Español

Guide · 30 October 2026

Black Friday incident checklist

Black Friday is 27 November and Cyber Monday 30 November. The outages that hurt most on those days are rarely new: they are the weak spots you already know about, under three times the load. Here are 30 checks for e-commerce and platform teams, in the order to do them.

Four weeks out: know your limits

• Peak traffic forecast agreed with marketing, per hour, including the promotion start times.
• Load test run at 1.5× the forecast peak against a production-like environment.
• The first component to fail under load is known, and its limit written down.
• Database connection pools, queue sizes and rate limits checked against peak, not average.
• Autoscaling tested to peak capacity, with cloud quotas and limits raised in advance.
• Every third party in the checkout path listed: payment provider, fraud check, tax, search, CDN, email.
• Payment providers and other critical partners told your expected volumes.

Three weeks out: see trouble early

• A single dashboard shows orders per minute, checkout errors, payment success rate and page latency.
• Alerts fire on business signals (orders dropping, payment failures rising), not only on CPU and memory.
• Alert thresholds adjusted for peak traffic, so normal Black Friday load doesn't page anyone.
• Synthetic checks run a full test purchase every few minutes.
• Third-party status pages and webhooks feed into the same alert channel.
• Noisy alerts reviewed and either fixed or silenced before the week.

Two weeks out: plan the response

• On-call rota for 26 November to 1 December published, with a named incident manager for each shift.
• Escalation contacts for every critical vendor tested, including out-of-hours numbers.
• Runbooks written for the five likeliest failures: payment provider down, checkout database saturated, CDN or cache failure, a third-party script slowing pages, stock sync failing.
• Each runbook has a "degrade gracefully" option, such as switching payment provider or turning off recommendations.
• Feature flags in place to switch off heavy, non-essential features under load.
• Rollback for the last release tested and timed.
• Who can decide to roll back, take down a promotion or pause marketing is written down.

One week out: freeze and rehearse

• Change freeze starts no later than 23 November, with an emergency-change route that needs two approvers.
• A 30-minute game day: simulate the payment provider failing and walk through the runbook.
• Status page and customer-service message templates ready for "checkout slow" and "payment issues".
• Customer service knows how to reach the incident channel and what they may tell customers.
• Backups verified with a restore test in the last 30 days.

On the day and after

• A war-room channel opens before the first promotion goes live, with updates every 30 minutes during an incident.
• Someone watches the business dashboard at each promotion start time.
• Every incident is logged with start time, impact and orders lost, even if it lasted minutes.
• Post-peak review booked for the first week of December, while memories are fresh.
• Anything that broke becomes a problem record with an owner, so it doesn't break again next year.

Questions