Commitments & Discounts

## The bargain a commitment makes Every large provider sells the same bargain under a different name: you promise something for a fixed **term**, and in exchange you are charged less per unit than the published, undiscounted rate. The provider gets predictable demand and predictable revenue; you get the lower rate and carry the risk that the promise outlives the need for it. The direction of that risk is worth stating plainly, because it is the opposite of the other cheap rate on a platform. A **commitment discount** is cheaper because *you* accepted the risk of not using what you promised. Capacity that the provider may reclaim at short notice is cheaper because the *provider* kept the right to take it back. They fail in opposite directions, and treating a commitment as something that can simply be handed back is the first confusion to avoid. Commitment instruments divide into two families by what the promise is made of. ## A usage commitment promises a quantity A usage commitment fixes a **quantity per unit of time of a particular resource shape** — units per hour of a given machine family and size class, a level of provisioned throughput, a quantity of stored data. Everything else about it follows from that: - **Only matching usage draws the discount.** Usage of a different shape is billed at the undiscounted rate, even when it is the same workload doing the same job. - **The promise is charged whether or not it is consumed**, at whatever granularity the instrument is measured at, usually the hour. An hour below the promised quantity is not refunded, and on hourly instruments it generally does not roll forward into a busier hour. - **It carries the deeper of the two rates**, because the buyer removed more uncertainty: not only how much money will be spent, but exactly what will run. - **It strands on a shape change.** A redesign onto a different family, a different size class or a managed tier leaves the commitment matching nothing while it keeps being billed. ## A spend commitment promises money A spend commitment fixes an **amount of money per unit of time**, and the provider applies it against whatever usage the programme says qualifies, commonly drawing it down against the highest-rate eligible usage first. Its properties are the mirror image: - **The shape may change and the discount follows**, as long as the new usage is still inside the qualifying set. - **The qualifying set still has edges.** It is not everything on the bill: some services sit outside it, and usage outside it pays the undiscounted rate however large the commitment is. - **It carries the shallower rate**, and the gap between the two rates is precisely what the flexibility cost. - **The floor is still a floor.** An hour in which qualifying usage falls below the committed amount is charged up to the commitment anyway. ## Side by side | | Usage commitment | Spend commitment | |---|---|---| | What is promised | A quantity per hour of a named shape | An amount of money per hour | | What draws the discount | Only usage matching that shape | Any usage inside the qualifying set | | An hour below the promise | Billed up to the promise | Billed up to the promise | | Usage above the promise | Billed at the undiscounted rate | Billed at the undiscounted rate | | After a shape change | Stops matching, keeps billing | Keeps applying if still qualifying | | Relative depth | Deeper | Shallower | ## Choosing on a flat baseline A year of flat spend is an argument for committing *something*. It is not an argument for committing all of it at the deepest rate. A workable order: 1. **Measure the floor, not the average.** Take the level the workload never drops below across a representative period, hour by hour, rather than the monthly mean. 2. **Ask how stable the shape is, separately from how stable the amount is.** Flat spend on a design that is about to be rewritten is steady money on an unsteady shape. 3. **Commit the durable part narrowly and the uncertain part flexibly**, or leave the uncertain part undiscounted. Holding a mix of instruments is normal; one large purchase is not the only option on offer. 4. **Re-measure once the term is running.** The purchase is judged afterwards by how much of it was consumed and how much eligible usage it covered, not by the headline rate. ## What candidates get wrong - Treating the discount as a **rebate applied to the invoice** rather than a different rate on matching usage. - Believing a commitment **caps** spend. It sets a floor you will pay; usage above it bills normally. - Assuming an unused hour **rolls forward** into a busier one. - Assuming the purchase can simply be **cancelled**. Most cannot; some platforms allow an exchange into another instrument or a resale and others allow neither, so providers differ — but none of them make the decision free.