One analyst's query cost £4,000 in a single afternoon. Leadership wants a policy. What do you actually implement?

What the interviewer is testing

Whether you reach for structural controls rather than a policy document, and whether you know where analytics spend actually goes.

Why a policy alone fails

A policy asks people to remember something while doing exploratory work under time pressure. The control must be in the platform, because a single query can cost more than a month of normal operation and nothing currently stops it.

The controls

Byte-scan limits per role. A query estimated to scan more than a threshold is rejected with an error explaining why. Exploratory roles get a low limit, production pipelines a higher one. This single control turns a £4,000 accident into a message.

Query timeouts per workload class.

Aggressive auto-suspend on idle compute — minutes, not hours. A surprising share of analytics spend is warehouses idling.

Workload isolation. Separate compute for dashboards, transformations and exploration, so exploratory work cannot degrade executive reporting and each pool's cost is attributable.

Partitioning and clustering on the large tables, so ordinary queries prune. On scan-priced platforms this is the largest structural lever, and the £4,000 query probably scanned a table that should have pruned.

The organisational half

Cost attribution and showback — per team, per dashboard, per pipeline, published. Nothing reduces spend faster than showing a team that one dashboard's refresh costs more annually than the analyst who built it.

Also audit the schedule: a material share of transformation spend refreshes tables nobody has opened in six months, and the catalogue's usage data identifies them.

What a strong answer adds

Treating the analyst sympathetically. They ran a reasonable query against a table that was not laid out for it, and no feedback existed until the invoice. Showing estimated cost before execution changes behaviour without any policy at all.

Common weak answers

Training and guidelines. Removing analyst access, which destroys the platform's purpose.