one of them.
Why do busy periods make alerting worse?
Because volume rises exactly when attention is scarce. A slow database or a struggling provider can fire many alerts that all describe one trouble. The on-call engineer is paged repeatedly, reads each alert separately and works out the connection by hand, while colleagues in support and marketing are asking whether the shop is broken.
The cure is not fewer monitors. What helps is a first reading of every alert before a person has to give it theirs.
How does Noci AI handle an alert during a busy period?
| The problem | What Noci AI does |
|---|---|
| A flood of alerts about one trouble | Triages every alert as it arrives and scores its noise, with the reasons shown |
| Is it the release or something else | Agents investigate with your telemetry and your recent code changes, and report a hypothesis with the evidence |
| Reaching the right person quickly | Paging by phone, SMS, Slack and Microsoft Teams; an AI voice agent phones and briefs |
| Which steps an agent may take on its own | You choose, step by step, whether an agent asks before it acts |
By default, paging goes out at once and AI triage runs alongside it, so nobody waits while the result is worked out. If you choose, triage and rules run before escalation and paging waits for them. A team that dreads a late page during a sale can keep the default.
How should a shop set the oversight choice before a busy period?
Decide in calm, not in the middle of the sale. Walk through the steps an agent could take and sort them:
- Visible to shoppers. A reply or message a shopper will read is a candidate to ask first.
- Hard to reverse. A step that is difficult to undo is a candidate to ask first until you have watched it behave.
- Hides a signal. A step that silences a monitor is a candidate to ask first, because being wrong means missing the next alert.
- Routine and local. A step that is easy to undo and affects one service can run on its own.
In Noci AI you choose, step by step, whether an agent asks before it acts. The principles behind them are in human in the loop for incident agents.
What does a person decide?
The agents’ investigation is a hypothesis with its evidence, and a person judges it. On-call schedules and escalation policies decide who is paged and in what order. Which agent actions wait for a person is your choice, step by step. To see how alerts and recent changes are lined up, read correlating alerts and deploys; for the sorting of alerts, read alert triage for SRE teams.
Questions
Can Noci AI connect an incident to a recent release?
Its agents read your telemetry and your recent code changes together and report a hypothesis with the evidence behind it. A person decides whether the release is the cause and what to do.
What happens to alerts during a busy period?
Noci AI triages every alert as it arrives and scores its noise, with the reasons shown. By default, paging goes out at once and triage runs alongside it; paging waits for triage only if you choose that.
Will an agent change anything in the middle of a sale?
That is your choice, step by step. You choose whether an agent asks before it acts, so you can decide in advance which steps wait for a person during a busy period.
How does the on-call engineer get reached?
By phone, SMS, Slack channel or Microsoft Teams, following your escalation policies. An AI voice agent can phone the on-call engineer and brief them on the incident.