The method, domain by domain
The nine phases of the methodology apply everywhere. What changes from one domain to another: the data, the read rights, the rules and the limits.
The whole invoice, not only virtual machines
Read-only access: Reader and Cost Management Reader on Azure, ReadOnlyAccess and CUR access on AWS, Viewer and Billing Viewer on Google Cloud. Seven families of sources: billing, inventory, metrics, native recommendations, backup, logs, prices.
The invoice is read in amortized and billed cost, then normalized to the FOCUS 1.2 format. Levers are quantified in the order they are executed:
- Buy less: orphaned and idle resources
- Consume better: double-evidence right-sizing, non-production Start & Stop, disk tiers, backup, generation migrations
- Pay less: reservations and Savings Plans computed after reduction, Azure Hybrid Benefit
- Beyond virtual machines: levers on managed services (PaaS)
- Production is never resized without your agreement
Tooling, without overpromising. Azure: complete chain, replayed on an anonymized pilot engagement (22 key totals out of 22 within a 1% tolerance). Google Cloud: collection through Cloud Shell from your BigQuery billing export. AWS: CUR 2.0, chain to be recalibrated on a first real export.
Contractual optimizations (EDP, MACC, SLA credits): see procurement, vendors and contracts.
Two days between access and the first figures
Scope
Scope, business rules, applications not to touch, read-only role.
Collect and compute
Seven families of sources read in read-only mode; the engine computes in twenty minutes.
Prioritize
Review of the findings with the consultant: effort, risk, prerequisites, horizon.
Read out
Summary to management, options to decide, dated execution plan.
From actual usage to signature, then to the invoice
- Contract register: for each contract, the vendor, the subject, the end date, the notice period, indexation, the consumption commitment, the entitlements and the exit window; kept in a workbook, reviewed at every committee.
- Renewal in four stages: inventory of actual usage, scenarios quantified at the same level (renew, reduce, switch, exit), negotiation position, then a check after signature.
- Invoiced price checks: line by line, SKU by SKU, against the contractual price grid; every gap is quantified and documented for the vendor.
- Vendor committees: consumption against commitment, billing incidents, upcoming renewal dates, decisions recorded; arbitration between marketplace and direct purchase.
Tooling. The schedule and the scenarios of cloud commitments (reservations, Savings Plans, CUD) are computed by our app at the observed discount. The contract register, tracking of consumption against commitments and invoiced price checks are kept in a workbook.
Scope. Cloud providers and software vendors. Out of scope: hardware, telecoms, service contracts.
Variable fee. As an option and in pilot only, on some negotiations, with a baseline written at scoping.
What KeeFinops does, what your teams do
| Topic | KeeFinops | Your teams |
|---|---|---|
| Inventory | Structures the register and matches contracts, invoices and usage | Provide contracts, amendments and invoices |
| Position | Quantifies the scenarios and prepares the negotiation file | Procurement and IT decide and negotiate |
| Signature | Reviews the economic terms before signature | Sign; your lawyer reviews the legal terms |
| Check | Checks invoiced prices and keeps the register up to date | Pass the gaps on to the vendor |
From entitlements to real usage
- Inventory of contracts, entitlements and applications
- Real usage compared with entitlements: dormant seats, reallocation rather than deletion
- Renewal calendar and negotiation positions
- Price alignment across entities and license clean-up
- Licenses in the cloud: license included or BYOL, Azure Hybrid Benefit, SQL Server editions
From inventory to decision, before the due date
Scope
Vendors, entities, due dates and available data; the duration is set here.
Collect
Contracts, entitlements, vendor usage exports, monthly SPLA reports.
Reconcile
Real usage against entitlements, reported against deployed, gaps quantified by vendor and by entity.
Decide and track
Rationalization plan, negotiation file, then tracking of renewals or reports.
What KeeFinops does, what your teams do
| Topic | KeeFinops | Your teams |
|---|---|---|
| Inventory | Structures the register of contracts and entitlements | Provide contracts, invoices and vendor exports |
| Reconciliation | Compares real usage and entitlements, reported and deployed | Confirm special cases: service accounts, test environments |
| Decision | Quantifies the options and prepares the negotiation | Procurement and IT decide and negotiate |
| Tracking | Keeps the calendar of renewals or reports | Apply the reallocation and report |
Report right, rebill right
Hosting providers, MSPs and IT services firms that report Microsoft licenses under SPLA every month and rebill them to their clients: since 1 October 2025, SPLA licenses can no longer be used on Listed Providers (Azure, AWS, Google Cloud, Alibaba), under the rule Microsoft announced to its partners in 2022.
The monthly report is due within ten days after month end (SPLA agreement, 2020 version) and, according to licensing practitioners, an audit can presume unreported usage from the start of the relationship with the end client.
- Reconciliation report between reported and deployed
- Audit risks and remediation plan
- Recommended path for each workload
- Monthly reporting procedure
- Rebilling model per end client
Meter, attribute, connect to outcomes
The FinOps Foundation defines token economics as the discipline through which AI consumption is metered, attributed and connected to business outcomes (FinOps Foundation).
We aim for that frame: cost per inference, per workflow and per outcome, and return-on-investment thresholds set before any new project. In the field, we already track cost per active user and per thousand tokens.
Baseline first, scale later
Baseline
Five consumption modes, activated seats against seats bought, cost by team and by application.
Instrument
LiteLLM or AI gateway: keys per use, budgets, spend logs, labels.
Set the rules
Model, token and cache policy; PTU or pay-as-you-go trade-off; quotas and alerts.
Track and allocate
Showback by entity and by team, KPIs per active user and per thousand tokens.
What you receive
Baseline
By consumption mode, by team and by application.
Dormant seats
The list, and a reallocation plan rather than deletion.
Token comparison
Cost per million tokens, by model and by channel (Azure, Bedrock, Vertex), discounts and commitments included.
PTU or pay-as-you-go
A quantified recommendation on your real utilization rate, with quotas and alerts.
Usage policy and gateway
Model, tokens, cache, routing; configuration of keys, budgets and labels.
Dashboard
Cost per active user and per thousand tokens; rebilling by client will come with the planned product module.
What KeeFinops does, what your teams do
| Topic | KeeFinops | Your teams |
|---|---|---|
| Baseline | Reads cost and usage data, never the content of conversations | Run the extraction on your side with your admin key, or provide the console exports; grant read-only billing roles |
| Instrumentation | Proposes the gateway configuration | Deploy and run the gateway |
| Rules | Quantifies the options: models, PTU, cache | Data and AI teams decide |
| Tracking | Computes the KPIs and the showback | Own the budgets by team |
Measure the footprint alongside the costs
- Scoping: read access to emissions is requested from the start: the Carbon Optimization Reader role on Azure; Carbon Footprint Viewer and read access to the export dataset on Google Cloud; read access to the emissions export of Data Exports on AWS. Definitions are written down and validated with your Green IT lead.
- Collection: the month's emissions are read after the provider publishes them, around the middle of the following month; steering indicators follow the cost collection.
- Execution: levers shared by cost and footprint move forward together: shutdown of non-production outside business hours, right-sizing, unused resources, storage, older generations, pausing data platforms outside production, and region only if your policy allows it.
- Verification and handover: monthly measurement on a fixed date, variances commented, handed over to your sustainability team, which remains the owner of the publication method.
Two limits, stated plainly. Changing region can cost more. Commitments (reservations, savings plans, CUD) lower the invoice, not the emissions.
Level of evidence. Green IT indicators: practiced and tooled. Emissions published by the provider, attached to costs: practiced in the field; import into our app: planned.
What is measured, and how often
| Indicator | Definition | Frequency | Level |
|---|---|---|---|
| Share of VMs correctly sized | A VM is only flagged as to be resized when three conditions are met: provider recommendation, metrics below a cautious threshold over the window, half size available. VMs that cannot be measured are shown separately. | Monthly | Tooled |
| Share of non-production switched off | Nights, weekends and public holidays; production is never concerned; exclusions by written rule. | Weekly | Tooled |
| Counters | VMs measured, not measurable, switched off all week. | Weekly | Tooled |
| Emissions of the month | Value published by the provider, by entity, application, environment and region, with its month, its method and its model version. | Monthly | Practiced in the field; in the app: planned |
| Spend intensity | Emissions per thousand euros and per VM, shown next to the euros, never added up. | Monthly | Practiced in the field; in the app: planned |
| Trend | Over 3 and 12 months, gap to a baseline. | Monthly | Planned |
Six rules that are not negotiable
The provider's value is the reference
It is quoted with its month, its method and its model version.
Never mixed
Measured and estimated values are never added up, nor two scope 2 calculation methods.
Monthly, never daily
Carbon indicators follow the provider's publication.
The cost dimensions
An emission allocated by cost, for lack of a fine enough grain, is flagged as allocated.
Cautious thresholds
Right-sizing requires a double confirmation; VMs that cannot be measured are never counted.
No avoided emission announced in advance
Not before the provider's measurement has confirmed it.
Frequently asked questions
What are the prerequisites by cloud?
Azure: billing history and 30 days of metrics can be read straight away with a Reader role, hence first figures on Day 2. Google Cloud: the Cloud Shell collection relies on your BigQuery billing export. AWS: on the CUR 2.0 export. These exports contain nothing from before their activation: if they are not already active, quantification waits until they cover 30 days.
Do we have to install anything?
No agent on your machines. The script pasted in the portal console needs no installation. The scheduled mode, however, deploys a job in your subscription (about €10 a month); a workstation can also collect with a Reader account.
Do we have to fix tagging first?
No. We quantify from the first collection; an environment category inferred from naming is flagged as an assumption, and tagging quality is measured in parallel.
Why two formats?
One scope is quantified in two days. A group in which every entity must validate its findings needs about six weeks: the time is not in the calculation, it is in the validation.
Why reallocate rather than delete?
Because a dormant seat here often answers a request elsewhere: reallocating it avoids a new purchase.
Do we need a specific tool?
Not to start: the baseline starts from the providers' cost APIs (for example the Anthropic Usage and Cost API, which requires an admin key: extraction then runs on your side), from Azure, Bedrock and Vertex billing and from seat administration consoles. The gateway comes next, for instrumentation.
Token prices are falling. Will the topic go away?
The unit price falls, but volumes and uses multiply: what needs tracking is the cost per workflow and per outcome, not only the token price.
Do you prepare vendor audits?
In pilot: we reconcile entitlements, deployments and reports, quantify the gaps and prepare the remediation plan before the auditors arrive.
Let's see how this methodology applies to your scope
Thirty minutes to choose the entry point, the format and the calendar.