Net Revenue Retention Across 12 Q2 Filings: Usage Pricing Leads, for Now

Net revenue retention across twelve public software companies in the latest quarter ran from 97% at Asana to 126% at Snowflake, a spread of 29 points. Sort the twelve by how they charge and the ranking almost sorts itself: metered vendors at the top, per-seat vendors at the bottom. That looks like proof that usage pricing retains customers better. Read the definitions in the filings and most of it is something else.

What net revenue retention was built to measure

Net revenue retention is the revenue you collect this period from a fixed group of customers, divided by what the same group paid a year earlier. New customers are excluded. Expansion counts in, and contraction and churn count against.

A figure above 100% means the existing base grew on its own. Below 100% means you need new logos just to stand still. That is why investors treat it as the single cleanest read on customer health.

The seat-era assumption inside the metric

The metric grew up in per-seat software. In that world a rising number had one main cause: your customer hired people, or rolled the product to another team, and bought more licences. Price increases happened at renewal and were visible.

So NRR worked as a proxy for adoption. More seats meant more people relying on the product. A buyer does not add 200 seats to a tool nobody opens.

Why a meter changes what the number means

Under usage pricing, revenue moves with the customer's workload rather than their headcount. A data team runs heavier queries, an app sends more messages, an AI product calls more inference. NRR climbs without a single new human user.

That is adoption of a kind. It is also exposure to things the vendor does not control: the customer's own traffic, their optimisation projects, and the price the vendor charges per unit. My view is that a metered NRR and a seat NRR are different instruments that happen to share a name. Comparing them on one axis is the most common error in SaaS commentary this year.

The 12 net revenue retention disclosures side by side

The table below collects the most recent disclosed figure from twelve public companies that report a retention metric. The period is calendar Q2 2026 for June year-ends, and the quarter ending 31 July 2026 for companies on January fiscal years.

Each company had to publish a numeric retention rate for the period, and the set had to span the three pricing structures that matter now: consumption, volume tiers, and seats. I excluded companies that stopped disclosing the metric, which is itself a signal covered in the limitations section.

Nine figures come from an earnings release or SEC filing. Three come from earnings-call coverage because the press release does not carry the number: HubSpot, Okta and DigitalOcean. Those three are marked, and they deserve less weight.

Box's year-earlier figure is from SaaStr's reading of the company's disclosures. Okta's and HubSpot's year-earlier figures are from call coverage. Klaviyo's release says "up one percentage point", from which 108% is implied.

The four usage-heavy vendors sit between 116% and 126%. The two volume-tiered marketing platforms sit at 109% and 110%. The five seat vendors run from 97% to 109%, with a median of 106%.

DigitalOcean breaks the pattern. It is metered, and its 102% sits beside HubSpot. That single exception is more informative than the eleven that conform, because it shows a meter does not produce high retention on its own.

Why usage-priced net revenue retention runs high right now

The obvious reading is that metered pricing captures expansion that seats leave on the table. That is partly true. The fuller reading is that the four top vendors sell infrastructure that AI workloads consume, and AI workloads grew fast across the past twelve months.

Snowflake and Datadog: consumption at scale

Snowflake describes its own model plainly in the release. It recognises revenue "based on platform consumption, which is inherently variable at our customers' discretion". Its retention rose from 125% at the January fiscal year-end to 126%, alongside 828 customers paying more than $1 million in trailing product revenue.

Datadog's quarterly filing puts trailing dollar-based net retention in the low 120% range, against about 120% a year earlier, and attributes the rise to larger usage growth from existing customers. Gross revenue retention, which ignores expansion, stayed in the mid-to-high 90s according to call coverage.

The largest-customer problem

Here is where workload-driven retention shows its cost. Datadog renewed its largest customer, an AI company, in what management described as a nine-figure deal. Management also said on the call that it expects that customer's usage to decline from the third quarter, and built the reduction into its outlook.

A seat vendor rarely faces that. One customer cutting its query volume can move a usage vendor's NRR by points, and the customer can do it without cancelling anything. I would treat any usage-priced NRR above 120% as partly a measure of the AI build-out, and price it as cyclical rather than structural.

DigitalOcean's chief financial officer said almost exactly this from the other direction. Call coverage reports Matt Steinfort saying the metric has become less representative as growth comes from the biggest customers and newer AI customers, and that the company would stop emphasising it. When the vendor tells you its own retention metric no longer describes the business, believe it.

Pricing-driven movement hides inside the NRR definition

Every company in the set defines retention slightly differently. The differences are not footnote trivia. They decide whether a price change shows up as customer health.

Paraphrased from the definitions sections of each company's Q2 release. Read the originals before comparing any two companies directly.

Klaviyo counts price increases as retention

Klaviyo's definition states that current-period ARR "includes any expansion, price increases" and reactivated subscriptions. That is honest disclosure, and it means a repricing lifts NRR exactly as adoption would.

Klaviyo gave a live example. On 18 February 2025 it began billing on all active profiles rather than only on contacts it emailed, with increases capped at 25% for existing customers. In the Q2 call, management cited the lingering effect of that enforcement as a drag on retention and said the trailing measure carries it through the first quarter of next year. A billing rule change is sitting inside a customer-health metric, in both directions, for two full years.

Twilio counts price cuts against itself

Twilio's release states that its expansion rate "decreases when these customers cease or reduce their usage of a product or when we lower usage prices". So Twilio's 116% is net of any price reductions it granted. That makes it a cleaner adoption signal than Klaviyo's, not a weaker one.

There is a second pricing effect in Twilio's numbers. Carrier fees for business messaging are passed through at cost, and the release strips $71.05 million of those fee increases out of organic revenue for the quarter. It does not say whether the same fees are stripped from the expansion rate. That is the question I would put to investor relations before crediting the 8-point gain to adoption.

Exclusions that flatter or depress

Cloudflare excludes free users who convert to paid, which keeps its self-serve funnel out of retention. Snowflake keeps zero-usage customers in its base, which is the conservative choice. Asana averages four quarterly rates, which smooths movement in both directions. None of these is wrong. All of them mean that a 3-point gap between two companies can be definitional rather than operational.

Seat vendors are adding credits, and NRR has not caught up

The five seat vendors all sit below the usage cluster, and four of them are now bolting a meter onto the seat. The pattern is consistent: keep the seat price, include a credit allowance, sell more credits on top.

monday.com and the credit allowance

monday.com's Q1 filing announced a seats-plus-credits model for its AI Work Platform. It applies to customers joining from 6 May 2026, with credit consumption for its Sidekick assistant starting 20 May and for agents on 8 June. Existing customers can opt in.

By the Q2 release, AI products represented 17% of net new ARR and AI ARR had doubled from the prior quarter. Overall net dollar retention fell from 111% a year earlier to 109%. The definition is a trailing weighted average of four quarterly calculations built on a year-ago cohort, and new platform customers are not in that cohort. Credits launched for new customers in May cannot yet appear in a metric that excludes new customers by design.

HubSpot, Box and Asana

HubSpot's 102% is down 1 point year over year, with customer dollar retention in the high 80s per call coverage. Summaries of the call say credit consumption grew after HubSpot cut credit prices in April. If accurate, that is a pricing action and an adoption signal pulling the same line in opposite directions, which the earlier analysis of HubSpot's credit repricing anticipated.

Box's 106% is the cleanest seat-side improvement, up from 103% a year earlier, and management credits Enterprise Advanced, its higher suite tier. That is packaging-driven expansion: the same seats moving to a pricier plan. It is real revenue, and it is a price-per-seat effect rather than more people using Box.

Asana sits at 97%. Its CFO said the company sees "a meaningful opportunity to build consumption- and outcome-based revenue streams alongside seats". Its AI Studio credits are a shared pool per billing account, so adding seats does not add credits. That design separates AI revenue from headcount, which is the point, and the mechanics of seat compression at renewal explain why Asana needs it.

Which way net revenue retention moved in a year

Level tells you where a company sits. Direction tells you what its pricing changes are doing. Eight of the twelve give a comparable year-earlier figure.

The two declines are both companies in the middle of a pricing transition. monday.com and HubSpot are moving seat customers toward credits, and both are seeing retention soften while their AI revenue grows. That combination is what a transition looks like from outside: the new meter is too young to lift the trailing number, while buyers renegotiate the old seats.

The same pattern turned up in private-company benchmark data on hybrid pricing, where the retention gap between models was only 6 points across medians. Public disclosures show a wider spread, mostly because the public usage vendors are larger and sell into AI infrastructure demand.

A reading guide for investors comparing NRR

Here is the checklist I use when a retention number appears in a deck or an earnings note. It takes about ten minutes per company, and it changes the conclusion more often than you would expect.

Gross retention is the most underused number on that list. It strips out expansion and shows only what customers kept. A company with 102% net and high-80s gross is expanding hard into a base that is leaking, which is a different business from one at 102% net and 97% gross.

For seat vendors adding credits, I would also ask for credit revenue as a share of net new ARR, which monday.com now discloses for AI products. Until credits are inside the year-ago cohort, that figure tells you more about the transition than NRR does. Buyers are already writing caps and true-up concessions into usage contracts, and those terms will compress metered expansion before it ever shows in a trailing rate.

Where this analysis is weakest

Twelve companies is a small and self-selected sample. Firms disclose retention when it looks good and quietly stop when it does not, so the set is biased upward. DigitalOcean is publicly walking away from the metric in this very quarter, which shows the mechanism at work.

The periods do not line up perfectly. Seven companies report a calendar quarter ending 30 June. Five report a fiscal quarter ending 31 July. A month of AI workload growth is enough to move a usage vendor by a point.

Three of the twelve figures, HubSpot, Okta and DigitalOcean, come from earnings-call coverage rather than a release I could open. They are Tier 2 sources and should be upgraded to the 10-Q or transcript before anyone relies on them.

The strongest objection is to my central claim. I argue that the usage cluster's lead is partly AI workload demand, not pricing. A fair critic would say that is a distinction without a difference: capturing workload growth is exactly what a meter is for, and seat vendors cannot capture it at all. That is a good argument. My answer is only that the same exposure runs in reverse, and Datadog's guidance on its largest customer is the first visible instance of it.

Finally, no company in the set splits its retention into price and volume. Every separation in this post is qualitative, reasoned from definitions and management commentary. A vendor that published that split would make this whole exercise unnecessary, and would earn a valuation premium for doing it.

Frequently asked questions

Where to start

If you hold or are evaluating any of these twelve, pull the latest release and copy its retention definition into the same sheet as the number. Mark one column for whether price increases are included and another for the window. Half the comparisons people make between these companies will not survive that sheet.

If you run a SaaS business, publish gross retention next to net, and split net expansion into price, packaging and volume in your board pack. You do not need to disclose it publicly to benefit. The exercise will tell you whether your expansion comes from customers using more, or from your pricing team, and those deserve very different valuations. The underlying question is which value metric survives automation, and retention is where the answer shows up first.

Related reading