Most AWS teams either over-commit locking in spend that later sits unused or under-commit, paying on-demand rates for baseline workloads that will never go away. Both failures trace back to treating a Savings Plan purchase as a one-time event rather than a structured, ongoing portfolio decision.
Compute Savings Plans deliver discounts of up to 66% on EC2, Fargate, and Lambda. EC2 Instance Savings Plans go up to 72% on specific instance families in specific regions. Database Savings Plans cover RDS, ElastiCache, and DocumentDB at 20–35% off on-demand rates. The gap between a well-layered portfolio and a poorly timed single purchase is routinely 10–20 percentage points of realized savings.
What Is an AWS Savings Plan Buying Strategy?
It's the deliberate approach to deciding which plan types to buy, how much hourly commitment to enter for each, when to purchase, and how to sequence renewals to maintain coverage without overexposure.
The stakes are concrete. A $2M/month AWS bill with 60% eligible workloads carries $1.2M/month in commitment-eligible spend. Getting the commitment level wrong by 15% in either direction means $180K/month in wasted commitments or $180K/month in foregone savings; a $2.16M swing over a 1-year term.
Three variables control outcomes: plan type selection, commitment sizing, and purchase timing.
Which Plan Type Goes First?
Compute Savings Plan first. Highest flexibility, broadest coverage, absorbs workload shifts between instance families and services. This is your baseline layer.
EC2 Instance SP second. Only after you have ≥90 days of stable, proven usage on a specific instance family (e.g., m6i in us-east-1). The higher discount justifies the tighter constraint.
Database SP third. If you have meaningful, stable RDS, ElastiCache, or DocumentDB spend. These plans do not overlap with Compute SPs.
A common mistake is reversing this order committing to an EC2 Instance SP on a specific family before validating that the usage pattern is genuinely stable.
If you want to understand how each plan type works under the hood, we covered it in detail here AWS Savings Plans Explained: Types, Pricing & How They Work
How Much Should You Commit?
Commit to your minimum baseline, not your average
Average hourly spend includes spikes. If your EC2 usage averages $8/hour but drops to $4/hour on weekend nights, a commitment based on the average produces waste 30%+ of the time.
The correct approach:
- Pull 90 days of hourly EC2 usage data from Cost Explorer
- Sort by hour and identify the true floor not the average
- Commit to 70–80% of that floor
- Leave 20–30% as buffer for measurement error and seasonal variation
Example: 90-day minimum hourly EC2 spend of $12.50/hour → commit $9.38/hour (75%). At a 40% Compute SP discount, that's ~$3.75/hour in savings, or ~$32,850/year.
The 75% rule exists because AWS Cost Explorer recommendations are calculated on historical data refreshed every 72+ hours. Any spend pattern change in the last three days is invisible. Committing to 100% of the recommendation means betting on 72-hour-old data.
Rightsize before you commit
A Savings Plan on an overprovisioned instance locks in the waste at a discount. An m5.4xlarge running at 12% CPU utilization, discounted 40% via a Compute SP, still costs 88% of what the right-sized instance at full utilization would cost without any commitment.
Pre-commitment checklist:
- Pull Compute Optimizer recommendations for all instances covered by the planned commitment
- Resize any instance with <30% average CPU utilization
- Wait 14 days post-resize before purchasing (allow new baseline to stabilize)
- Re-pull the minimum-baseline calculation on post-resize usage data
Skipping this step is the single largest source of over-commitment waste in FinOps audits.
Rightsizing alone doesn't solve the full picture Why Cloud Resource Optimization Alone Doesn't Fix Cloud Costs
What Is Layering and How Do You Build a Portfolio?
Layering is holding multiple AWS Savings Plans simultaneously, each covering a different segment of your workload, with purchase and expiration dates deliberately offset from each other.
A layered portfolio might look like:
- Base: Compute SP (1-yr), $9.00/hr, purchased Jan 1, expires Jan 1 next year
- Flex: Compute SP (1-yr), $3.50/hr, purchased Apr 1, expires Apr 1 next year
- DB: Database SP (1-yr), $4.00/hr, purchased Jan 1
- Growth: EC2 Instance SP (1-yr), $2.00/hr, purchased Jul 1
Why this outperforms a single large plan:
- Expiration staggering; if all plans expire the same day, you face a cliff: full on-demand rates until you analyze and repurchase. Offset expirations keep you partially covered.
- Risk isolation; if a workload shrinks, only one tranche is at risk, not the entire portfolio.
- Rolling optimization;each renewal cycle is an independent decision point using fresh usage data.
Staggering rule of thumb: No two plan expirations within 60 days of each other. No single plan representing more than 40% of total committed spend.
When Should You Buy?
Avoid end-of-month purchases. A plan bought on the 29th gives you 2–3 days of savings before the bill closes, but the commitment clock starts immediately. Buy on or before the 5th of the month.
Post-change waiting periods. If your team has recently migrated workloads, deployed significant new capacity, completed a rightsizing exercise, or moved from EC2 to Fargate or Lambda, wait at least 45 days before purchasing. Cost Explorer data will reflect the old workload mix for 30–45 days after a change.
Seasonality awareness. Pull the usage floor from your lowest historical quarter, not the most recent 90 days especially if your workload is seasonal.
1-Year vs 3-Year
The Compute SP no-upfront discount is ~34% on 1-year and ~54% on 3-year. The practical rule: never buy a 3-year plan for a workload you haven't observed for at least 12 months. The 20-point discount improvement doesn't offset a usage drop of more than 15% when you're locked in for 36 months.
Payment Options
- All Upfront: lowest effective rate, highest total discount. Choose this when the commitment level is validated and the workload has 12+ months of stable history. The NPV advantage over No Upfront on a $100K/year commitment can exceed $8,000 over the term.
- No Upfront: highest flexibility, lowest discount. Right for high-growth environments, changing infrastructure, or a first purchase where you're validating the model.
- Partial Upfront: useful when finance limits capital available for upfront cloud commitments. Run the break-even math for your specific commitment size before selecting.
Multi-Account Organizations
Savings Plans purchased in the payer (management) account apply automatically across all linked member accounts. Plans purchased in a member account apply only to that account unless sharing is enabled.
Always purchase from the payer account. Purchasing from member accounts creates fragmented coverage and prevents portfolio-level optimization. If a member account's workload shrinks, the payer account's other member accounts absorb excess commitment reducing, not eliminating, waste.
What Happens If You Over-Commit?
Over-commitment is an operational reality. When it happens, these are your options:
- AWS Marketplace: standard RIs can be listed there, but Savings Plans cannot.
- Convertible RI exchange: Convertible RIs can be exchanged for different instance families or regions. Savings Plans are not exchangeable.
- Wait for natural expiration: let the plan expire without renewal, or reduce the renewal commitment. The plan continues applying to any eligible usage; it only produces waste when usage falls below commitment.
- Platform-level buyback: this is the structural gap that Flex Commitment models (like Usage.ai Flex Commitments) are built to address. The platform holds the commitment on your behalf and provides cashback on underutilization, so the financial risk of a usage drop doesn't transfer to your AWS bill.
For teams managing $1M+/month in commitment-eligible spend, a single over-commitment event can exceed the platform fee many times over.
Before committing, it helps to understand the full RI landscape AWS Reserved Instances: Complete Guide to Pricing, Types & Savings
The 72-Hour Recommendation Lag Problem
AWS Cost Explorer Savings Plan recommendations refresh every 72+ hours. At $6,000–$12,000/day in uncovered on-demand spend, a 3-day lag means your commitment sizing is based on a usage picture that may already be $18,000–$36,000 out of date. Size a commitment to 90% of coverage based on that stale data, and you're potentially mis-committed from day one.
Usage.ai refreshes recommendations daily, purchases commitments automatically, and provides cashback on underutilization so the precision gap and its financial exposure don't land on your AWS bill. The fee model is a percentage of realized savings only; if no savings are generated, no fee applies.
Post-Purchase Monitoring
Two metrics matter after purchase:
- Utilization rate: percentage of your hourly commitment actually consumed by eligible usage. Below 80% means you over-committed for that period.
- Coverage rate: percentage of eligible on-demand spend covered by Savings Plans. Below 80% means you have significant spend that could be covered by additional commitments.
The relationship: high utilization + low coverage = under-committed. Low utilization + high coverage = over-committed. High utilization + high coverage = optimal.
Pull both reports from Cost Explorer → Savings Plans → Utilization Report and Coverage Report weekly during the first 90 days post-purchase. Monthly monitoring is sufficient after that for stable workloads.
Workload-Specific Notes
EC2-heavy (>70% of bill): Lead with a Compute SP covering 70% of minimum EC2 baseline. After 90 days, layer an EC2 Instance SP for the single highest-spend instance family. Don't purchase EC2 Instance SPs across more than 2 families simultaneously.
Containers (ECS/Fargate): Compute SPs are the only type that covers Fargate. Include Fargate hourly spend in your baseline calculation, not just EC2. Many teams under-commit their Compute SP because they calculated the floor on EC2 data only.
Serverless / Lambda: Lambda is covered by Compute SPs at ~17% discount. For high-invocation-volume workloads, the absolute dollar impact is material. Include Lambda spend in your Compute SP baseline.
Database-heavy (RDS, ElastiCache, DocumentDB): Database SPs are entirely separate from Compute SPs. If you spend $50K/month on RDS and $200K/month on EC2, you need a $200K-scale Compute SP and a $50K-scale Database SP. One type does not cover everything.
Multi-account Organizations: Purchase all Savings Plans from the payer account. Coordinate timing to avoid staggering collisions.
A Practical Example: Building a Portfolio from Scratch
Scenario: Mid-size SaaS, $350K/month AWS bill: $220K EC2, $60K RDS, $40K Fargate, $30K Lambda + other.
Step 1: Pull 90-day minimum baseline: EC2 floor $14.00/hr, Fargate $3.50/hr, Lambda $1.20/hr → total Compute-eligible floor: $18.70/hr
Step 2: First Compute SP purchase: 75% of $18.70 = $14.03/hr commitment, No Upfront. At ~34% discount: saves ~$5.33/hr → ~$46,700/year
Step 3: Database SP: RDS floor $7.50/hr → commit $5.63/hr. At ~30% discount: saves ~$1.69/hr → ~$14,800/year
Step 4: 90-day review: Utilization 94%, coverage 68%. Coverage gap signals room to increase. New floor calculation yields a $3.00/hr incremental tranche.
Step 5: Layer 2 Compute SP: $3.00/hr, purchased 90 days after Layer 1. Expiration now offset by 90 days.
Total annual savings: ~$61,500/year. No single expiration date represents more than 55% of committed spend.
(All pricing estimates are illustrative. Verify current rates at aws.amazon.com/savingsplans/pricing.)
What's the biggest mistake you've made or seen with AWS Savings Plan sizing? Over-committed on a workload that shrank, or left significant on-demand spend uncovered?
Explore the end-to-end breakdown here → AWS Savings Plan Buying Strategy: Layering, Timing & Right-Sizing Commitment
Top comments (0)