A SaaS company should start AI on the support tickets it already counts per product, and count the hours freed there against next year's support hiring plan.
Support cost sits in cost of revenue, inside the gross margin a buyer reads first. The 2026 SaaS & AI Performance Benchmarks, published June 1, 2026 by Aleph, which sells financial planning software, and Benchmarkit, a benchmarking firm, draws on full-year 2025 results from 342 B2B SaaS and AI-native software companies, 228 to 232 of them for gross margin. The median software gross margin is 80%, and AI costs have not dented it at the median yet; the blended total-revenue margin, with services and other revenue included, slipped modestly from 77% to 76%, likely transitional cost from integrating AI.
ICONIQ, an investment firm, based the July 2026 edition of its State of AI report on a Q2 2026 survey of about 300 executives at software companies building AI products, AI builders rather than every SaaS company. Spending on AI for internal productivity, including direct and indirect spend, is projected to rise from 11% of revenue in 2025 to 16% in 2026, and headcount growth is uneven by function, with R&D, sales, and product and design expanding while customer support and G&A have contracted. A company about to spend a sixth of revenue on internal AI, while it is already shrinking support, should count the tickets before the tool arrives, or a saving cannot be told from a cut.
A separate survey, of 321 customer service and support leaders worldwide across industries, run September through October 2025 and published April 28, 2026 by Gartner, which sells research and advisory, found 85% expanding human agent responsibilities as AI reduces contact volume, and 31% having implemented or planning AI-driven layoffs through the first quarter of 2027.
In month one the head of support exports a year of tickets by product, category and handle time, and the VP of engineering ties each week's tickets to that week's releases. The CFO ranks categories by tickets times handle time per product, never on a blended total, and the head of support strikes any category without a written answer or needing a change inside the customer's account. In month two AI drafts the reply on the top category of one product, a support agent sends it, and the head of support reads every reply in week one; if closed tickets per support agent fall, the tool widens no further until they recover. In month three the head of support compares closed tickets per support agent per week and the reopen rate on that category with the same category on the untouched products, same months, the release tie explaining any spike, and the CFO sets the hours freed, and the tool's fee, against next year's support hiring plan. A deflection target is met by closing tickets, so the count kept is reopens.
The company pays, out of cost of revenue: the CFO books the tool's fee, whether charged per seat or per resolved ticket, beside support salaries. The general counsel checks customer contracts and data processing terms before ticket text goes into a tool. An operating partner can ask every portfolio company for the same four numbers, tickets per product, handle time, tool fee and support hiring plan, and a company that reports one blended margin has to split it first.
ICONIQ also finds that almost half of companies say their AI agents still need a human to step in on at least 30% of tasks, and the most common failure is multi-step workflows that break partway through.
