AI has moved from experiment to income statement. ICONIQ's latest State of AI survey shows that software companies building AI products expect those products to make up a growing share of revenue, with better margins and more usage-based pricing. For B2B leaders, the numbers are a useful benchmark for pricing, hiring and where to invest next.
The report, ICONIQ's 2026 State of AI: The Builder's Economy, is based on a Q2 2026 survey of roughly 305 executives at software companies building AI products. SaaStr pulled out ten headline metrics. We group them into five themes and add what we think each one means in practice.
AI is becoming the majority of revenue
Respondents said AI products made up 32% of their revenue in 2025, are projected at 42% in 2026 and are heading towards 53% in 2027. In other words, for this group of builders, AI stops being an add-on and becomes the core business within two years.
That has a knock-on effect for buyers. If your vendors are rebuilding their products around AI, expect pricing, packaging and roadmaps to change quickly. If you are a vendor, expect customers to compare you against AI-native competitors, not the incumbents you used to benchmark against.
Related: Unlocking the Power of Website Analytics for B2B Leaders: Best Practices for Founding Teams (2026)
Margins are improving, and pricing is shifting to usage
- Gross margin on AI products: 45% in 2025, projected 53% in 2026 and 59% in 2027.
- Consumption-based pricing: rose from 35% to 42% of respondents in six months.
- Outcome-based pricing: rose from 18% to 23%.
- Blended models: companies now use 1.7 pricing models on average.
Improving margins mean the cost of running models is falling faster than prices. The shift to consumption and outcome pricing means vendors increasingly want to charge for the work done rather than the seats occupied. For buyers, that is good news when usage is predictable and a budgeting headache when it is not. Ask every AI vendor for a usage forecast and a cap before you sign.
Builders use several models, and vertical apps lead
Licensed frontier model APIs are used by 80% of respondents, down from 85%, while open-source use rose from 37% to 40% and 37% run proprietary in-house models. On average, builders use 3.3 model providers. Vertical AI applications make up 43% of what is being built, ahead of horizontal apps at 20% and consumer at 11%, with financial services and healthcare climbing fastest.
Two lessons follow. First, multi-model is the norm, so no product should depend on a single provider. Second, the biggest opportunity is in AI that understands one industry's data and workflow deeply, which is harder to copy than a general assistant.
Related: Unlocking the Secrets of Anonymous Website Visitors: A Guide for B2B Leaders (2026)
Agents are the priority, but not yet autonomous
Two-thirds of respondents (66%) rank agentic AI as their top investment priority for the next 12 months. Yet only 5% say their agents rarely need human intervention. The gap between ambition and reality is large.
That matches what most teams find when they deploy agents in sales and support: they are excellent at drafting, triage and research, and still need a person to approve anything that touches money, contracts or a customer relationship. Plan agent rollouts with a human review step and remove it only where the error rate is proven low.
Organisations are getting flatter
- 78% plan a different mix of roles or smaller teams.
- 72% of AI-driven companies run with one to four layers between the CEO and junior staff.
- 50% plan to scale up forward-deployed engineers, and 76% of those engineers' variable pay is tied to retention or renewal.
- High-growth companies report 48% productivity gains from coding assistants, compared with 32% for others.
The forward-deployed engineer figure stands out. Builders are admitting that AI products need hands-on help to land inside customers, and they are paying for it based on whether the customer stays. Among companies with more than $500M in revenue, data readiness also jumped from 4% to 22%, a sign that larger firms are finally cleaning up the data AI depends on.
What this means for B2B sales and marketing
- Revisit your pricing. If competitors are moving to usage or outcome pricing, test a component that scales with customer value.
- Fix your data first. Agents and scoring models are only as good as the CRM and intent data behind them.
- Keep humans on high-stakes steps. Let AI draft, research and prioritise; let people send the important messages until you trust the output.
- Tie customer-facing roles to retention. The builders surveyed are doing it for engineers; it works for success teams too.
- Benchmark productivity honestly. Measure time saved on real tasks, not tool adoption.
At Happierleads, the same thinking shaped our own product flow: identify website visitors, qualify them against an ideal customer profile, and only then engage. Clean, qualified data at the top makes every automated step after it more reliable. If you want to see how that works on your own traffic, the pricing page includes a free trial.
How to use these benchmarks
Remember who was surveyed: executives at software companies building AI products. Their projections are optimistic by design, and a typical B2B services firm or manufacturer will not match them. Use the figures as a direction of travel, not a target. The useful questions are whether your pricing, team structure and data readiness are moving the same way, and how far behind you would be if they are.
The takeaway
ICONIQ's numbers describe a market where AI is becoming the product, margins are improving and pricing is moving towards usage and outcomes, while agents still need human oversight. For B2B leaders, the practical moves are to get your data in order, test pricing tied to value, and deploy agents where a person can still check the work.
