Cutting cloud spend by 38% at a growing edtech platform

Client

A South African online education platform serving learners nationally. Named details are withheld at the client's request; the figures below are from the delivered engagement.

Business challenge

Monthly cloud spend had roughly tripled over eighteen months while user numbers grew far more slowly. Nobody could say which product areas were responsible, so every cost conversation ended in the same place: a general instruction to spend less, with no way to act on it.

Existing technical constraints

• No tagging standard, so spend could not be attributed to a team or feature
• Instances sized during an early load test and never revisited
• Development and staging environments running continuously, including overnight and at weekends
• Object storage with no lifecycle policy, holding several years of processed media
• Sharp seasonal peaks around term start, which made average-case sizing misleading

Proposed architecture

The work was sequenced so that attribution came first. Without it, every later change would have been unmeasurable.

1. A tagging standard covering environment, team and product area, enforced through policy rather than convention
2. Cost allocation reporting split by those dimensions, published to both engineering and finance
3. Rightsizing based on observed utilisation over a full seasonal cycle, not on peak-day headroom
4. Scheduled shutdown of non-production environments outside working hours
5. Storage lifecycle policies moving processed media to colder tiers on a defined schedule
6. Commitment purchases sized only after the above had settled, so the baseline was real

Implementation

Guardrails were introduced before optimisations, so savings could not quietly erode. Tag policy was applied at the account level, budget alerts were routed to the owning team rather than a shared inbox, and anomaly detection was configured against the new per-team baselines. Rightsizing was applied in waves, with a week of observation between each.

Security and governance

Cost tooling was given read-only access through a dedicated role. Environment scheduling was implemented so that production was structurally excluded from any automated shutdown, rather than relying on a naming convention. All changes were applied through Infrastructure as Code and reviewed.

Outcome

• 38% reduction in monthly cloud spend, sustained across the following quarter
• Spend attributable to a team and product area, so decisions had an owner
• No measurable change in application latency or availability
• Seasonal peak handled on the new sizing without manual intervention

Lessons and next phase

The single largest contributor was not waste but attribution: once teams could see their own costs, several optimisations were proposed by the teams themselves rather than imposed. The next phase covers commitment coverage monitoring and a per-feature cost view for the product group.