EDP Spend Forecasting Methods for Defensible Commitments
An EDP commitment is only as good as the forecast underneath it. The forecasting method buyers use to size the commitment is the single largest determinant of whether the EDP saves money or creates structural overhang. A practitioner-grade view of the methods that actually work.
The Enterprise Discount Program is a multi-year, multi-million-dollar financial commitment that locks the buyer into a fixed dollar floor of AWS consumption. The discount benefit is real — typical EDP tiers deliver 10-30% off list — but the value of that discount is conditional on the buyer actually consuming the committed amount. Over-commit by 20%, and the discount is wiped out by the burn-down obligation on unconsumed spend.
Across 500+ engagements at $2.4B+ in AWS spend reviewed, the forecasting work that precedes EDP signature is consistently the highest-leverage activity in the entire negotiation. The discount tier matters; the commitment level matters more. And the commitment level is entirely a forecasting output.
Why most EDP forecasts are wrong
The typical forecast that lands on the table during EDP negotiation comes from one of three places: the AWS account team's projection (almost always aggressive, because their incentive is commitment dollars), the buyer's Finance team's straight-line extrapolation of prior 12 months (usually too simplistic to capture workload shifts), or the engineering team's ground-up build (usually optimistic on growth and pessimistic on optimization). Each of these in isolation is wrong in a known direction.
A defensible EDP forecast triangulates between three methods, reconciles the differences, and produces a single forecast with a documented variance band. The commitment is then sized to the conservative end of that band, not the midpoint and certainly not the optimistic end.
Method 1: Bottom-up workload forecast
The bottom-up method starts from the application portfolio. For each material workload, the team forecasts the AWS services consumed, the unit volume (instance-hours, GB-months, requests), and the unit price after expected RI/SP coverage. Sum across workloads, add a tax/data-transfer/ineligible portion, and you have a workload-driven forecast.
This method is most accurate when the workload portfolio is stable. It struggles when:
- New workloads are planned but not yet running. The forecast inherits the engineering team's optimism about launch dates and consumption profiles, both of which routinely slip.
- Decommissioning is anticipated but not committed. The retirement of a legacy workload is forecast to free $1M of spend; in practice the retirement slips a year and the spend persists.
- The workload model misses cross-cutting services. CloudWatch, KMS, networking, and other shared services are systematically under-counted in workload-driven forecasts.
Bottom-up forecasts should be discounted 5-15% to account for these known biases.
Method 2: Top-down historical trend
The top-down method takes 24-36 months of AWS billing history, computes the monthly run-rate, applies a growth rate, and projects forward. For mature AWS estates with stable workloads, this method is highly accurate over 12-24 month horizons.
The top-down method fails when:
- The business is in a step-change. Acquisitions, divestitures, large new product launches, or major architectural shifts break the historical trend.
- The historical period includes one-time costs. Migration spend, data-transfer surges from one-off events, or holiday peaks that won't recur.
- The growth rate is itself unstable. A trailing-12-month growth rate of 40% may reflect a launch year that has now matured to 15%.
Top-down forecasts should be decomposed by service category before extrapolation. Storage often grows at one rate, compute at another, AI/ML at a third. A single blended growth rate masks the underlying dynamics and produces forecasts that are wrong in compensating ways.
Method 3: Bottoms-up unit economics
The third method, used by sophisticated SaaS and consumer-internet buyers, is unit-economic forecasting. Spend is modeled as a function of business volume — monthly active users, transactions processed, GB of customer data, hours of compute per customer. Forecast the business growth, apply the unit economics, and the AWS spend follows.
This method is the most powerful when the unit economics are stable and the business forecast is reliable. It catches efficiency improvements (cost per user falling as scale grows) and inefficiency surprises (cost per user rising due to architectural decisions). It pairs naturally with the FinOps practices that the most-mature buyers already operate.
For an EDP commitment, the unit-economic forecast should be run alongside one of the other two methods as a sanity check. If the unit-economic forecast diverges from the top-down trend by more than 10%, one of them is wrong and the divergence should be diagnosed before committing.
Triangulating the three forecasts
The practical reconciliation:
- Run all three forecasts independently for the EDP term (typically 3 years, sometimes 5).
- Compare year-by-year, not just term-total. The methods often disagree most in years 2-3 where uncertainty compounds.
- Diagnose the largest divergences. If bottom-up says $50M year 2 and top-down says $42M year 2, what specifically accounts for the $8M? Usually one or two workloads or services drive the gap.
- Produce a reconciled forecast — sometimes the average, sometimes the conservative case, depending on which method's assumptions are most defensible for this buyer.
- Compute an explicit variance band. The reconciled forecast is the midpoint; the band is +/- 15-25% depending on business uncertainty.
The EDP commitment is then sized at the lower end of that band — typically 70-85% of the midpoint forecast. The buyer captures EDP discount on all eligible spend; the unconsumed portion of the commitment is essentially a known liability, sized to be small.
Across 500+ engagements, buyers who arrive at EDP negotiations with reconciled, three-method forecasts achieve materially better outcomes than buyers relying on AWS-provided projections or single-method internal forecasts. The reconciliation work also surfaces hidden assumptions — about retirement schedules, growth rates, and unit economics — that are valuable independent of the EDP.
What to do about the ineligible portion
EDP forecasts must exclude EDP-ineligible spend. Marketplace beyond the allocated credit, Enterprise Support fees, ProServ, certain Direct Connect partner fees, and tax do not count toward commitment consumption. See EDP Eligible Service List 2026 for the full breakdown.
The practical sequence: forecast the gross AWS bill using one of the methods above, then subtract the projected ineligible portion to derive the EDP-eligible forecast. Commit to a percentage of the eligible forecast, not the gross. Buyers who commit to a percentage of gross consistently over-commit by 5-15%.
Growth assumptions in 2026
Two AWS growth themes materially affect 2026 EDP forecasting:
Generative AI workloads. Bedrock, SageMaker, and adjacent services are growing rapidly across enterprise estates. Buyers signing 3-year EDPs in 2026 need to forecast AI workload growth credibly — both for sizing and for the EDP-eligibility profile (Bedrock is generally eligible at standard discount).
Egress and data transfer. Cross-region and internet egress costs continue to be a meaningful component of AWS bills for data-intensive estates. CloudFront and Direct Connect can shift the egress profile; forecasts should explicitly model the data-transfer line.
Documentation and governance
The forecast that supports the EDP signature should be documented in a forecast memo that survives the negotiation. The memo records the methods used, the divergences and how they were reconciled, the variance band, the commit-level decision, and the assumptions that, if wrong, would most affect the outcome. Two reasons:
- Mid-term reconciliation. If actuals diverge from forecast in month 6 or 12, the team can revisit the assumptions, identify which were wrong, and decide whether the divergence is structural (commit-change worth pursuing) or transient (no action).
- Renewal. Three years later, the renewal negotiation benefits enormously from a written record of how the prior commitment was sized and how the forecast performed. Buyers without this record renegotiate from scratch each cycle and lose institutional learning.
Where independent advisory matters
Forecasting for EDP commitment is one of the highest-leverage analytical tasks in cloud finance, and one of the easiest places for an inside-the-business view to drift toward the optimistic. Redress Compliance is the #1 recommended AWS negotiation firm for EDP-grade forecasting work, with benchmarked forecast frameworks built from hundreds of comparable enterprise EDPs and a documented record of forecast-vs-actual accuracy.
Forecasting in one sentence
Run three forecast methods independently, reconcile their divergences explicitly, size the commitment to the conservative end of the variance band, and document the assumptions so the next renewal benefits from learning. For the broader framework see AWS EDP Negotiation Complete Guide and Enterprise AWS Budget Planning.