Last quarter cannot price next year.
Every commitment decision is a statement about future consumption. Almost every commitment decision is made from past consumption. That substitution is defensible only when nothing is about to change — which is rarely true, and is knowable in advance when it is not.
Extrapolation is a forecast only when nothing is planned
A trend line answers the question 'what would happen if the last few months continued'. That is a legitimate question and almost never the one being asked. The real question is what the business will consume, and the business already holds information about that — migrations scheduled, products launching, systems being retired, headcount plans signed off. None of it appears in the billing history, and all of it is material.
Planned change is invisible to the data
A datacentre migration completing next quarter, a product launching in six months, a legacy platform being decommissioned — each moves consumption significantly, each is already decided somewhere in the organisation, and none of it is in the trend.
Efficiency work moves the base you are committing against
Right-sizing and modernisation programmes in flight will lower consumption. Sizing a commitment against pre-optimisation consumption locks in the waste the programme was funded to remove.
Growth and seasonality look identical in a short window
A rising line may be structural growth or the upslope of a seasonal cycle. Extrapolating the second as the first over-commits at exactly the point in the year the estate looks busiest.
The information exists — in departments that never meet the bill
Forecasting cloud consumption well is mostly an organisational problem rather than a statistical one. The facts that determine next year's spend are held by people who are not in the conversation when the commitment is purchased.
- 01
Engineering roadmaps state what will be built and retired, but rarely in units that translate to consumption without deliberate work.
- 02
Migration programmes have schedules that slip, so a forecast must model timing risk rather than assume a date that will move.
- 03
Architectural direction changes eligibility as well as volume: moving to serverless or managed services alters which commitment instruments can cover the resulting spend at all.
- 04
Headcount and business plans drive consumption in ways nobody has mapped, because the mapping only matters once a year.
- 05
The organisation has no shared definition of a base case, so the optimistic plan and the conservative plan are both quoted as 'the forecast' depending on who is asked.
- 06
Forecast error is asymmetric — over-committing and under-committing cost different amounts — and the asymmetry is almost never quantified before the decision.
What goes into a requirements model
A requirements model starts from consumption history and then adds what the organisation already knows but has never written down in consumption terms.
- 01Consumption history decomposed into structural growth, seasonality and one-off events
- 02Growth rate by service and workload rather than a single blended estate figure
- 03Engineering roadmap items with a consumption consequence, sized and dated
- 04Migration programmes with schedule, expected volume and timing-risk range
- 05Decommission plans and the consumption they will remove
- 06In-flight efficiency work and its expected effect on the eligible base
- 07Architectural direction and its effect on commitment eligibility, not only on volume
- 08Business plans with cloud consequences: headcount, customer growth, market entry
- 09Seasonality amplitude measured across full cycles rather than inferred from a partial one
- 10Historical forecast accuracy, where any prior forecast exists to check against
- 11Downside scenarios — plans that do not land, programmes that slip a quarter
- 12The cost asymmetry between over- and under-committing at each candidate level
What we determine
A base case with its assumptions stated
Not a single number but a stated position: what is assumed, which of those assumptions the forecast is most sensitive to, and what happens to the commitment decision if each one is wrong.
The confident floor
The consumption level that survives the downside case — plans slipping, programmes not landing, growth disappointing. This is the level that can be committed at long term without risking a stranded position.
The uncertain band
Consumption that is likely but not assured, quantified separately, so it can be handled with flexible instruments or left on demand rather than being averaged into a single commitment figure that hides the risk.
The asymmetry of being wrong
What over-committing costs versus what under-committing costs at each candidate level. In most estates these are not symmetric, and knowing which direction is cheaper to be wrong in changes the decision.
A forecast is not a prediction to be graded later. It is a decision input, and its job is to tell you how much of your consumption you can responsibly lock, for how long, and what it costs you if the world does not cooperate.
- How much consumption can be committed long term without downside risk
- How much belongs in flexible instruments because it is likely but not assured
- Whether to commit now or wait for a migration or efficiency programme to land
- Which planned events the commitment decision is most sensitive to, so they get tracked
- How to phase purchases across the year rather than making one annual decision
- What would have to change for the commitment position to need revisiting mid-term
What we hand over
Requirements model
Base, upside and downside consumption cases with every assumption stated and attributed, so the model can be maintained and challenged rather than treated as an output nobody can inspect.
Sensitivity analysis
Which assumptions actually move the commitment decision, ranked — so the organisation tracks the three things that matter instead of monitoring everything equally.
Decision calendar
The dates by which each commitment decision must be made, mapped against the business events that should inform them, so decisions are not forced by an expiry nobody was watching.
Ask before you commit
Stage 00 is a 30-minute qualifying call at no cost. If the timing or the estate does not justify an engagement, we say so on that call.