Traffic arrives all at once, or not at all.
Media and entertainment systems spend most of their life idle and are judged entirely on the twenty minutes when tickets go on sale or an episode drops. Average load is a meaningless number here.
The whole system exists for one twenty-minute window.
An on-sale is the clearest engineering brief in any sector we work in: a year of preparation judged by whether the platform stays up while several times the normal annual traffic arrives in a quarter of an hour, all of it contending for the same scarce inventory.
That rules out several ordinary approaches. Optimistic locking on seats produces oversells; naive queueing produces abandonment; caching the wrong thing produces customers buying seats that have already gone.
It also makes the failure modes brutal in a way most sectors do not experience. There is no graceful degradation for an on-sale, either it worked or it is on social media.
We have built ticketing and box office platforms with event management, scanning and real-time reporting in production, which is the experience that informs how we approach the rest of this sector.
Six pieces of media work we do repeatedly.
Ticketing & box office
Inventory, on-sale, scanning and real-time reporting designed for the spike rather than the average.
Inventory · payments · scanningContent & rights management
Assets, versions, windows and territories held in one place, with rights enforced by the system, not by a spreadsheet.
MAM · rights · schedulingStreaming workflows
Ingest, transcode, package and deliver, with the operational tooling to see where a title is stuck.
Ingest · transcode · CDNMetadata & discovery
Enriching and normalising metadata, which decides whether anyone finds the content at all.
Catalogue · taxonomy · searchAudience analytics
Completion, drop-off and cohort reporting on a governed model instead of four conflicting extracts.
Playback events · warehouse · BIArchive intelligence
Making decades of unindexed material searchable through transcription and automated tagging.
Archive · transcription · taggingThree decisions that determine whether it holds.
Everything else about a ticketing platform can be fixed next week. These three cannot be fixed during the twenty minutes that matter.
Queue honestly, reserve properly
A visible queue with a realistic position beats a spinning page; customers tolerate waiting but not uncertainty. Behind it, inventory is held for a bounded window while payment completes and released reliably on abandonment. Getting the reservation wrong produces either oversells or phantom sell-outs.
Visible position Bounded holds Reliable releaseNever cache availability
Cache the page, the imagery and the description as aggressively as you like. The seat map is the one thing that must be live, and it is the thing most often cached by accident, which is how customers end up buying seats that have already gone.
Live inventory Cache everything else No stampedeDecide in advance what you shed
There is no graceful degradation for an on-sale unless you designed one. Choose beforehand what gets dropped under extreme load (recommendations, imagery, analytics) so the checkout path survives, and instrument the funnel live so you can see which step is failing while it happens.
Planned shedding Checkout protected Live funnelBefore your next on-sale.
Three questions we are asked by almost every media client.
Our platform fell over at the last on-sale. Where do we start?
With the funnel data from that event, not with an architecture review. Almost every on-sale failure has a single dominant bottleneck (a database contention point, a payment provider limit, a cache stampede on the seat map) and the traffic data usually identifies it within a day. Rebuilding broadly before you know which one it was is expensive and often misses it. Once the bottleneck is clear, the fix is frequently narrower than expected and can be proven with a load test before the next event.
Is AI useful here or is it a distraction?
Mostly it is useful away from the transactional path. Metadata enrichment, archive transcription and automated tagging deliver real value, particularly for organisations sitting on decades of unindexed material that is currently unfindable. Audience analytics and churn prediction are reasonable applications too. What we would keep AI out of is the on-sale path itself, that needs to be deterministic, fast and boring, and adding a model there buys nothing and introduces a variable you cannot afford.
Should we build ticketing or use a platform?
Use a platform unless ticketing is your business. The established ticketing platforms have absorbed the on-sale problem over many years and their fees are usually less than the cost of learning those lessons yourself. Building makes sense when you have unusual inventory rules, when the ticketing data is strategically yours to own, or when you are selling the platform to others, which is the case for several of the systems we have built. We will tell you which situation you are in before quoting.