Sector
Retail & marketplaces
Catalogue, inventory and checkout under traffic that arrives in spikes — where availability during the spike is the entire commercial case.
The detail
Retail load is not a curve, it is a cliff. The system spends most of the year comfortable and then has a handful of days that decide the commercial year, which means capacity planning, cache strategy and graceful degradation matter more than steady-state efficiency.
We design for the cliff. Read paths served from materialised views that survive the write path failing, inventory that degrades to an honest "low stock" rather than overselling, and checkout isolated hard enough that a recommendations outage cannot take payment down with it. Load tests are run against the real topology, not a scaled-down copy that proves nothing.
Search and catalogue get treated as their own product. Relevance is measured, indexing is incremental, and the merchandising team gets levers they can pull without a deploy.
Typical engagements: checkout reliability, catalogue and search re-platforming, inventory consistency across channels, and readiness work in the quarter before a peak trading period.