Kubernetes cost management for a national consumer campaign

Cloudreason supported an international marketing agency for three years on a national consumer campaign with a traffic profile most platforms never face: extreme, time-bound peaks, concentrated into campaign windows, in front of a very large national audience.

The problem

Campaign infrastructure inverts the economics of normal cloud workloads. A typical platform earns its keep every day; campaign infrastructure earns its keep in short, scheduled bursts, on a budget agreed months in advance, for an audience that arrives all at once, driven by advertising spend that makes the peak both enormous and largely predictable in timing, if not in size. Between windows, demand can be a tiny fraction of the peak.

That profile punishes both of the easy answers. Provision for the peak permanently and the budget is consumed by idle compute. Idle capacity is the norm, not the exception, in this class of infrastructure: analyses of production Kubernetes clusters consistently find average CPU utilisation in the low double digits, with most clusters significantly overprovisioned. Provision lean and the campaign fails in public, at the precise moment the most people are watching, with the media spend that drove them there already committed. There is no quiet failure mode. A national campaign that falls over trends on social media the same evening.

Kubernetes is the right tool for this problem, built as it is to scale workloads up and down, but it is not an automatic one. Autoscaling that responds to load that has already arrived is too slow for a spike measured in minutes; scaling policies, node provisioning and the platform’s own architecture all have to be designed around the campaign’s calendar rather than left to defaults.

What we did

The work spanned both sides of the problem, because on campaign infrastructure cost and reliability are one engineering problem, not two competing ones.

Cloudreason advised on the Kubernetes deployment underpinning the campaign: the cluster architecture and scaling behaviour that let the platform breathe with demand, expanding for the windows that justified it and contracting the moment they closed. Capacity was planned against the campaign calendar rather than discovered through incidents: scaled and rehearsed ahead of each window, held through the peak, and brought back down promptly afterwards, so the budget was spent on serving the audience rather than on insurance against it.

Alongside the engineering, we ensured the infrastructure ran within its agreed budget across load that varied by orders of magnitude, tracking spend through each campaign period against a plan the agency could stand behind with its client, with no end-of-campaign surprises.

The results

Three years of campaign windows, delivered within budget, through traffic peaks that varied by orders of magnitude, and an agency able to make infrastructure commitments to a major consumer brand with confidence, year after year.

Campaign work is the sharpest version of a general truth: cost efficiency and reliability are not opposing forces. They are the same discipline. It comes down to knowing exactly what the platform needs, when it needs it, and having the engineering control to provide precisely that.