Cloud providers offer substantial discounts for committing to spend or capacity over one or three years. The discounts are genuine and the arithmetic is compelling. The risk is that you are being asked to forecast, and most organisations are bad at forecasting their own consumption.
What you are actually buying
Broadly two shapes. Commit to a quantity of specific resources, which gives the deepest discount and the least flexibility. Or commit to a level of spend, which discounts less and lets you move between services.
The deeper discount always carries more specificity, and specificity is what turns into waste when your plans change.
The failure modes
Committing to the wrong shape. Three years on a specific instance family, then the application is rearchitected, or a new generation arrives that is better value, and you are holding a commitment to yesterday’s hardware.
Committing to a migration that does not happen on schedule. The commitment starts billing on day one. The workloads arrive in month nine. You have paid for nine months of nothing, which frequently exceeds the discount.
This is the most common and most expensive version, and it happens because the commitment is negotiated by people optimistic about a migration timeline.
Committing at the peak. Sizing the commitment against peak consumption rather than the floor. The whole point is to cover the baseline that definitely exists.
Nobody owning it. The commitment is made, the person moves on, and nothing is tracked against it until renewal.
The approach that works
Commit to the floor, not the forecast. Look at the last twelve months and find the level below which consumption never dropped. Commit to that, at the longest term. It is the part you are certain about.
Layer the rest. Above the floor, use shorter commitments or none. The incremental discount on the uncertain portion is rarely worth the lock-in.
Stage commitments to match migration reality, not the plan. If a migration is meant to complete over a year, do not commit to the end state on day one. Add commitment as workloads actually land.
Track it monthly. Actual consumption against commitment, with a name attached. This single discipline prevents most of the waste, and it takes fifteen minutes a month.
The question to ask before signing
How much of this commitment is covered by workloads that are running today, in this account, right now?
If the answer is most of it, the commitment is low risk and you should take the deepest discount available.
If a significant portion depends on things that have not migrated yet, you are not buying a discount. You are underwriting a project plan, and project plans slip. Size the commitment to what exists and revisit in six months.
The negotiation point people miss
These terms are more negotiable than the published pages suggest, particularly at meaningful scale and particularly around flexibility: the ability to change shape mid-term, to apply commitment across accounts, or to pause during a specific event.
Ask for flexibility rather than for a bigger discount. A slightly smaller discount on a commitment you can adjust is worth considerably more than a deep discount on one you cannot, and the second is what gets signed when the conversation is only about percentage.