top of page

Cost Explorer tells you what happened. It doesn't tell you why.

  • nemo01292
  • Aug 10
  • 3 min read

Why we rebuilt AWS bill analysis as an AI problem instead of a spreadsheet problem.


AWS Bill Analysis is an internal tool we built at Jedi Hill to read a cloud bill the way an experienced FinOps engineer would. It runs against single accounts and against Organizations multi-account setups, detects cost spikes, generates root-cause analysis and optimisation recommendations, and outputs a finished visual HTML report. The intent behind it is narrow and specific: move cloud cost work from manual reconciliation to intelligent governance.


The three walls

Once cloud usage scales past a certain point, almost every organisation runs into the same three problems.


The bill becomes unreadable. Cost Explorer surfaces a dozen dimensions and several different cost measures, and cross-analysing spend by service, account, region and tag turns into manual Excel work that never quite keeps pace with the environment it is describing.


Anomalies go uncaught. Monthly cost spikes are usually noticed after the fact, and tracing one back to a root cause costs hours of digging through usage detail that nobody has budgeted time for.


Optimisation has no direction. Are Reserved Instances and Savings Plans actually being used well? Are tags compliant? Are service quotas approaching their limits? There is no single lens that answers those questions, so for the most part they go unanswered.

None of this is glamorous work. That is precisely what makes it a good place to put AI.


From reconciliation to governance

The shift here is not a nicer dashboard. It is automation plus an AI layer that reads the bill and then writes down what it found.


Instead of a person exporting CSVs and assembling pivot tables, the tool identifies the services that grew significantly, pulls the UsageType, Operation and Region detail sitting behind that growth, and hands that snapshot to a large language model. What comes back is a plain-language explanation of the root cause, a set of concrete optimisation recommendations, and a risk-monitoring plan. The output is a report a stakeholder can read, rather than a spreadsheet they have to interpret.


That is the distinction the whole tool is built around. Manual reconciliation tells you what happened last month. Intelligent governance tells you why it happened, and what to do next.


What it does

AI root-cause analysis. The tool flags services with significant growth, captures the UsageType, Operation and Region detail behind the increase, and uses a large language model to produce root-cause interpretation, optimisation recommendations, and a risk-monitoring plan. It works against any OpenAI-format API, hosted or self-deployed.

Multi-dimensional cost analytics. A three-month trend overview, Top-N analysis by service, account and region, and historical trend tracking, with support for multiple cost measures including Amortized, Net Amortized, Blended and Unblended.


Commitment governance. Dual-metric monitoring of utilisation and coverage for both Reserved Instances and Savings Plans, so waste and under-coverage show up in a single view instead of two separate investigations.


Tag governance. Tag-compliance analysis, identification of untagged spend, and tag-penetration analysis. This is the foundation for cost attribution that anyone will actually trust.


Anomaly and quota alerting. Account-level cost-spike detection, plus two-tier warning and critical threshold alerts on multi-region service quotas.


Zero-friction reporting. Fully automated HTML reports with visualisations, archived by account and by month. Nothing to assemble before someone can read it.


The use case nobody would demo

There is a pattern worth naming here. When organisations look for a first AI use case, they tend to reach for the customer-facing one — the chatbot, the assistant, the thing that shows well in a demo. The direct P&L impact is often sitting somewhere less visible, inside an internal cost centre.


The cloud bill is one of those places. Anyone can graph a spike; the value is in turning "spend went up" into "this UsageType, for this reason, here is what to do about it." And the leaks that matter most tend to be structural rather than dramatic — commitment coverage that drifted, tag hygiene that decayed — the kind of thing that drains money slowly and invisibly and needs a unified lens rather than another spreadsheet.

 
 
 

Comments


bottom of page