Salesforce’s woes underline marketing’s agentic AI problems

Salesforce’s Agentforce rollout has become a case study in the gap between agentic AI’s promise and the operational reality inside most enterprises. When Marc Benioff declared the company “all in on Agentforce” at its 2024 launch, the narrative was clear: autonomous agents would handle service, sales, and marketing workflows without human hand‑holding. Eight months later, only 34 percent of Salesforce’s 150,000 customers have adopted the platform, and the vendor has shed more than $200 billion in market value. The question for marketers is not whether agentic AI works in a demo — it is whether their own data and processes can support it.

The adoption gap

KeyBanc Capital Markets and Bernstein both downgraded Salesforce on the same day, an unusual convergence that signals deep skepticism. KeyBanc’s research puts the active Agentforce base at roughly 23,000 organizations, most of them running proof‑of‑concept projects rather than production deployments. A CIO survey conducted by the firm found more respondents planning to cut Salesforce spend over the next year than increase it. Partners report that Agentforce deals are only now entering the pipeline, and even those are scoped to narrow use cases such as automated case routing or lead enrichment. The pattern mirrors what happened with Einstein GPT a year earlier: heavy marketing, light enterprise rollout.

Data readiness is the bottleneck

Agentic AI agents need clean, connected, and contextualized data to make decisions — assign a case, trigger a nurture flow, rewrite a subject line. Most CRM instances fall short. Records are duplicated across clouds, contact roles are inconsistent, and critical fields like purchase intent or lifecycle stage are either missing or maintained in spreadsheets outside the platform. Marketing teams often discover that the “agent‑ready” data set they thought existed is a patchwork of custom objects, legacy integrations, and manual exports. Preparing that foundation consumes as much engineering time as building the agent itself, a reality that Benioff’s launch keynote glossed over.

Product maturity still lags

Beyond data, the agent framework itself is immature. Agentforce’s reasoning engine, prompt orchestration, and guardrails for compliance are still evolving. Early adopters report that agents hallucinate next‑best‑action recommendations when fed incomplete opportunity histories, and that handoff logic between service and sales agents breaks when custom validation rules fire. Salesforce’s own roadmap shows critical features — multi‑agent coordination, audit trails, and granular permissioning — slated for the second half of 2025. Until those land, most marketing ops leaders will keep agents in sandbox environments, using them for simple tasks like auto‑reply drafting rather than end‑to‑end campaign execution.

What this means for marketing

For CMOs and marketing operations heads, the lesson is blunt: agentic AI will not fix a broken data model. If your lead‑to‑account matching relies on fuzzy email matching, an autonomous agent will amplify the error at scale. If your segment definitions live in a BI tool that the CRM cannot see, the agent cannot personalize in real time. The path to value runs through data governance — master data management, identity resolution, and a single source of truth for customer intent — before any agent is deployed. Marketing teams that invest in those foundations now will be the ones able to move from pilot to production when the platform matures.

Wall Street’s verdict

The market has already priced in the delay. Salesforce shares sit more than 50 percent below their December 2024 peak, erasing the $200 billion premium investors assigned to Agentforce as the next growth engine. Analysts are not questioning the long‑term thesis — autonomous agents will eventually run large swathes of front‑office work — but they are doubting Salesforce’s ability to monetize it before competitors like Microsoft’s Copilot Studio or HubSpot’s Breeze AI reach parity. The vendor’s pricing model, which bundles Agentforce credits into existing editions, also makes it hard to isolate revenue attribution, further clouding the investment case.

The road ahead

Salesforce’s next earnings call will be watched for two signals: the conversion rate of proof‑of‑concept projects into paid deployments, and any shift in net‑new ARR attributable to Agentforce. For marketing buyers, the pragmatic play is to treat agentic AI as a 2026‑2027 capability, not a 2024 quick win. Prioritize data hygiene, standardize handoff protocols between marketing automation and CRM, and run controlled pilots that measure time‑to‑value rather than feature checklists. The agents are coming. The enterprises that prepare their house will be the ones that actually use them.