Usage-Based Pricing vs Seat-Based Pricing

The most important fact about how a SaaS company prices is not the number on the invoice. It’s what that number is attached to.

Tie price to seats and revenue grows when a customer hires. Tie price to usage and revenue grows when a customer succeeds. Those are different growth curves, and the choice between them quietly decides net revenue retention, forecasting accuracy, gross margin, and the entire shape of the sales motion.

That is the thesis: usage-based pricing ties revenue to value delivered and lifts net revenue retention through expansion, while seat-based pricing trades that expansion for predictability. The decision is strategic, not a billing detail.

This piece reads the tradeoff through public filings, not vendor decks. Every retention figure below ties to a specific SEC filing or shareholder letter and fiscal period. The framing is analytical: how to think about the position, not what to do about any stock.

---

Two pricing models, two growth trajectories

Seat-based pricing charges per user. Add a person, add a license. The revenue line tracks headcount, and headcount grows in steps a buyer has to approve.

Usage-based pricing charges per unit of consumption: queries run, gigabytes stored, messages sent, compute-seconds burned. The revenue line tracks what the customer actually does with the product, and it can grow without anyone signing a new order form.

Said plainly: seat-based revenue grows when the customer decides to grow. Usage-based revenue grows when the customer’s workload grows, decision or not. That single difference is why the two models produce such different retention numbers, and why the choice is upstream of almost everything else in the unit economics.

---

Usage-based leaders show expansion through consumption growth

The clearest evidence sits in the net revenue retention (NRR) line, the percentage of last year’s revenue a cohort of existing customers still represents this year, after churn and after expansion. Above 100% means the surviving base spent more, net, even before any new logos.

Usage-based companies post the strongest NRR in software, and the mechanism is structural. When a customer’s product grows, its consumption grows, and the bill grows with it, automatically.

Per Snowflake’s Form 10-K for the fiscal year ended January 31, 2025, net revenue retention was 126%. Per Datadog’s Form 10-Q for the quarter ended September 30, 2025, dollar-based net retention was approximately 120%, stable with the 120% reported for the quarter ended June 30, 2025. Per Twilio’s Form 8-K for Q4 2024, the dollar-based net expansion rate was 106%.

None of those numbers required a sales rep to negotiate a seat increase. They are the byproduct of customers using more.

---

Seat-based trades expansion for predictability

Seat-based pricing gives up some of that organic expansion. In exchange, it buys something usage-based models struggle to match: a revenue line you can forecast.

A seat is contracted. It renews on a known date at a known price for a known quantity. Finance can model next quarter from the renewal schedule and the pipeline of new seats, with consumption volatility largely out of the picture.

Atlassian is the instructive case because it is seat-based and still posts elite retention. Per Atlassian’s Q2 FY26 shareholder letter (February 2026), Cloud net revenue retention was above 120%. But the composition is different: that expansion comes from contracted multi-year deals, edition upgrades, and intentional seat growth, not from a meter ticking in the background.

The contrast appears when expansion slows. Per Asana’s Q3 FY26 earnings announcement (December 2025), dollar-based net retention was 96%, stable with the 96% reported in Q2 FY26. Below 100%, the existing base is contracting net of expansion. A seat-based model at 96% is telling you the same thing every quarter: growth has to come from new logos and explicit upsell, because the meter is not doing the work.

---

Real data: pioneers versus incumbents

The split is sharpest side by side. The table below is an original analytical asset; every retention figure is sourced to the filing cited in its row.

Company	Pricing model	NRR / net expansion	Period	Source
Snowflake	Usage-based	126%	FY2025 (ended Jan 31, 2025)	Form 10-K
Datadog	Usage-based	~120%	Q3 2025 (ended Sep 30, 2025)	Form 10-Q
Twilio	Usage-based	106%	Q4 2024	Form 8-K
Atlassian Cloud	Seat-based	120%+	Q2 FY26 (Feb 2026)	Shareholder letter
Asana	Seat-based	96%	Q3 FY26 (Dec 2025)	Form 8-K

Read the table by mechanism, not just by magnitude. The three usage-based names cluster at or above 106% on consumption that grows without a purchase decision. The two seat-based names split: Atlassian sustains 120%+ through deliberate, contracted expansion, while Asana sits below 100% because seat expansion is a sale that has to be made, repeatedly, and right now it is not clearing the churn line.

The headline numbers can converge (Datadog and Atlassian both near 120%) while the engine underneath them is completely different. That distinction is the whole point.

---

The NRR arbitrage: why expansion economics diverge

Call it the NRR arbitrage. The same retention percentage can be earned or it can be collected, and those are not equally durable.

In a usage-based model, expansion is collected. The customer scales its own product, consumption rises, the invoice rises. The vendor’s marginal cost to capture that dollar is close to zero, which is why usage-based NRR tends to run high without a proportional jump in sales spend.

In a seat-based model, expansion is earned. Every incremental dollar above flat renewal is a deliberate motion: a new team onboarded, an edition upgraded, a multi-year deal negotiated. That can absolutely work (Atlassian’s 120%+ proves it) but it consumes sales and customer-success effort that the usage meter would otherwise have done for free.

This is why the models diverge on gross margin pressure too. Usage-based revenue is often coupled to a real cost of goods, the cloud compute under the meter, so a chunk of each expansion dollar funds the infrastructure that produced it. The relationship between pricing, cost of goods, and durable margin is its own subject, treated in Why Gross Margin Is Destiny in SaaS. Seat-based revenue, by contrast, is decoupled from incremental delivery cost, which is part of why its gross margin can look cleaner even when its NRR looks weaker.

Methodology: how to read these retention figures

• Inputs: NRR / net expansion rates from each company’s cited filing (Snowflake FY2025 10-K, Datadog Q2 and Q3 2025 10-Q, Twilio Q4 2024 8-K, Atlassian Q2 FY26 shareholder letter, Asana Q2 and Q3 FY26 8-K).
• Assumption: each company computes the metric on a comparable cohort basis (trailing-period existing customers, net of churn and expansion). Definitions vary slightly by issuer, so cross-company comparison is directional, not exact.
• Sensitivity: usage-based NRR is the more volatile input. A consumption-led base can swing several points quarter to quarter on customer workload alone, whereas a contracted seat base moves slowly.
• What this misses: NRR says nothing about new-logo growth, gross retention, or absolute dollar size. A 96% NRR company adding logos fast can out-grow a 126% NRR company that has stopped landing accounts. Retention is one input, not the scoreboard.

---

Where usage-based is vulnerable

A credible analysis names the holes. Usage-based pricing has three.

Revenue is volatile by construction. If the bill rises when customers do more, it falls when they do less. A demand shock, a customer cost-cutting cycle, or a single large account optimizing its consumption can move the revenue line in ways no renewal schedule predicts. The same meter that delivers 126% NRR in a good year delivers the disappointment in a bad one.

Forecasting is opaque. Finance cannot model a meter the way it models a contract. Quarterly guidance becomes a probabilistic read on aggregate customer behavior, which is why usage-based companies often carry wider guidance ranges and more guidance risk than their seat-based peers.

NRR decays as the base matures. This is not hypothetical. Per Snowflake’s FY2025 10-K and investor materials, net revenue retention fell from 177% in FY2022 to 126% in FY2025. The product did not get worse. The base got bigger and normalized: once a platform is broadly adopted inside its accounts, the rate of net-new usage expansion per account naturally slows. A still-excellent 126% is the mature shape of what used to be a 177% land grab.

---

Where seat-based is vulnerable

Seat-based pricing has the opposite failure mode: it leaves expansion on the table.

Price is decoupled from value. A customer can extract enormous value from a seat-based product, automating work, displacing headcount, running mission-critical workflows, and still pay exactly the contracted per-seat rate. The vendor captures none of that upside automatically. Worse, a product that makes each user more productive can reduce the number of seats a customer needs, putting the pricing model in direct tension with the product’s own promise.

Expansion is a recurring sales cost. Because growth above flat renewal must be sold, seat-based NRR depends on a customer-success and upsell motion that has to be funded and executed every quarter. When that motion stalls, retention drifts toward 100% and below, which is the story Asana’s 96% (Q3 FY26 8-K) is telling. The expansion that usage-based peers collect for free, seat-based vendors have to go win.

The AI era sharpens this. Agentic and AI-assisted software increasingly delivers outcomes rather than occupying a seat at a desk, which is precisely why so many AI products are reaching for consumption or outcome metering instead of per-user licenses. The cost dynamics that push usage-priced infrastructure businesses around are visible in AWS Margin Pressure and the Cloud Reset.

---

Founder playbook: when to choose each, when to hybridize

The choice is not ideological. It follows from what your product actually scales with.

Choose usage-based when consumption tracks value. If the customer gets more out of the product by doing more with it (running more queries, storing more data, sending more messages), meter the thing that grows. You will accept volatility and harder forecasting in exchange for expansion you don’t have to sell. This is the land-and-expand engine behind Snowflake, Datadog, and Twilio’s numbers above.

Choose seat-based when value is per-person and predictability is the priority. If the product is a tool a defined set of humans use (project management, design, collaboration), seats map cleanly to value and give you a forecastable revenue line. You accept capped organic expansion in exchange for being able to model the business. Reaching elite retention here, as Atlassian does, means building a deliberate expansion motion, not waiting for a meter.

Hybridize when a single dimension misprices someone. Many durable models combine a seat or platform base with a usage component on top: a committed floor for predictability, a meter above it for upside. The committed-use contract, prepaid credits drawn down by consumption, is the same instinct, smoothing usage volatility into something finance can forecast while preserving consumption upside.

Here is the mechanism at founder scale, as an illustrative, hypothetical example (numbers invented to show the logic, not drawn from any company). Suppose a product charges 25 dollars per seat and a customer buys 40 seats: 1,000 dollars a month, flat until they hire. The same product metered at, say, a fraction of a cent per processed item, against a customer running 3 million items a month, grows its bill every time that customer’s own volume grows, with no new order form. Same product, same value delivered, two completely different revenue curves. The only variable that changed was what the price was attached to.

That is the decision. Not how to bill. What to bill for.

---

How the pieces fit together

Usage-based and seat-based pricing are not better and worse. They are bets on different sources of growth.

1. Usage-based bets that expansion you collect automatically beats the volatility it imports. The payoff shows up as elite NRR (Snowflake 126%, Datadog 120%, Twilio 106%) and the cost shows up as forecasting risk and NRR decay at maturity.
2. Seat-based bets that a forecastable, contracted revenue line is worth capping organic expansion. The payoff is predictability; the cost is that every dollar above flat renewal has to be sold, and when that motion slows, NRR drifts below 100% (Asana 96%).
3. The headline retention number can match across models while the engine differs entirely (Datadog and Atlassian both near 120% on opposite mechanics), so read composition, not just magnitude.
4. The strongest position is usually a hybrid: a committed base for the forecast, a meter on top for the upside.

The framing carries across the portfolio. The same surface-versus-payload logic that governs platform strategy in Google’s AI Strategy Is a Distribution War shows up here as model-versus-meter: the pricing surface, not the feature, decides who captures the value the product creates.

That’s the whole tradeoff. The rest is instrumentation and nerve.

---

Analysis, not investment advice. Retention figures are drawn from the cited public filings and shareholder letters (Snowflake, Datadog, Twilio, Atlassian, Asana) and tied to their fiscal periods inline. Frameworks here are for understanding pricing strategy and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for reading filings like this, the NRR-decomposition worksheet, the pricing-model decision matrix, and the unit-economics scorecard used above? It’s in the Tech Business Analysis Playbook.