A documented Coinbase campaign analysis and load-test checklist for aligning media reach, landing-page capacity, throttling and conversion recovery.
Short answer: answer: Coinbase’s 2022 Super Bowl campaign shows that media approval and capacity approval must be one decision. Model the one-minute arrival burst, load-test the complete registration journey beyond that forecast, define graceful degradation and measure successful completions—not only scans or visits. If the landing page, identity service, offer ledger or fraud controls cannot survive the agreed peak with headroom, cap exposure or redesign the call to action before buying the media.
Calling the campaign a failure misses what worked. Coinbase reported more than 20 million landing-page hits in one minute and engagement six times its previous benchmarks. A bouncing QR code turned an expensive television slot into a direct, measurable action and generated enormous attention.
Calling it an unqualified success also misses the operating lesson. Coinbase said the volume exceeded projections and led it to temporarily throttle systems. CDM’s position is that attention that outruns the promised path is not a creative trophy; it is a partially fulfilled demand obligation—even when the interruption is brief and the campaign ultimately performs well.
What the documented record shows
During Super Bowl LVI on 13 February 2022, Coinbase ran a 60-second spot built around a colour-changing QR code moving across a black screen. Scanning led viewers to a promotion that offered eligible new users $15 in Bitcoin and gave existing users a chance to enter a sweepstakes.
Coinbase later said its teams had load-tested the site for millions of simultaneous hits. The company reported more than 20 million landing-page hits in one minute, described the volume as historic and unprecedented, and said engagement was six times higher than prior benchmarks. It also said the volume caused temporary throttling and that engineers restored access quickly.
Those are company-reported figures, not an independently audited campaign dataset. The public record does not provide the complete traffic distribution, error rate, number of failed registration attempts, lost conversion value or exact duration and scope of each degraded service. That limitation is central: hits demonstrate demand, while reliability requires completion evidence.
The Demand Shock Readiness Test
The Demand Shock Readiness Test has six gates. Marketing owns the demand assumptions; engineering owns tested capacity; product owns journey integrity; finance owns cost exposure; and one incident commander owns the live decision. No function can approve the event alone.
Gate 1: forecast arrivals, not impressions
Translate audience into requests per second. Estimate viewers during the exact spot, scan rate, QR-resolution burst, repeat scans, bot traffic, refresh behaviour and downstream registrations. Use low, expected, high and extreme scenarios rather than one point forecast.
The timing shape matters more than the daily total. Twenty million requests spread across a day is a different system from 20 million landing-page hits in one minute. Television, livestreams, creator drops, ticket releases and flash sales can synchronise behaviour, defeating autoscaling designed for gradual growth.
Gate 2: test the entire customer path
Drive production-like synthetic traffic through DNS, CDN, landing page, authentication, identity verification, database, offer eligibility, fraud controls, notification and confirmation. Testing a static page proves little if the first dynamic request becomes the bottleneck.
AWS’s Well-Architected guidance warns against testing only components, using unlike environments or stopping at expected load. Google Cloud guidance similarly recommends controlled load tests, second-by-second analysis for rapid spikes and checking quotas. The relevant principle is platform-neutral: discover the breaking point before customers do.
Gate 3: agree service objectives and headroom
Set thresholds for availability, p95 latency, error rate, completed registrations, duplicate awards and recovery time. Load beyond the high forecast until the first constraint fails, then document the safe operating envelope.
CDM recommends at least 2× the high credible one-minute forecast for a single-shot national event when the infrastructure and budget permit. That is an operating recommendation, not a universal engineering standard. If 2× is impractical, leadership must explicitly accept a lower envelope and reduce media concentration or simplify the journey.
Gate 4: design graceful degradation
Decide what survives when capacity is exceeded. A static page can explain demand, preserve the offer deadline and place users into a controlled queue. Nonessential personalisation, analytics calls or account features can shed load. Throttling can be a sound protective control when it is planned, observable and fair.
The customer must not see an ambiguous failure that encourages frantic retries. Retries can multiply the shock. Provide a stable status, preserve eligibility where reasonable and prevent duplicate registrations or awards.
Gate 5: rehearse live control
Run a game day with marketing, engineering, customer support, security, fraud and legal. Practise pausing paid amplification, swapping the QR destination, activating a queue, extending an offer, publishing status and preserving incident data.
The incident commander needs thresholds, not vibes: for example, freeze additional traffic when successful registration falls below the agreed rate for three consecutive minutes. The exact rate and window come from the service objective and customer consequence.
Gate 6: reconcile attention with completion
After the event, connect ad exposure, scans, unique landing sessions, queue entries, successful accounts, verified eligibility, funded actions, support contacts and fraud. Report abandoned users separately from rejected users. A throttled request is not automatically a lost customer, and a hit is not automatically a person.
Calculate cost per successful eligible action and the value of recovered attempts. This prevents the biggest top-of-funnel number from defining the campaign’s history.
Reader asset: campaign load-test checklist
| Readiness item | Evidence required | Live trigger |
|---|---|---|
| Burst model | Low/expected/high/extreme requests per second | Actual arrival exceeds tested envelope |
| Full journey | End-to-end production-like test results | Downstream error breaches objective |
| Capacity | Quotas, autoscaling, warm capacity and breaking point | Headroom falls below declared margin |
| Degradation | Queue, static fallback, retry control and offer policy | Activate before uncontrolled failures |
| Command | Named incident lead and channel decision owners | Pause traffic at predeclared threshold |
| Customer recovery | Eligibility preservation and support script | Extend or restore when users were blocked |
| Measurement | Hits-to-completion reconciliation | Do not report success until matched |
Attach the checklist to the media authorisation. The total spend should not be releasable until the technical owner signs the tested envelope and the campaign owner signs the traffic assumption. Re-run the critical path after final creative, offer or identity-flow changes.
Was temporary throttling a failure?
It was evidence of a capacity mismatch between observed demand and the unthrottled system, but not necessarily catastrophic incident management. Deliberate throttling can preserve core services and enable recovery. Coinbase said systems returned quickly and celebrated the extraordinary response.
The fair conclusion is mixed: the creative generated exceptional, measurable attention; the demand exceeded projections and required temporary constraint; public data is insufficient to quantify lost value. Strong case analysis holds those facts together rather than selecting the version that best supports a predetermined lesson.
Related guides
Frequently asked questions
Did Coinbase’s Super Bowl ad crash its website?
Coinbase said the campaign produced more than 20 million landing-page hits in one minute and that it temporarily throttled systems because volume exceeded projections. Contemporary users and reports described access delays, and Coinbase announced that it was back online. “Crash” is a common shorthand, but “temporary throttling and degraded access under exceptional demand” is closer to the company’s own account.
The caveat is that Coinbase did not publish a full event postmortem with error rates, precise duration or affected-service breakdown, so the scale of failed customer journeys cannot be calculated from public evidence.
Was the Coinbase QR campaign successful?
It clearly succeeded at generating attention and direct response: Coinbase reported 20 million-plus landing-page hits in one minute and engagement six times prior benchmarks. Whether it achieved profitable, trustworthy acquisition requires verified registrations, funded activity, incentive cost, fraud, retention and lost attempts—data not fully disclosed publicly.
The responsible verdict is therefore “creative and demand-generation success with a documented capacity constraint,” not a simple win or loss. The caveat is that a company may reasonably value awareness and cultural salience beyond immediate account economics, but it should label those objectives separately.
How much traffic should a campaign landing page be able to handle?
Enough to clear the high credible burst forecast with explicit headroom while meeting customer-level service objectives. Model requests per second and concurrency across the whole journey, then test until failure. CDM’s 2× high-forecast suggestion suits some one-shot national events, but cost, architecture and consequences differ.
A lower tested envelope can be acceptable when the campaign uses a queue, traffic shaping or staggered exposure. The caveat is that autoscaling and third-party services have quotas and warm-up behaviour; theoretical capacity is not tested capacity.
Is throttling bad during a marketing traffic spike?
No. Planned throttling can be the safest response to prevent cascading failure, fraud or damage to unrelated customers. It becomes poor experience when users receive ambiguous errors, eligibility is lost, retries amplify load or the team cannot measure and recover affected attempts. Define fairness, messaging, queue behaviour and offer treatment in advance, then rehearse activation.
The caveat is regulated or financial activity: limits may also serve security and compliance needs, so marketing must not bypass them merely to improve conversion.
What should marketers ask engineering before a major launch?
Ask for the tested requests-per-second envelope, full-journey bottleneck, quotas, headroom, autoscaling delay, graceful-degradation plan, incident thresholds, dashboard and recovery-time objective. Supply engineering with the spot timing, audience concentration, call-to-action mechanics and traffic scenarios. Both teams should sign the same record. Include the offer ledger, identity provider and fraud vendor rather than testing the landing page alone.
The caveat is that engineering cannot guarantee unknown demand or eliminate every failure. The goal is an evidence-based operating range and bounded response, not a promise of infinite capacity.
Which metric matters most during a demand shock?
Successful customer completion within the promised service level matters more than raw hits. Monitor it alongside error rate, p95 latency, queue depth, duplicate actions and affected users. Hits help explain demand; they do not show whether people obtained the offer or completed registration.
The caveat is incident triage: infrastructure saturation indicators may lead the first technical response even though completion remains the business outcome. Use both views and reconcile them after recovery rather than allowing one dashboard to define success.
Next decision: What Should a Marketing Team Do When a Critical Tool Fails on Launch Day?
Related reading: How Do You Repair a Landing-Page-to-CRM Handoff That Stops Recording Leads? · How Do You Track Campaign Dependencies and Risk in monday.com? · Why Do Marketing Campaigns Fail Even When the Strategy Looks Sound?
Sources and research notes
- Coinbase campaign review — company-reported creative rationale, traffic, engagement and throttling; checked 26 September 2026.
- Coinbase 2026 release recalling the 2022 result — later company confirmation of more than 20 million hits in one minute; checked 26 September 2026.
- Bloomberg/Fortune contemporaneous report — live chronology and company executive statements; checked 26 September 2026.
- AWS Well-Architected load-testing guidance — production-like, end-to-end and beyond-expected-load testing principles; checked 26 September 2026.
- Google Cloud load-testing practices — spike control and second-by-second analysis guidance; checked 26 September 2026.
- Limitations: Coinbase has not published a complete audited postmortem or funnel dataset for the event. The readiness gates, 2× suggestion and live triggers are CDM recommendations, not Coinbase’s internal standards or cloud-provider guarantees.
This article is editorial guidance. Apply the principles in proportion to your market, evidence, and responsibilities.
