What 300 software executives reveal about scaling AI products

40% of companies with $500M or more in revenue are actively deploying AI agents, with another 35 to 45% running pilots. Agents are being granted read and write permissions across infrastructure and developer tools. They are making autonomous decisions at scale. (Onspring)

The measurement gap this creates is significant. Most of these deployments have system metrics: latency, uptime, token consumption. They do not have task-level visibility into which agent decisions users trust and which they override or ignore.

This matters because agent trust is not uniform. Users will trust an AI agent to draft a first response. They may not trust it to make a pricing decision or handle a sensitive customer escalation. Knowing where trust breaks down, by task type, by user segment, by department, is what makes it possible to deploy agents confidently rather than speculatively.

Cost management requires task-level data

AI gross margins are projected to reach approximately 52% on average in 2026, up from 41% in 2024. The improvement is real. The pressure on margins from inference costs is also real. (Vectra AI)

Model inference has become the dominant variable cost as AI products scale. Teams running multi-model strategies, which the report shows averaging 3.1 providers per company, route workloads across models to balance cost, latency, and reliability. But without task-level data, that routing is based on system performance, not business outcomes.

The practical problem: a team can optimize inference costs and still be spending disproportionately on tasks that generate low business value, while underinvesting in the tasks that drive retention and revenue. Cost efficiency at the infrastructure level does not automatically produce cost efficiency at the business level. Only task-level visibility connects the two.

Pricing is moving toward outcomes

37% of companies plan to change their AI pricing model in the next 12 months. The primary drivers are customer demand for consumption-based pricing (46%), competitive pressure (39%), and the need to demonstrate outcomes rather than just activity. (Onspring)

Companies experimenting with outcome-based pricing most often tie outcomes to cost savings (36%) or revenue generated (18%). This reflects a broader shift in how enterprise buyers are evaluating AI: not on capability, but on proven business impact. (Vectra AI)

For enterprise AI teams, this shift has a direct implication. Customers are increasingly asking for evidence that AI has delivered the productivity gains or revenue contribution it was purchased to produce. Teams that can answer that question with task-level data, showing which workflows improved, how much time was saved, and what revenue was influenced, are in a fundamentally stronger position than those pointing to usage counts and latency figures.

What this means in practice

The ICONIQ report describes a market in transition. The competitive advantage that came from being an early AI adopter is eroding. The new differentiator is execution: the ability to scale AI reliably, measure its impact accurately, and prove its value to leadership and customers alike.

Three things separate the companies that are scaling from those that have stalled at pilot stage.

Task-level visibility. Not aggregate usage, but which specific tasks are succeeding and which are failing, and why. This is the measurement infrastructure that makes improvement systematic rather than reactive.

Automated evaluation. Behavioral signals that surface failure patterns before users complain, so teams can act while the impact is still limited rather than after adoption has already declined.

ROI connection. Linking task success to the business outcomes that matter: productivity gains by department, cost reduction by workflow, revenue influenced by customer-facing AI interactions. This is what moves the conversation from "the system is running" to "the investment is paying off."

The 49-point gap between the 88% of enterprises that have deployed AI and the 39% that report measurable financial impact is not a technology gap. It is a measurement gap: most organizations have not built the infrastructure to connect AI activity to business outcomes. The ICONIQ data shows that the companies building that infrastructure are the ones widening the distance between themselves and the rest. (Nextiva)

Nebuly

Nebuly is the ROI platform for enterprise AI. It connects to the AI agents your business runs on, the assistants your customers interact with, and the tools your employees use every day, including Claude, ChatGPT, and Copilot, and translates that activity into business value. How much time is being saved across teams. What revenue your AI is influencing. What adoption and AI proficiency look like in practice, across departments and geographies. All aggregated at the organizational level, never tied to individuals.

If you need clarity on what your AI investment is actually delivering, book a demo.