Committed-Use Discounts Are the 2026 NRR Shock Absorber for Usage-Based SaaS

The sharpest 2026 pricing signal is not a discount percentage. It is the gap between independent SaaS retention benchmarks and public usage-infrastructure expansion results. SaaS Capital’s 2026 benchmark reports 103% median NRR for bootstrapped SaaS companies with $3M to $20M ARR, while the 90th percentile is 117.9%.

Against that floor, Datadog reported about 120% trailing 12-month dollar-based net retention, and MongoDB reported approximately 121% net ARR expansion. Those are company-reported figures, not independent market-wide benchmarks. Still, the contrast frames the 2026 question: how should usage-based infrastructure vendors protect expansion when consumption is powerful but uneven?

Keep PAYG, but stop giving away enterprise discounts

The answer is not to abandon pay-as-you-go. PAYG remains useful for trial, workload discovery, and uncertain consumption. The sharper move is to reserve meaningful enterprise discounting for annual commit-to-consume pools, prepaid units, or order-form ACV tiers.

Snowflake’s public service consumption table shows the structure clearly: capacity credit pricing applies a Snowflake Credit Discount from the customer Order Form to the on-demand credit price. It also shows AI credit pricing tied to annualized contract value ranges, with global AI credits listed at $2.00 on demand and $1.88 at the $40M plus ACV tier. Negotiated enterprise discount percentages remain order-form specific, so the public evidence shows the mechanism rather than every customer’s realized deal.

Commitment converts volatility into a renewal floor

Usage infrastructure has a forecasting problem as much as a monetization problem. MongoDB says Atlas revenue is primarily usage-based and billed either monthly in arrears or paid upfront, while also noting that Atlas usage can vary and revenue timing is less predictable when usage fluctuates. That is exactly where committed consumption can act as a shock absorber.

The goal is to turn part of the customer’s expected consumption into a committed revenue floor, then preserve upside through overage. Datadog’s 2025 Form 10-K says subscription agreements are primarily monthly, annual, or multi-year, with the majority of revenue from annual subscriptions and committed contractual usage plus overage mechanics. Datadog also reported, in its own filing, about 120% trailing 12-month dollar-based net retention.

That does not prove the commitment model caused the retention result. It does show that high public usage-infrastructure expansion can coexist with annual subscriptions, committed usage, and overage rather than pure PAYG alone.

Prepaid pools need guardrails, not just procurement approval

Prepaid usage units are another public example. Microsoft’s Azure Databricks documentation says customers can prepurchase Databricks commit units and use them at any time during the purchase term. Usage is automatically deducted from the prepaid DBCU pool, while compute, storage, and networking remain separate charges.

That model is commercially attractive because it gives the vendor more committed demand and gives the buyer a clearer discount rationale. The risk is conversion_rate. If the required commitment is too high for a new or uncertain workload, the buyer may stay in PAYG longer, start smaller, or avoid the vendor entirely.

The operating discipline is therefore practical: offer PAYG entry, ramp commitments as usage becomes observable, show drawdown transparently, and make overage rules clear before they appear on an invoice. Without those controls, unused commits become renewal friction and unexpected overage becomes bill shock.

For 2026, treat committed-use discounting as a retention design, not a procurement concession. Use the 103% median NRR and 117.9% 90th percentile private SaaS benchmarks as the renewal floor to beat, then reserve enterprise discounts for customers willing to commit annual consumption while keeping PAYG for adoption and overage for expansion. Under those guardrails, commitments are a defensible NRR shock absorber; without them, they are just future contraction risk sold upfront.