The First Truly Mobile Olympics: London 2012's Lessons for Every Site

One week into the London Games, the interesting numbers are not on the medal table. The BBC, serving its home audience, reports Olympic traffic at volumes past anything in its history. Mobile and tablet devices carry a share that would have sounded absurd when Beijing closed four years ago. On peak days, roughly a third of its Olympic traffic, and higher in the evenings. NBC reports streaming records alongside the same shape of device data.

And the shape is the story. Mobile usage spikes during live moments, evenings, and event finishes. Which is to say, precisely at the peaks. The second screen is not hypothetical anymore. It is measurable, it lights up next to the television, and it changes a planning assumption we have been circling for two years.

Beijing was our masterclass in delivering video at scale. London is the masterclass in audience behavior at scale, and the lesson applies to every organization with a big day on its calendar.

Peak behavior is mobile behavior

Here is the assumption London breaks. Most organizations, when they think about mobile at all, work from their average. Mobile is fifteen or twenty percent of our traffic, therefore mobile is a fifth of our problem.

But averages describe ordinary Tuesdays. The Games data shows what many analytics dashboards have been hinting. At moments of high interest, audiences reach for the nearest glass rectangle, and the nearest one is increasingly in a pocket or on a couch arm. The BBC’s evening mobile surge is people watching television with a browser open. The lunchtime spike is desks abandoned for phones.

When attention peaks, the device mix shifts toward mobile. Which means your traffic spikes skew more mobile than your averages do, and every capacity plan built on average device mix under-provisions for the exact moments it exists to survive.

The inauguration analysis taught us that peaks are concentrated in time. London adds the corollary. They are concentrated on small screens.

What this changes in the planning we already teach

• Capacity modeling gets a device column. The peak-math exercise from our campaign playbook, audience times response rate across the arrival window, should now estimate the mobile share of that peak, and test accordingly. Cellular network latency, smaller pipes, and the retry behavior of impatient thumbs all shape load differently than office broadband does. Test your peak scenario with mobile user agents and throttled connections, not just concurrency counts.

• Page weight becomes peak-survival engineering. A two-megabyte page is an annoyance on a desktop and a failure on a cellular connection under load. The lightweight fallback templates we prescribe for surge days earn double under a mobile-heavy peak. The image-weight discipline from this spring’s Retina coverage stops being a quality conversation and becomes a capacity one. Light pages serve more people from the same infrastructure. That was always true. The mobile peak makes it decisive.

• The interaction layer must assume touch at the worst moment. The visitor arriving during your peak is disproportionately on a phone, stressed, and task-focused. That holds on decision day, on giving day, and in a crisis. The responsive commitments our design practice has made this year are not aesthetics. They are the difference between that visitor completing the task and calling the overwhelmed phone line instead.

The infrastructure holding, and why

Credit where the engineering earns it: the delivery layer is holding, and the reasons are the familiar ones. The BBC’s Olympic architecture, per its own engineering posts, leans on aggressive caching and edge delivery for the pages everyone requests. Schedules, medal tables, live text. Origin work is reserved for what genuinely varies. That is the Beijing separation, four years matured, now serving a mobile-heavy load its designers explicitly modeled.

The streaming volumes ride the CDN pattern that is no longer news, which is itself the news. What was record-breaking heroics in 2008 is provisioned expectation in 2012. The down-market trend our CloudFront coverage traced means the same patterns now protect institutional sites at institutional budgets.

The failures visible so far are instructive at the edges. Third-party widgets and social embeds buckling under load their vendors did not provision. The dependency lesson from June 25, 2009, at Olympic scale. And occasional app backends straining while the websites hold, a quiet argument for the open web’s caching maturity.

Nothing this fortnight has revised the playbook. Everything has confirmed it at the largest scale yet, on the smallest screens yet. Which is exactly the confirmation worth having before your own next peak.

The second-screen content lesson

A quieter observation from watching the Games coverage patterns. The mobile surge is not primarily people streaming video on phones. It is people seeking context while the television carries the picture. Schedules, results, athlete backgrounds, the answer to the argument happening on the couch.

The content that wins the second screen is fast, factual, and findable. Exactly the structured, well-labeled content our information-architecture and structured-data work produces.

For institutions, the transferable version. During your organization’s live moments, audiences on phones want the schedule, the names, and the where-do-I-go. The game, the gala, the commencement: same story. The organizations that make that context effortless collect the engagement. The ones that make phones pinch through a desktop brochure lose it to a search engine and whoever answers there.

Peak content strategy is a real discipline, and it is mostly about respecting what a person with one thumb and divided attention actually needs.

Reading your own data, and the December question

The homework, as always, is local. Pull your last three traffic spikes (campaign days, event days, news moments) and compare their device mix against your monthly average. Our client data almost universally shows the London pattern in miniature. Spike-day mobile share runs well ahead of baseline, and the gap widens each year. That gap is your planning number.

Then look at the calendar. Whatever your organization’s next scheduled peak, it will be the most mobile moment your site has ever served, because every peak now is.

The Games will close in two weeks and the retrospectives will fixate on streaming totals. Keep the device chart instead. Four years from now in Rio, the mobile share will not be a third. The trend only points one way, and the sites that internalize it now get four years of compounding readiness. The averages will catch up to the peaks eventually. Planning should get there first.