Software companies that charge separately for AI grew faster last year than those that fold it into existing plans, but the extra growth has not yet shown up as higher net retention, according to the 17th annual software benchmark from KeyBanc Capital Markets, released on Nov. 13.
KeyBanc, the corporate and investment banking unit of KeyCorp, has run the benchmark for 17 years under the name SaaS Survey. This year it renamed it the AI x Software Benchmark and widened it to cover how AI changes product strategy, monetization, unit economics and operating models at private software companies. Companies that charge separately for AI products reported median ARR growth of 27% in 2025, against 17% for those that do not.
The findings that follow come from KeyBanc’s release. The release does not say how many companies took part or how the sample was weighted, so the figures describe KeyBanc’s respondents rather than the whole software market.
Advertisement
Where the growth and the margins sit
KeyBanc split respondents into three groups by how central AI is to what they sell. Companies selling AI as their primary product reported median ARR growth of 34% in 2025. Companies embedding AI in broader products reported 22%, and companies still developing customer-facing AI reported 9%.
The order flips on profitability. Core AI product companies reported a median EBITDA margin of negative 12% in 2025, against negative 1% for embedded AI companies. Embedded AI companies also led on revenue per employee, with ARR per full-time employee of $201,000 against $117,000 for core AI product companies.
AI revenue is growing fast from a small base
Respondents expect AI-driven ARR to grow 128% in 2026, up from 74% in 2025. Even so, AI would account for only 12% of total ARR. Overall median ARR growth was 24% in 2025 and is projected at 23% in 2026, while participants are planning for growth to accelerate to 40% in 2027. The overall median EBITDA margin is expected to move from negative 3% in 2025 to 1% in 2026.
“AI is accelerating growth and compressing development cycles, but AI spend is running ahead of budget and the economics are still catching up,” said Scott Peterson, Managing Director, Software Investment Banking at KeyBanc Capital Markets. “2026 is a proving year for AI-driven revenue.”
AI spend is landing in gross margin
Forty-four percent of respondents reported AI spend over budget so far this year, and 24% were more than 20% over. Higher-than-expected token usage was the driver they cited most. Companies spending more than 5% of revenue on AI reported a median gross margin of 66% in 2025, compared with 78% for companies spending less than 2%.
The benchmark also tracks how AI is changing the way these companies build and staff. Ninety-two percent of respondents said AI is shortening product development cycles, and 91% use AI internally in engineering. Peterson said companies are reinvesting developer productivity gains “to ship more product faster.” Sixty-five percent reported no AI-related headcount reduction or reallocation in 2025. Looking 12 to 24 months out, that share falls to 34%, with R&D (45%) and customer service (37%) named as the functions most likely to be affected.
Agents are still early
Only 20% of respondents offer fully autonomous agents to their customers. Fifty-seven percent said their customers are building alternative products with AI coding tools.
What it means for the sales leader
Most readers of this publication sit on one of two sides of these numbers. Some run sales at software companies deciding how to package AI. Others buy software from those companies and negotiate the AI line in the contract.
If you sell software with AI in it
In our reading, the 27% against 17% growth gap supports pricing AI as its own line, and the retention finding is the warning attached to it. KeyBanc found that the growth advantage has not yet translated into higher net retention, so a separate AI price can lift new bookings while leaving the renewal case unproven. Sales leaders who carry a net retention target should ask customer success for usage and outcome data on the AI add-on before the first renewal cycle arrives, and should keep the AI line visible in renewal forecasts instead of netting it into the core subscription.
The cost side matters for deal desks too. When 44% of companies report AI spend over budget and token usage is the main cause, discounting on an AI add-on eats into a margin that is already lower. We covered the same pressure when Salesforce, Five9, SAP and Microsoft moved to sell contact-center AI as an add-on, and in Anthropic’s IPO filing, which prices by usage. Approval rules for AI discounts deserve their own thresholds.
If you buy software with AI in it
The same data gives buyers a negotiating position. Vendors that sell AI as a separate line are growing faster, and 44% of respondents are spending above their AI budget, so buyers should expect pressure to move AI from a fixed fee to metered use. We have argued that sales leaders should ask agent vendors to price on work done and stop paying for AI agents by the seat they do not fill. The KeyBanc figure that only 20% of respondents offer fully autonomous agents is a useful check in any demo: ask the vendor which tasks its agent completes with no person approving each step.
The finding that 57% of respondents see customers building alternatives with AI coding tools also changes renewal talks. A buyer with an internal build option has a stronger hand, and sellers should expect that option to come up in procurement conversations.
What to do next
Revenue operations leaders can take three steps from this report. First, split AI revenue from core subscription revenue in pipeline and renewal reporting so the retention effect KeyBanc has not yet found can be measured in your own book. Second, set deal desk approval thresholds for AI discounts that reflect the lower gross margin on AI-heavy products. Third, track how often buyers raise an internal build alternative in late-stage deals and log it as a loss reason in the CRM.
The full 2026 AI x Software Benchmark is available from KeyBanc at key.com.
