Does Adding AI Features Hurt SaaS Gross Margin? Figma’s Fell 5 Points While Its Net Dollar Retention Was 136%
TL;DR: Adding AI features hurts SaaS gross margin when usage grows faster than the revenue attached to it. Figma, Amplitude, and GitHub each hit that gap and responded with credits, cost offsets, or token billing. Growth teams miss it because AI feature gross margin needs usage, billing, and cloud cost data joined.
Does adding AI features hurt SaaS gross margin?
It depends on whether the revenue attached to AI usage keeps pace with the cost of serving it. Figma’s GAAP gross margin fell to 84% from 89% in the quarter ended June 30, 2026, according to its earnings release. Cost of revenue rose to $60.5 million from $27.9 million while revenue grew 48%. CFO Praveer Melwani called it the company’s “first full quarter of AI credit monetization.”
Atlassian saw the opposite. Its gross margin rose from 83% to 87% that quarter as cost of revenues grew 2% and revenue grew 28%, however, the release doesn’t say why cost stayed nearly flat.
That evidence also answers four related questions. Inference takes a bigger share of AI product cost as products scale, according to ICONIQ. A seat price is fixed while inference cost varies with every request, which is why seat pricing breaks. A feature’s margin needs data from three systems, so a growth dashboard can’t compute it. And a free tier needs a metered allowance.
Why did Figma’s margin fall while its growth metrics rose?
Figma’s CFO pointed to free beta usage on its Q2 earnings call. “We do not charge our customers for their usage of products that are currently in beta, and we bear the cost of inference without offsetting consumption revenue,” Melwani said. Once AI credits were in place, non-GAAP gross margin rose 2.5 points from the prior quarter, which he called “the result of our first full quarter of AI credit monetization.”
Figma’s growth metrics told a different story in that release. More than 80% of paid customers above $10,000 in ARR were consuming AI credits weekly, and net dollar retention was 136% as customers expanded both seats and AI credit add-ons. Adoption and retention looked strong in the quarter gross margin fell five points, so the metrics a growth team steers by were rising while the one finance steers by was not.
How did Amplitude and Duolingo handle AI cost?
Amplitude took the cost in gross margin and planned to recover it in operating expenses. Non-GAAP gross margin was 71%, down about four points from a year earlier. On Amplitude’s Q2 earnings call, CFO Andrew Casey said higher customer adoption of AI capabilities and greater data ingestion “reduced our gross margins by an additional 2 points versus Q1,” on top of a drag from integrating Statsig. He expects margins to stay “in the low 70s” and plans to “offset that with a commensurate reduction in operating expenses.”
Duolingo held its margin by pacing the rollout. Gross margin was 72.6% against 72.4% a year earlier, which management attributed to “our measured pace of AI-powered feature expansion” and “AI cost efficiencies.” It still guides to about 71.0% for the next quarter.
What are companies doing about the cost of AI usage?
GitHub said in an April 27, 2026 announcement that “a quick chat question and a multi-hour autonomous coding session can cost the user the same amount.” It added that it had “absorbed much of the escalating inference cost behind that usage” and that the premium request model was “no longer sustainable.”
On June 1, GitHub replaced premium requests with AI Credits consumed by token use, and plan prices held steady.
Atlassian’s September 1 announcement said Rovo credits, automation steps, and AI agent resolutions will start billing on December 3, 2026, arguing that “the value of software outgrows what seat count alone can capture.” In-line summaries, rewrites, and Rovo Search stay free. Its first-quarter fiscal 2027 guidance calls for a GAAP gross margin of about 85%, from 87% in the quarter it just reported. A metering change like this moves retention numbers too, as one pricing-model change did to net revenue retention at a different company.
ICONIQ’s July 2026 State of AI report, drawn from surveys of over 300 software executives, found that consumption-based pricing rose from 35% to 42% in six months and outcome-based pricing from 18% to 23%. Two-thirds report improving per-query unit economics, yet inference’s share of AI product cost rises as products scale.
GitHub and Figma both moved to metering after absorbing the cost, by their own accounts. That ordering is the risk for a growth team. The cost shows up first, the price comes second, and the margin hit lands in between.
Why can’t a growth team see the margin of an AI feature?
AI feature gross margin is a cross-source metric, and a cross-source metric is one that cannot be computed inside any single system of record. Product analytics knows who used which feature and how often. Billing knows what each customer paid, including credits. The cloud or model invoice knows what the inference cost. Each system is right about its own corner, and a dashboard built on any one of them answers a different question.
Consider a hypothetical team that ships an AI meeting summary inside its Pro plan, with illustrative figures. Product analytics shows 62% of weekly active accounts using it. Billing shows it included at no extra charge. The cloud invoice shows it costing $0.40 per active account per week. The team celebrates the adoption number, and the margin hit arrives in a finance review a quarter later. It has the shape of the Four Cross-Source Questions, with a cost system added. I spent years in market research before building Scry, and the pattern that repeated most was a number reported with confidence from the one system able to produce it.
How do you measure the gross margin of a single AI feature?
Three steps give you a first answer from exports you already have.
Attribute inference cost to a feature and a customer, using model-provider invoices, gateway logs, or cloud tags. Count usage in free tiers and betas as cost with no revenue.
Attribute revenue to that feature. Credits sold, a share of plan price, and add-on subscriptions each need a written rule.
Join both to usage by plan and cohort, then compare each feature’s margin with the margin the business needs.
Run it as a standing question and rerun it whenever the model or routing changes, because cost per request moves with both. A customer intelligence layer does that job. Scry is built to read those exports together against a standing objective such as “which AI features earn back their inference cost, by plan?” and to report where the sources disagree. A new cost-tracking tool on its own adds another silo, which is how this problem started.
Should free trials and freemium tiers include AI features?
Include them with a metered allowance sized from cost per active user. Figma doesn’t charge for products in beta and bears the inference cost as a result. GitHub ended its fallback to cheaper models, so credits and admin budgets now govern usage. Atlassian’s new meters come with alerts before customers reach a limit and options to pre-pay or pay as they go. Duolingo pairs longer free trials with a measured pace of AI feature rollout.
An unmetered free tier turns AI usage into a cost no one set a budget for. Before the next AI feature ships, ask which system would show its margin by plan. If the answer is none, build that view first.
Caveats
This is four public companies, one platform vendor, and one survey, which falls short of a controlled study. Gross margin moves for reasons other than AI. Amplitude’s CFO attributes a drag to the Statsig integration, and Atlassian’s release doesn’t explain its rise. GAAP margins also include stock-based compensation and acquisition amortization. ICONIQ’s figures, including average AI product gross margins of 45% in 2025 and a projected 53% for 2026, are self-reported by executives at companies building AI products, and two-thirds of them report improving per-query economics, so model routing and falling prices may narrow the gap without any pricing change. The claim here is narrower than “AI hurts margin.” You can’t tell which case you’re in without a per-feature view.
—Steven Rencher, Founder of Monadux
FAQ
Does adding AI features hurt SaaS gross margin?
It hurts when AI usage grows faster than the revenue attached to it. Figma’s gross margin fell from 89% to 84% in the quarter ended June 30, 2026, while Atlassian’s rose from 83% to 87%. Track whether each AI request is paid for, absorbed, or unpriced.
What is AI feature gross margin?
AI feature gross margin is the revenue an AI feature brings in, minus the inference it consumes, as a share of that revenue. It is a cross-source metric. Computing it needs product usage, billing, and the cloud or model invoice joined together, and no single system holds all three.
Why does seat-based pricing break when AI features cost per use?
A seat price is fixed while inference cost scales with every request. GitHub said a quick chat question and a multi-hour autonomous coding session can cost the user the same amount, and that absorbing the inference cost behind that usage was no longer sustainable. It moved Copilot to token-based AI Credits on June 1, 2026.
How do you measure the gross margin of a single AI feature?
Join three sources: product analytics for usage by feature and customer, billing for the revenue attached to that usage, and the cloud or model invoice for inference cost. A cross-source metric is one that cannot be computed inside any single system of record. AI feature gross margin is the result of that join.
Should free trials and freemium tiers include AI features?
Include them with a metered allowance sized from cost per active user. Figma doesn’t charge for products in beta and bears the inference cost, and GitHub removed its fallback model so credits and admin budgets govern usage. An unmetered free tier turns AI usage into a cost with no budget behind it.