Revenue leaders now get a large share of their “how is AI agent adoption actually going” signal from a single source: the vendor selling the agents. Salesforce’s second Agentic Enterprise Index, published August 7, is the most-discussed example of the genre this month, and the past two weeks of trade coverage make an unplanned case study out of it. Three of the most substantive independent reads of the same numbers, from Futurum Group, CX Today and Diginomica, do not agree on what the data means, and that disagreement is the real story for anyone using vendor usage indices to judge where agentic CRM actually stands.
What Salesforce actually published
The index draws on aggregated usage data from Agentforce, Salesforce’s agent platform, covering the period from February 2025 to April 2026. The headline numbers are large: the average number of activated agents per organization nearly tripled year over year, the time to build and deploy a new agent fell from roughly four days to 1.9 days, and the average skill set per agent grew from two to six. Salesforce also reported that agents now handle 170 times more customer-service conversations than they did at the start of the measurement window, resolving seven in ten of those without human help, and that retailers running agents saw four times higher online sales growth than those that had not deployed them.
Salesforce also broke the growth out by industry. Retail accounted for 22% of total monthly agent output and grew 18 times over the measurement period, financial services grew 13 times, travel grew seven times, and the public sector, from a much smaller base, grew 227 times. Fifty-five percent of organizations named reliability and hallucination management as their top adoption challenge, even as 73% of enterprise software buyers told Salesforce that agentic AI is now a top purchasing priority.
Advertisement
Joe Inzerillo, Salesforce’s president of enterprise AI and technology, framed the scale of deployment as proof of concept rather than experimentation: “Whether spinning up agents at scale or orchestrating deep multistep pipelines, they ship real value.”
PenFed chief information officer Shree Reddy, whose credit union is a named Salesforce customer in the release, offered the closest the index comes to an outcome-based endorsement rather than an activity count: “By pairing robust governance with our unified platform, we’ve safely deployed multi-action agents like Ace and Echo that perform real, complex banking tasks.”
Futurum Group: a paradigm shift in competitive dynamics
Futurum Group took the numbers largely at face value and read them as a market-structure story. Analyst Keith Kirkpatrick’s write-up treated the near-tripling of active agents, the 53% cut in deployment time and the finding that 43% of enterprises now prefer autonomous agents to copilots as evidence that agentic AI has moved from pilot to default deployment mode, with Microsoft, ServiceNow and SAP all racing to match the pace. Futurum’s read is the most vendor-friendly of the three: it treats the index as a leading indicator of where the whole CRM market is headed, not just where Salesforce’s own customers are.
CX Today: adoption is outrunning readiness
CX Today accepted the activation numbers but paired them with a readiness gap the index itself does not resolve. Only 27% of the commerce organizations Salesforce surveyed have fully unified customer data, and only 32% have defined AI success metrics and KPIs at all. CX Today’s framing is that Salesforce is measuring activity, not outcomes, and that a company can triple its agent count while still lacking the data foundation to know whether any of those agents are working. RocketPhone.ai chief executive Muj Choudhury, a former Salesforce director quoted in the piece, argued that enterprises do have a real structural advantage here in the volume of historic data they hold, but the point of his comment cuts both ways: having the data is not the same as having unified it.
Diginomica: the number that undercuts the headline
Diginomica‘s Stuart Lauchlan gave the most openly skeptical read, and he is the only one of the three to flag a data point that complicates Salesforce’s own framing: average skills per agent peaked at 6.4 in November 2025 and had fallen to 3.7 by the time of publication, even as the index’s headline claim is that skill sets are expanding. Lauchlan also noted that the index only counts organizations with continuously activated agents throughout the measurement window, a sampling choice that filters out anyone who tried agentic deployment and gave up, and that the “Agentic Work Unit” metric Salesforce uses to describe output is a metric Salesforce itself invented. He set this against a broader mood of AI-investment skepticism, including MIT research finding that a striking share of enterprise AI spending has not delivered the value expected, and finance teams reportedly startled by unexpected usage-based AI bills.
Where the three accounts actually disagree
All three outlets accept that Salesforce’s raw activation and deployment-speed numbers are real. Where they diverge is on what those numbers are evidence of. Futurum treats scale itself as the signal, competitors are being forced to match Salesforce’s pace, so the market has already shifted. CX Today treats scale as necessary but not sufficient, pointing to the data-readiness gap as the reason activity metrics will not translate cleanly into customer experience gains. Diginomica goes a step further and treats the index’s own internal numbers as partially self-undermining, using Salesforce’s stated decline in average agent skill depth to question whether the “more capable, more numerous” narrative holds up under its own evidence. That is not a minor difference of emphasis. It is three analysts reading the identical dataset and reaching three different verdicts on whether it should reassure or worry a buyer.
What it means for the RevOps and sales leader
The practical lesson is not that Salesforce’s numbers are wrong. It is that a vendor-published adoption index cannot be the only input into a revenue organization’s AI roadmap, because even trained analysts reading the same index cannot agree on what it proves. A RevOps leader benchmarking their own agent rollout against Salesforce’s figures should treat the readiness metrics, unified data, defined KPIs, sustained (not just trial) usage, as the more decision-useful numbers in the report, not the activation counts. Those readiness figures are the ones that predict whether an agent deployment will still be running in a year, and they are also the numbers most sales and RevOps teams can audit against their own CRM instance today without waiting for a vendor’s index to tell them.
The index also arrives alongside a broader pattern this publication has tracked in CRM delivery, where the bottleneck on agentic AI has shifted from the software itself to the partners and processes that implement it. And it lands the same week Gartner’s latest CRM rankings began weighting agentic depth more heavily than seat count, a shift covered here that reinforces the same conclusion: capability claims are proliferating faster than the tools to independently verify them.
How to evaluate the next vendor index
Three questions travel well from this specific dispute to the next vendor-published adoption report a sales or RevOps leader will be handed. First, does the sample include organizations that tried and abandoned the technology, or only those still using it. Second, is the headline metric one an outside analyst could recompute from the vendor’s raw numbers, or is it a proprietary unit the vendor defined itself. Third, does the report separate activity (agents deployed, actions taken) from outcome (revenue attributable, retention improved). Salesforce’s index answers the first and third questions in ways that favor the vendor’s narrative, which is not an accusation of dishonesty so much as a description of what every vendor-funded index will tend to do. The signal worth acting on is the one three independent readers agree on: enterprises are deploying agents faster than they are proving those agents work.
Source: Salesforce
