Salesforce’s own State of Sales report for 2026, based on a survey of more than 4,000 sales professionals, paints an unambiguous picture of adoption. AI, and increasingly AI agents specifically, is now the top tactic sales teams say they are relying on to drive growth this year. According to Salesforce’s announcement of the findings, 87% of sales organisations now use some form of AI, 54% of individual sellers report having used an agent directly, and adoption is expected to reach nearly nine in ten sellers by 2027. Those are striking numbers by any measure. The more useful question for anyone managing a licensing budget is what they actually mean for return on investment, and the honest answer is more nuanced than the adoption headline alone suggests.
The productivity numbers are real and specific
Unlike a lot of AI adoption data, which tends toward vague sentiment, Salesforce’s figures here are unusually concrete. Sellers using AI tools report saving an average of twelve hours per week, with research time per prospect dropping by 34% and email drafting time dropping by 36%. In one internal example cited in coverage of the report, Salesforce used its own AI agents to work leads that no human rep had capacity for, contacting 130,000 leads over four months and generating 3,200 new opportunities that would otherwise have gone untouched in the CRM. That level of specificity, according to an independent breakdown of the report’s key data points, is unusual for a vendor commissioned survey, and it is worth noting explicitly that the underlying respondent base skews toward Salesforce’s own customer population, so the numbers should be read as a strong directional signal from within the ecosystem rather than a fully independent, vendor neutral benchmark.
Where the productivity story runs into a harder ROI question
Here is where the picture becomes considerably more complicated, and considerably more relevant to anyone deciding how much of next year’s budget to allocate toward AI agent licensing. A separate, independent survey of enterprise IT decision makers found that the pure productivity argument for AI agent ROI is losing ground as the primary success metric. According to Futurum Research’s own analysis of enterprise decision makers, productivity gains fell from 23.8% to 18.0% as the top ranked measure of AI success among IT decision makers between successive halves of the year, while direct financial impact metrics, combining top line revenue growth and bottom line profitability, nearly doubled to 21.7% of the top ranking over the same period. In plain terms, the people actually approving AI agent budgets are shifting away from asking whether an agent saves time and toward asking whether it moves revenue or profit, and that is a materially harder bar to clear.
That shift is reflected starkly in a separate and quite candid piece of research specific to the Salesforce ecosystem. According to IBM’s State of Salesforce 2025-26 report, based on a broad survey of Salesforce customers, only 33% of AI initiatives are currently meeting their expected ROI, 62% of organisations report concern about unpredictable AI related costs, and only 21% of respondents strongly agree they have the governance structures in place to manage agentic AI responsibly. Perhaps most tellingly for licensing teams specifically, 74% of Salesforce customers surveyed say they are still struggling to move the needle on customer experience and engagement outcomes, despite the scale of AI investment already underway.
The gap between adoption and measured return
This is the gap that matters most for anyone sitting on the licensing and procurement side of an AI agent decision. Adoption numbers, unmeasured cost, and confidence in governance are three genuinely different things, and the State of Sales figures speak primarily to the first of those, while the IBM data speaks far more directly to the second and third. A broader review of AI ROI measurement across sales and marketing functions makes a related point worth internalising, arguing that unmeasured ROI is, structurally, no ROI at allbecause a benefit that cannot be measured cannot be defended in a future budget cycle, cannot be scaled deliberately, and cannot be optimised with any confidence. The organisations capturing genuinely strong return from agent licensing are, by that same analysis, not simply the ones deploying the most agents, but the ones that built measurement infrastructure into the programme from the very start, rather than bolting a measurement exercise on afterward to justify a renewal.
There is also a wider market context worth keeping in view when evaluating how much weight to put on any single vendor’s adoption statistics. Salesforce’s own financial results this year have shown Agentforce revenue crossing the one billion dollar annualised run rate mark with triple digit year over year growth, yet the company’s stock performance and full year guidance have drawn a more cautious reaction from financial markets, with coverage of the company’s most recent quarterly results noting that full year revenue guidance came in slightly below Wall Street expectations even as Agentforce itself outperformed. That combination, strong product level adoption alongside more measured overall financial guidance, is a reasonable proxy for the same tension showing up inside individual customer organisations, strong usage signals sitting alongside genuine uncertainty about aggregate return.
Governance spend deserves its own line item, not an afterthought
One figure from the IBM research deserves particular attention from anyone building next year’s Salesforce budget. With only 21% of respondents strongly confident in their agentic AI governance, and nearly three in four saying digital labour raises a genuine need for stronger risk management, there is a clear implication for how AI agent licensing should actually be budgeted. Governance, monitoring, and change management capacity around AI agents should be treated as a real, planned cost alongside the licence fees themselves, rather than as a soft, deferred concern to be addressed once problems surface. Organisations that treat governance as an afterthought tend to discover the cost of that decision later, in the form of an agent that behaved unpredictably in production, a data quality issue that only became visible after several months of agent activity, or a renewal conversation where nobody can clearly articulate what value the current deployment has actually delivered.
A useful discipline here, borrowed from how mature organisations already treat new hires, is to define success criteria for a new AI agent deployment before it goes live, not after. What is this agent meant to accomplish, what does the performance baseline look like before deployment, and what does success look like at 30, 90, and 180 days. Framing an agent deployment this way turns a vague productivity narrative into something a licensing team can actually track against, renew with confidence, or decline to expand if the evidence does not support it.
A simple framework for reading any vendor’s AI adoption statistics
Given how central AI adoption statistics have become to nearly every software vendor’s messaging this year, not just Salesforce’s, it is worth having a consistent internal framework for reading them rather than evaluating each new report from scratch. Three questions are worth asking of any vendor published AI statistic before it informs a budget decision. First, whose population was actually surveyed, since a vendor’s own customer base will naturally skew toward more engaged, higher adopting users than the market as a whole, which tends to inflate headline adoption figures relative to a fully independent sample. Second, is the metric being reported a usage metric, such as hours saved or tasks completed, or a financial outcome metric, such as revenue growth or cost reduction, since these are genuinely different things and only the latter maps directly onto a licensing ROI decision. Third, is there an independent, non vendor source corroborating the finding, even directionally, since a claim that shows up consistently across a vendor’s own report, an independent analyst firm, and a customer facing survey carries meaningfully more weight than one that appears in a single vendor commissioned study alone.
Applying that framework to the data discussed above, the underlying adoption trend, that sales teams are turning to AI agents in large and growing numbers, holds up reasonably well across multiple independent sources, including Futurum’s separate enterprise decision maker survey. The financial ROI picture is considerably less settled, and the IBM data in particular, drawn from a broad Salesforce customer base rather than a narrow group of top performers, is a useful corrective to any assumption that adoption automatically implies return. Both pictures are true at once, and a licensing budget built on only one of them will be incomplete.
A short list of questions worth bringing to your next agent renewal
To make this concrete, five questions are worth having answered in writing before any AI agent licensing renewal or expansion goes to sign off. What specific, measurable business outcome has the current deployment produced against a defined baseline, not simply a usage statistic. What would the same team have achieved without the agent, as a rough comparison point. What is the current monthly cost trend for consumption based components such as Flex Credits or per conversation fees, and how confident is the team in projecting that forward twelve months. What governance review, if any, has been applied to the agent’s outputs since it went live. And finally, what happens to the workflow and the people currently supported by the agent if the licence is not renewed, since a clear answer to that question is often the most honest signal of how essential the deployment has actually become.
What this means for building next year’s agent licensing budget
For licensing and procurement teams, the practical takeaway is not to be sceptical of AI agent adoption data, which is genuinely strong and worth taking seriously, but to insist on separating usage metrics from financial return metrics before committing further budget. Before expanding any AI agent licensing commitment, ask for the specific, measurable outcome the current deployment has produced against a defined baseline, not simply the hours saved or conversations handled. Ask what governance structure exists to catch cost overruns before they compound across a full contract year, given that 62% of the IBM survey’s own respondent base named unpredictable cost as an active concern. And treat a vendor’s own adoption statistics, however impressive, as a starting point for due diligence rather than a substitute for your own internal measurement of whether the specific agents you have licensed are actually producing the return your organisation needs them to produce.
