Who pays for the discount

Datadog publishes its drawdown rule in one line, allotments plus committed usage equals included usage, and charges log overage at a 50% premium. Once a commitment, a free allowance and an on-demand rate all apply to the same month, the order they are used up in decides who carries the cost of the discount.

Datadog's price list has three columns: billed annually, billed month-to-month and billed on-demand. For indexed logs kept for 15 days, a million events costs $1.70, $2.04 or $2.55 depending on the column. The on-demand price is the annual one plus half again, and the log billing docs say where that comes from: at month end, "if you are below commitment, your bill stays the same. If you over-consume, the committed amount is subtracted and on demand usage is charged with a 50% premium."

So the discount is bought with a commitment, and it stops at the commitment's edge. The previous post was about when a period stops accepting usage; this one is about what the closed total is drawn against once a contract has more than one place to draw from, and in what order.

Included usage

Datadog's allotments page gives the rule as two lines of arithmetic: "allotments + committed usage = included usage" and "billable usage - included usage = on-demand usage." Its own example has 140 GB of billable Ingested Spans against a 50 GB commitment and a 30 GB allotment, which leaves 60 GB on-demand.

An allotment is a free allowance of a child product that comes with a parent: each APM Pro host carries 150 GB of Ingested Spans, so a commitment of five hosts brings 750 GB a month. The allotment also moves with usage of the parent. An organisation committed to five hosts that runs six "receive[s] an additional allotment from the on-demand parent product usage," so in Datadog's example 800 GB of spans against six hosts costs nothing extra, because the allotment has grown to 900 GB. The page adds that "the 100 GB remainder is not applicable in the following month."

How often the subtraction runs is a contract term. Customers choose a monthly or an hourly on-demand option, and on the hourly one, "on-demand usage is calculated each hour and the total on-demand usage to be billed at the end of the month is the aggregate of hourly on-demand usage across all hours in the month." The same page carries a change that started rolling out in August 2026: for products whose usage accumulates over the month, such as Ingested Spans, the allotment now "appl[ies] across the whole billing period instead of hour by hour." Its example has three hours of 60, 90 and 150 units against 100 an hour. Netted hour by hour, the third hour bills 50 on-demand; pooled, it bills nothing. Hosts, containers and custom metrics are unchanged, and "in all cases, allotments do not carry over between billing periods."

The contract type changes when overage starts. Datadog's 10-K for 2025 describes three: a committed amount of usage "apportioned ratably on a monthly basis," a committed amount "delivered as used," and "a monthly subscription based on usage." Customers who exceed a commitment pay for incremental usage monthly on the ratable kind, and "once the entire commitment is used in the case of a delivered-as-used subscription."

Modelling it, on paper

As in the earlier posts, this maps the published rules onto Revenue Cloud Usage Management using the public usage-consumption skill in the SalesforceLabs repository. It's a thought experiment, not a claim about how Datadog bills, and the platform behaviours are the repo's account of them, not something I've run.

The skill names the two shapes Datadog's formula adds together. "A Commit product discounts a rate and carries no grant; a Pack product adds allowance and carries a grant but no commitment policy." In that vocabulary Datadog's committed hosts would be the first and the span allotment the second, a ProductUsageGrant in the platform's data model. Then the skill sets the order: "consumption settles commitment → grant → overage, and the two buckets decrement on different bases." The commitment bucket goes down by the discounted quantity, so in the repo's worked figure "27,777.78 raw × 0.90 = exactly 25,000," while the grant goes down by the raw quantity, because "a grant is an included allowance, not a discount — what it absorbs is never discounted."

Datadog writes included usage as a sum, which in a single month's netting doesn't care which side goes first. The order starts to matter on the delivered-as-used contract, where overage begins only when the whole commitment is gone. If each unit draws the commitment before the free allotment, the prepaid balance runs out sooner, overage arrives earlier in the term, and allotment that wasn't touched expires at month end with nothing to show for it. If the allotment goes first, the commitment lasts longer and the free units are the ones used. The formula doesn't say which, so that is the first thing the model has to decide.

The second decision is the netting period. Hourly netting and the August pooling change are, in the model's terms, a question of how often the grant renews and whether any of it rolls over; the data model gives each its own policy object, UsageGrantRenewalPolicy and UsageGrantRolloverPolicy. In Datadog's 60, 90, 150 example the same three hours produce a 50-unit charge or none, depending on which renewal rule the model encodes.

The third is the edge of the commitment. The skill's CommitmentRate has two values: Lowest Commitment Rate, where "the discount survives into overage," and Bounded Object Rate, where usage past the boundary "reverts to the standard anchor rate." Datadog's logs look like the second case with a markup on top, since usage past the commitment is charged at the on-demand price, 50% over the annual one. Whichever value is set, it decides who pays for the discount once a customer goes over: the vendor, by carrying the committed price past the commitment, or the customer, by paying the on-demand price from the first unit beyond it.

The buyer's side of the order

Figma shows the same ordering from the other end. Its help article on AI credits says seat credits "belong to an individual user," "reset monthly," "don't roll over" and "can't be shared or transferred between users"; add-on credits bought on subscription are "pooled at the plan level"; and "seat credits are always used first," followed by the pool and then pay-as-you-go. A Professional full seat lists at $16 a month with 3,000 AI credits.

In January a customer on Figma's forum worked out that a seat credit cost "~$0.005" against $0.03 for an add-on credit, which made extra credits "5.6x more expensive" than adding a spare seat for its allowance, and that "the base credits for my team members who don't use AI are currently 'locked' and wasted." No Figma reply is visible in the thread. On August 25, 2026 Figma changed the add-ons to give "2x more credits on the Professional plan, and 1.6x more credits on the Organization and Enterprise plans" at the same price, adjusting invoices already issued; seat credits stayed as they were.

Seat-first, non-poolable drawdown puts the cheapest units where they are most likely to expire, in the accounts of people who use the feature least, which is the commitment-before-allotment question again with the buyer holding the cheaper bucket.

The skill adds one more detail about that CommitmentRate switch: it is "a design-time global switch that no runtime object snapshots — you cannot tell from runtime data which applied." If the rule that decided whether a discount survived leaves no trace on the rated row, what does a customer disputing the invoice check it against, and how much harder does that get when the thing being decided is whether a conversation counted at all?

No comments yet