New Study of 2,025 Agentic AI Leaders: First To Launch Isn’t Fastest to ROI

Agent deployments more than doubled over the past year, according to Salesforce's platform data, and those agents are driving real results. Retailers running AI agents grew online sales at four times the rate of those that didn't. For the many organizations now deploying their first agents, that raises a sharper question: Among those already seeing returns, what sets them apart? Salesforce's State of Agentic AI in the Enterprise, a global survey of 2,025 agentic AI decision makers, points to preparation rather than pace. Companies that deployed first weren't necessarily the first to reach meaningful ROI. Rather, operational factors (e.g., having clean, well-governed data available to agents at their time of need; clearly defined agent scope) were most predictive of success.

Other key results: Among the 30% already running agents in production, deployments reach ROI in about eight months, with an employee adoption rate of 53% and a 29% average lift in customer satisfaction.

"Every boardroom is asking whether it's moving fast enough. Two years into the agentic shift, the answer from the data is that the advantage was never in starting first; it's in starting deliberately. The organizations getting real returns got specific about a shortlist of things before conditions were perfect: the data they made trustworthy for the job, the point where a person stays in the loop, and the guardrails they built before they needed them." - Shibani Ahuja, SVP, Data & AI Strategy, Salesforce

Dig deeper

The industries fastest to ROI aren't always the ones leading in deployment

If speed of deployment drove returns, the earliest and largest deployers would also be the fastest to ROI. They aren't.

• The industries furthest along in their AI agent deployment are different from the industries reaching ROI fastest - evidence that being first to adopt doesn't guarantee being first to returns.
• Two of the industries with the smallest share of companies that have graduated to full deployment - Professional & Business Services and Supply Chain & Logistics - attained meaningful ROI the fastest.
• Conversely, the High Tech industry is one of the biggest deployers of AI agents but posts one of the slowest times to ROI at 10.1 months.

Perfect data isn't a prerequisite for deployment

Data readiness doesn't separate those who have deployed AI agents from those who haven't, though it correlates with better outcomes once deployed.

• Deployers credit two factors above the rest for the success of their most autonomous agents: clean, accessible data at the moment an agent acts and a tightly bounded use case. Neither requires fully unified data across the business.
• In fact, only 31% of deployers fully unified their data before launching AI agents. The other 69% were still integrating sources, working around gaps, or operating on fragmented data.
• However, organizations that unified relevant data before deploying their AI agents achieved meaningful ROI sooner than those who deployed agents first then addressed data infrastructure gaps afterward - 7.3 months versus 8.8.

"People think they need to boil the ocean - get all their data perfect in one place before starting. What we're finding is you can go use case by use case: get the data accurate, mechanized, and semantically described so agents understand what it is and how to use it. That semantic layer is what unlocks the value." - Joe Inzerillo, President, Enterprise & AI Technology, Salesforce

Agents deliver more when they're embedded in the workflow, not bolted on

Where an agent lives shapes what it can reach.

• The average surveyed organization runs 58 separate business applications, yet fewer than half (42%) have AI natively embedded across them.
• Employees use AI regularly at an incrementally higher rate where it's natively embedded (55%) than where it's connected but outside core systems (47%).
• Ninety-four percent of deployers say embedding AI into core workflows delivers more value than running it as a stand-alone tool.

Governance tends to arrive late - and there's a real speed-versus-durability tradeoff

Most deployers went live with some oversight and built the rest afterward.

• Deployers averaged two governance structures - such as real-time monitoring, escalation frameworks, and audit logs - before deploying agents, and three after.
• That sequence tracks with speed: Organizations with lighter oversight reached positive ROI in 7.2 months, versus 9.3 months for those with heavier governance.
• Organizations with below-average governance were nearly twice as likely to discover an agent operating outside of their parameters only after a consequential error occurred - 32% versus 18% of those with above-average governance.
• For many, this is a genuine tradeoff: Launch faster with lighter governance in place, and you're likely faster to ROI but slower to catch what goes wrong.
• More than a third of respondents whose AI initiatives slowed, stalled, or failed (38%) named stronger governance frameworks and escalation protocols among what they'd do differently.

"Good enough to learn is the bar. Define guardrails sensibly, but don't overbuild them to the point where you kill the innovation." - Joe Inzerillo, President, Enterprise & AI Technology, Salesforce

Trend in action: Asymbl

Asymbl, a workforce orchestration company, experienced rapid growth in 2025, with candidate volume outpacing the recruiting team and applicant records scattered across disconnected systems. So it built Rosa, a digital recruiter agent that drafts job descriptions, screens candidates, schedules interviews, and sends offers. Here's the stack that makes it work: Rosa operates inside Slack, where recruiting already works. Underneath, Data 360 provides a centralized knowledge foundation so Rosa runs on consistent, trusted records rather than scattered ones. MuleSoft pulls in Github, Hira, and Google Drive so recruiters aren't hunting across systems mid-pipeline. The result? Asymbl hired 100 people in 100 days with Rosa's support, at 96% retention.

Explore further

• Read the full State of Agentic AI in the Enterprise report
• Become an Agentic Enterprise: A Refreshed Step-By-Step Guide

Methodology: Data are from a double-blind survey of 2,025 agentic AI decision-makers across 20 countries on five continents, fielded May 14-28, 2026. All respondents influence AI agent purchasing decisions. Respondents are grouped as deployed (30%), piloting (47%), or evaluating (23%); most findings reflect the 30% that have fully deployed ("deployers"). All outcome metrics - including time to ROI and the 29% customer-satisfaction lift - are self-reported. Unless otherwise noted, figures reflect the full respondent base; percentages are rounded and may not sum to 100%.