In short
- The stakes are public: Adobe Analytics put US online spending on Black Friday 2025 at $11.8 billion, and Shopify’s merchants peaked at $5.1 million a minute at 12:01 p.m. Eastern. A store’s own number is its revenue per minute in its peak hour, and that is the figure every test below is priced against.
- The six-week plan runs the money flows first (record them), then the same load at every release (compare the percentiles), then the layout (baselines before the banners change), then security and accessibility, and ends with a freeze and monitors, not with new tests.
- The failure to plan for is not your code: Cloudflare’s November 2025 outage took three hours and was a configuration file that doubled in size. The rehearsal that matters is what your checkout shows when a dependency answers 5xx.
About this article
- Format
- Scenario walkthrough: how a team would use the platform for one job; a scenario, not a customer story.
- Written for
- QA leads and test managers · Engineering managers · Product owners
- Where it applies
- The rules it cites are specific to United States, Ireland, Germany, France and Netherlands. The spending and lawsuit figures are United States numbers. The European Accessibility Act applies to stores selling into the EU wherever the company is based, and every store should plan with its own revenue per minute.
This is a scenario walkthrough of a pre-peak test plan. It is not a customer story; the store is invented so the plan is concrete, and every outside number is in the sources at the end. The plan assumes the Team plan’s limits and says where they bite.
The stakes, in public numbers
Adobe Analytics reported $11.8 billion of US online spending on Black Friday 2025, up from $10.8 billion a year earlier, with shoppers spending $12.5 million a minute between 10 a.m. and 2 p.m. Shopify’s merchants sold $14.6 billion over the Black Friday to Cyber Monday weekend, 27% more than the year before, with sales peaking at $5.1 million a minute at 12:01 p.m. Eastern on the Friday; Shopify notes its figures are unaudited. A mid-sized store does not see those totals, but it sees the same shape: a peak hour that carries a disproportionate share of the season, which is why a slow checkout at 12:01 costs more than a dead one at 3 a.m.
The outage to plan for was not a store’s own. On 18 November 2025 a permissions change in one of Cloudflare’s database systems produced duplicate rows in the file that configures its Bot Management product; the file doubled in size, crossed a limit of 200 features in the software that reads it, and the network began returning errors at 11:20 UTC. Core traffic was flowing again by 14:30. Cloudflare states the incident was not caused by an attack of any kind. Every store behind that network, including stores that had tested everything they owned, was down for three hours. The lesson for a test plan is that the dependency drill is not optional.
The scenario
The store takes about 2,000 orders on an ordinary day and expects ten times that on the peak day. One checkout, card and a buy-now-pay-later provider, a search page, a promo banner that marketing changes weekly from now until January, and a mobile share of traffic above 70%. The team is five engineers and one QA lead, on the Team plan, with a code freeze planned a week before the peak.
Week six: record the money flows
Record the journeys that earn money or trust: search to product to cart to paid order with each payment method, sign-in, the guest checkout, the returns request. Each recording keeps every locator it saw per element, the network calls each step caused and a screenshot, and replays as a regression test straight away. Record against a test account with synthetic data; a login typed during a recording is kept, encrypted, so the replay can sign in, and that is a reason not to record with a real customer account. Put the flows in a folder and schedule it nightly. Scheduled folder runs use Chromium; the WebKit and Firefox replays run on the execution grid, which the Team plan includes.
Week five: the same load, every release
Here the limits need stating first. Dokimore generates load from its servers in India, so latency includes the distance to the target, and the Team plan enforces a cap of 250 virtual users per run when the run starts. Two hundred and fifty concurrent users against a store expecting twenty thousand orders in a day is not a capacity test of the peak, and this plan does not pretend it is. What it is, is a regression instrument: the same profile, built from the recorded checkout flow, run after every release with the same virtual users, ramp and duration, so that P50, P95, P99 and the error rate are comparable from week to week. A release that moves P95 on checkout from 800 ms to 1.4 s at the same load has a problem that will be worse at the peak, and the plan catches it in week five rather than on the day.
Capacity for the peak itself belongs to the hosting provider’s own tooling and a load generator sized for it; a profile set up by hand on the load screen can be taken there as a k6 script (a plan built from the recorded flow cannot), there is no JMeter import or export, and the plan says which tool carries the peak.
Week four: baselines before the banners
Marketing will change the promo banner every week, and a banner is a layout change on the page that earns the money. Capture a visual baseline per page and per viewport now, with the mobile viewport first given the traffic share, and set the pixel threshold. From here on, an intended change becomes the new baseline with one click and an unintended one, a collapsed product grid or a pay button pushed below the fold, is a finding with a diff image. The element-drift check on the DOM catches the button that moved even when the pixels look similar.
In the product
Week three: security, with the ownership proof done early
A security scan against the store runs only after ownership is proven by a DNS record, a token file or a meta tag, so do that in week three rather than discovering it in week one. The scan covers the OWASP Top 10 (2025) and the OWASP API Security Top 10 and reports a finding as confirmed only with the request, the response and a proof of concept a reviewer can replay; everything else is an observation. Fix the confirmed findings, scan again, and keep the evidence bundle; if a payment provider or an acquirer asks what was tested before the season, that bundle is the answer.
Week two: the checkout a screen reader can use
Seventy percent of the 5,114 digital accessibility lawsuits UsableNet counted in the United States in 2025 targeted e-commerce, and 36% of the top 500 online retailers were sued at least once that year. Audit the checkout pages with axe-core against WCAG 2.2 A and AA, fix what automation finds, and read the manual-review list, because automation judges a minority of the criteria and the pay button’s label is the kind of thing a person has to check. If you sell into the EU, the European Accessibility Act has applied since 28 June 2025.
Week one: freeze, monitors and the dependency drill
No new tests in week one. Set the API collection that exercises checkout, search and a request that depends on the payment provider, a test-mode charge or its status endpoint, to run on a schedule as a monitor, with alerts to Slack, Microsoft Teams, Opsgenie or a webhook, so a failing assertion on the weekend reaches a person. Then run the drill the Cloudflare outage argues for: on the staging store, make the buy-now-pay-later provider answer 500 and watch what the customer sees. If the answer is a spinner, the fix is a message and a fallback to card, and it is a week-one fix because it is a page change, not a platform change.
On the day, nobody runs a load test against production. The monitors run, the folder runs nightly, and the evidence from the last six weeks is what the team reads if something breaks.
For the manager reading this
Divide last year’s peak-hour revenue by sixty. That is the cost of a minute, and the figure every item above is priced against. The plan costs the QA lead about a day a week for six weeks. Recording, replay and the seven evidence files per session are on every plan, including Free; load testing, visual baselines and security scans start at Solo Pro, and the grid and the 250 virtual-user cap are Team.
What to do next
- Divide last year’s peak-hour revenue by sixty: that is the cost of a minute, and the number every check below is priced against.
- Record the money flows this week with a test account and synthetic data, put them in a folder and schedule it nightly (recording your first test).
- Build one load profile from the checkout recording and run it at the same virtual users after every release, so P95, P99 and the error rate are comparable week to week (load testing basics).
- Capture visual baselines per page and viewport before marketing changes the banner, the mobile viewport first.
- Prove ownership of the storefront domain early so the security scan can run in week three, and audit the checkout pages against WCAG 2.2 A and AA in week two.
- Freeze a week before the peak, run the API collection as a scheduled monitor with alerts, and rehearse the checkout on staging with a payment provider answering 500.
Questions people ask
Can a 250 virtual-user load test prepare a store for Black Friday?
Not as a capacity test, and this plan does not use it as one. Within the Team plan’s enforced cap it is a regression instrument: the same checkout profile run at the same load after every release, so P95, P99 and the error rate are comparable week to week and a release that slows checkout is caught before the peak. Capacity for the peak itself needs a generator sized for it, usually with the hosting provider.
How much did Black Friday 2025 take online?
Adobe Analytics reported $11.8 billion of US online spending on Black Friday 2025, with $12.5 million a minute between 10 a.m. and 2 p.m., and Shopify reported $14.6 billion across its merchants over the Black Friday to Cyber Monday weekend, peaking at $5.1 million a minute. Shopify states its figures are unaudited. The number a store should plan with is its own revenue per minute in its own peak hour.
What broke in the Cloudflare outage of November 2025?
A permissions change in a database system caused duplicate entries in the configuration file for Cloudflare’s Bot Management product. The file doubled in size, exceeded a 200-feature limit in the software that reads it, and the network returned errors from 11:20 UTC until core traffic recovered at 14:30. Cloudflare says it was not caused by an attack. For a store, the lesson is to rehearse what checkout shows when a dependency fails.
Why record the flows with a test account?
A recording captures what the browser sees, including what is typed. A login typed during a recording is kept, encrypted, so the replay can sign in, and card details belong to a payment sandbox, never to a real customer. Record against a test account with synthetic data and the replay stays safe to run nightly.
Does the test plan change for a store on Shopify or another hosted platform?
The shape stays the same and the ownership proof changes. Recording and replay work against any page you are permitted to test; a security scan needs proof you own the target, which on a hosted platform means the DNS record or meta tag for your own storefront domain, and some platform pages are out of scope because you do not own them.
How we measured this
- The store, its order volumes and its team are a scenario; no customer is described. Every outside figure is quoted from the source named beside it, on the date it was read.
- Dokimore’s behaviour is the product on the article date: load generated from India within an enforced cap of 100 virtual users on Solo Pro and 250 on Team, a profile set up by hand on the load screen exports as a k6 script and as a JSON file, a plan built from a recording has no export, no JMeter import or export, scans only against proven-owned targets, scheduled folder runs on Chromium, API collections as scheduled monitors with Slack, Microsoft Teams, Opsgenie and webhook alerts.
Sources
- TechCrunch, "Black Friday sets online spending record of $11.8B, Adobe says" (29 November 2025): $11.8 billion, $12.5 million a minute between 10 a.m. and 2 p.m. read
- Shopify, "Shopify merchants generate record-breaking $14.6 billion in Black Friday Cyber Monday sales" (2 December 2025): up 27%, $5.1 million a minute at 12:01 p.m. EST, figures unaudited read
- Cloudflare, "Cloudflare outage on November 18, 2025": 11:20 UTC start, 14:30 core recovery, feature file doubled past a 200-feature limit, not an attack read
- UsableNet, 2025 Year-End Digital Accessibility Lawsuit Report: 5,114 lawsuits, 70% e-commerce, 36% of the top 500 retailers sued read