Commitment discounts: you sell flexibility
How much should you commit, and why does a multi-year term save less than its headline?
A commitment discount runs the trade in the opposite direction. You sell flexibility. The vendor buys demand certainty, because certainty lets them provision against a smaller variance. They monetize it back to you as a discount of 30–60%.
Sizing from the shape of the loss
The correct commitment level is derivable from the asymmetry of the loss.
Measure both sides in the same unit, the list price of one unit. With a discount d (say 40%), a committed unit costs 1 − d = 0.6 whether you use it or not.
- Under-committing: each unit of demand above the commitment is bought at list instead of at 0.6, so it costs the discount, d = 0.4.
- Over-committing: each committed unit nobody uses is paid for in full at the committed price, 1 − d = 0.6.
The cheapest commitment is the level c where one more committed unit stops paying: where the chance that demand falls below c equals 0.4 ÷ (0.4 + 0.6). That ratio is just d. So commit at the demand quantile equal to the discount: the P40 of your demand forecast for a 40% discount. That is below the mean, but not far below, and a larger discount moves it up. Worked example: demand forecast mean = 60 units, peak = 100 units, P10 = 40 units. With a 40% discount the cheapest commitment sits near the P40, about 56 units (the exact number depends on the forecast's shape). Almost everyone commits at the mean (60) or at last quarter's peak (100), and the waste is invisible because it appears as a flat line and not as an anomaly. That is exactly Priya's situation.
A multi-year commitment is a bet against price deflation
A multi-year commitment is a bet against price deflation, and the bet is usually not stated.
In year 3 the discount has nearly disappeared. The saving over the term is (0.40 + 0.20 + 0.04) ÷ (1.00 + 0.80 + 0.64) = 0.64 ÷ 2.44 ≈ 26%, against a headline of 40%. That is about two-thirds of the headline, and all of the benefit comes from the early years. One-year terms at smaller discounts often beat three-year terms at larger ones for this reason alone, which is the opposite of how they are marketed.
Data table
| List prices fall per year | Real saving over the term |
|---|---|
| 0 | 40% |
| 0.004 | 39.8% |
| 0.008 | 39.5% |
| 0.011 | 39.3% |
| 0.015 | 39.1% |
| 0.019 | 38.9% |
| 0.023 | 38.6% |
| 0.026 | 38.4% |
| 0.03 | 38.2% |
| 0.034 | 37.9% |
| 0.038 | 37.7% |
| 0.041 | 37.5% |
| 0.045 | 37.2% |
| 0.049 | 37% |
| 0.053 | 36.7% |
| 0.056 | 36.5% |
| 0.06 | 36.3% |
| 0.064 | 36% |
| 0.068 | 35.8% |
| 0.071 | 35.5% |
| 0.075 | 35.3% |
| 0.079 | 35% |
| 0.083 | 34.8% |
| 0.086 | 34.5% |
| 0.09 | 34.3% |
| 0.094 | 34% |
| 0.098 | 33.8% |
| 0.101 | 33.5% |
| 0.105 | 33.2% |
| 0.109 | 33% |
| 0.113 | 32.7% |
| 0.116 | 32.5% |
| 0.12 | 32.2% |
| 0.124 | 31.9% |
| 0.128 | 31.7% |
| 0.131 | 31.4% |
| 0.135 | 31.1% |
| 0.139 | 30.8% |
| 0.143 | 30.6% |
| 0.146 | 30.3% |
| 0.15 | 30% |
| 0.154 | 29.8% |
| 0.158 | 29.5% |
| 0.161 | 29.2% |
| 0.165 | 28.9% |
| 0.169 | 28.6% |
| 0.172 | 28.4% |
| 0.176 | 28.1% |
| 0.18 | 27.8% |
| 0.184 | 27.5% |
| 0.188 | 27.2% |
| 0.191 | 26.9% |
| 0.195 | 26.6% |
| 0.199 | 26.3% |
| 0.202 | 26% |
| 0.206 | 25.7% |
| 0.21 | 25.4% |
| 0.214 | 25.1% |
| 0.217 | 24.8% |
| 0.221 | 24.5% |
| 0.225 | 24.2% |
| 0.229 | 23.9% |
| 0.232 | 23.6% |
| 0.236 | 23.3% |
| 0.24 | 23% |
| 0.244 | 22.7% |
| 0.247 | 22.4% |
| 0.251 | 22.1% |
| 0.255 | 21.7% |
| 0.259 | 21.4% |
| 0.263 | 21.1% |
| 0.266 | 20.8% |
| 0.27 | 20.5% |
| 0.274 | 20.1% |
| 0.277 | 19.8% |
| 0.281 | 19.5% |
| 0.285 | 19.1% |
| 0.289 | 18.8% |
| 0.292 | 18.5% |
| 0.296 | 18.1% |
| 0.3 | 17.8% |
Two things change the calculation. Convertible commitments (exchangeable between machine types) and marketplaces that resell unused commitments lower the cost of over-committing, and so move the right level up. And every commitment consumes a demand forecast whose error bars nobody states and which, in most organizations, nobody owns.