Why the cloud bill escapes traditional budget control
FinOps was born of a simple observation: the cloud has changed who decides on spending. In a data centre, buying servers goes through a request, a quote and an approval. In the cloud, an engineer creates a resource in a few moments, and the cost runs from that point. The purchasing decision has dissolved into thousands of everyday technical actions.
The usual control tools are not suited to this. The annual budget discovers the variance after the fact. The invoice contains thousands of lines expressed in technical units that finance cannot link to a product. Teams, for their part, often do not know what their choices cost: they do not see the invoice.
The result is not always excessive spending. It is unexplained spending, which is more troublesome: no one can say whether the month’s increase reflects growth in activity or a forgotten test environment.
Visibility, allocation, optimisation: the three FinOps workstreams
Visibility means knowing what is spent, on what and for whom. It relies on the provider’s billing exports, on a readable account structure and on tags applied systematically. An optional tag is rarely applied: its enforcement must be automated at deployment time.
Allocation links each cost to an owner: a product, a team, an internal customer. The difficulty lies in shared costs (network, common platform, support, security), which must be distributed using a simple, accepted key. A perfect key that no one understands is worth less than an approximate key that everyone accepts.
Optimisation comes last, and that is a logical order. You can only usefully optimise what you know how to allocate, and it is the owners who know whether a resource can be reduced.
For the vocabulary and reference framework of the discipline, see our glossary entry on the definition of FinOps.
Usage, architecture, pricing: three levers not to be confused
| Lever | Examples | Who acts | Main risk |
|---|---|---|---|
| Usage | Rightsize, shut down outside hours of use, remove orphaned resources, move cold data to cheaper storage | The owning teams | Degrading performance by reducing without measuring |
| Architecture | Autoscaling, choosing pay-per-use services, reducing data transfers between zones | Architects and product teams | A development effort to be weighed against the benefit |
| Pricing | Consumption commitments, interruptible capacity for tolerant workloads, contract negotiation | FinOps team, procurement, finance | Committing to usage that is set to fall |
The order matters. A pricing commitment calculated on unoptimised consumption locks in the waste for its entire term. We therefore address usage first, then pricing on the share of consumption that has become stable.
Some items are regularly overlooked because they do not look like servers: outbound data transfers, log retention, backups, licences. Analytics platforms and artificial intelligence workloads deserve particular attention, because their cost often depends on the queries executed rather than on reserved resources. This point is addressed with the teams concerned in our data platform offering.
Our approach: making cost visible to those who incur it
Our consultants work within your teams; decisions remain yours.
Establishing a first accurate picture. Starting from the billing data, we reconstruct spending by product and measure the share that cannot be allocated to anyone. This unallocated share is the first indicator to bring down.
Showback before chargeback. Each team receives a view of its costs, in its own terms. Showback alone often reveals forgotten resources that their owners remove of their own accord.
Introducing a unit cost. The total bill says nothing about efficiency. The cost per transaction, per case processed or per active user, on the other hand, can be compared from one month to the next. A rising bill with a falling unit cost is good news.
Establishing routines. A periodic review brings technical, product and finance together around variances and the decisions to be made. Anomaly alerts flag a drift within a few days instead of waiting for the month-end close.
Building cost into design. The cost estimate becomes part of the architecture file, and tagging is built in by default to the golden paths of the platform described in our DevOps and platform engineering offering.
FinOps and sobriety: where they converge, where they diverge
Cost and environmental footprint often move in the same direction. An oversized machine, an environment running at night with no users, data kept for no reason: eliminating them lightens both the bill and resource consumption.
The convergence is not complete. A pricing commitment reduces the price, not the usage. The cheapest region is not necessarily powered by the lowest-carbon electricity. Very cheap storage encourages keeping everything. Conversely, the footprint is not limited to the electricity consumed: manufacturing the equipment accounts for a share of it, which the bill does not reflect.
We therefore recommend tracking footprint indicators alongside cost indicators, bearing in mind that the data supplied by hosting providers rests on calculation methods that differ from one to another. Comparison over time, with the same provider, is more reliable than comparison between providers.
False economies
- The one-off campaign. A reduction plan carried out once produces an effect, then spending climbs back up. FinOps is a continuous practice.
- The central team optimising alone. Without knowledge of the applications, it reduces what should not have been reduced, or dares not touch anything.
- Tagging by decree. A tagging rule without automated enforcement is only followed for a few weeks.
- The bill as sole judge. Penalising a team whose spending grows with its activity discourages precisely what the business is seeking.
- Migration without follow-up. The estimates made before a cloud migration must be compared with actual costs as soon as the first applications have been moved.
The indicators of an established FinOps practice
- the share of spending allocated to an identified owner;
- the share of resources correctly tagged;
- the gap between forecast and actual spending;
- the coverage rate of stable consumption by commitments, and the utilisation rate of those commitments;
- the trend in unit cost of the main products;
- the time between a cost anomaly appearing and it being dealt with;
- the share of optimisation recommendations actually acted on by teams.
None of these indicators is a savings figure. Savings are a consequence. What we seek to establish is the organisation’s ability to explain its spending and to decide on it.