92% of Enterprises Will Increase AI Budgets—But Half Cannot Measure ROI

Budget Growth Without Proven Returns

Ninety-two percent of enterprise AI decision-makers expect budgets to grow over the next twelve months, yet nearly half cannot identify a measured return on investment, according to a September 22 Futurum Research report published with Google Cloud. Only 18.5% of surveyed organizations rated their AI maturity above the midpoint on a five-point scale. The gap between spending commitments and operational readiness signals that procurement criteria are shifting from model performance toward deployment capabilities, governance tooling, and ROI measurement.

The Futurum study, From Pilots to Production: Why Agentic Transformation Runs Through the Ecosystem, identified agentic AI—systems that take multi-step actions with minimal human oversight—as the fastest-rising enterprise software priority. The finding favors vendors that can deliver integration frameworks, observability, security controls, and proof of business value over those selling foundation models alone. Google Cloud competes in this broader platform market with Microsoft Azure AI, Amazon Web Services, Salesforce, ServiceNow, IBM watsonx, and Oracle.

AI Funding Remains Unsettled Across Finance and IT

Forty-one percent of finance-function respondents in the Open Future Forum's September 2026 AI Transformation Report said net-new money was the largest source of AI funding. Twenty-eight percent reported having no clear AI budget, and 20% funded AI by reallocating existing software spend. Among finance teams in the largest cohort, 71% already used Claude or another AI tool, 22% were evaluating options, and 8% had not started.

The data shows that budget ownership is distributed across finance, IT, and business units. Vendors requiring a separate AI allocation may face approval friction. Anthropic's Claude is competing for enterprise deployments with OpenAI ChatGPT Enterprise, Microsoft 365 Copilot, Google Gemini for Workspace, and proprietary tools from Salesforce and ServiceNow. Competition is increasingly occurring inside existing application and finance workflows rather than only at the model layer.

Buyers should evaluate whether proposed AI spending is genuinely incremental or simply displacing licenses, contractors, or existing automation platforms. The 28% without a clear AI budget indicates that approval processes and chargeback models remain immature.

Procurement AI Moves From Drafting to Negotiation

NPI and Zip announced a partnership on September 24 to integrate NPI's Price Pulse benchmarking capability into Zip's AI procurement-orchestration platform. The integration uses an MCP-based connection so Zip's Procurement Superagent can call NPI's IT-price benchmarking data inside a procurement workflow and return pricing insights in real time. NPI described itself as the launch member of Zip's AI Data Partner ecosystem and the selected partner for large-enterprise IT-deal benchmarking.

The move positions Zip against procurement and sourcing platforms adding AI capabilities, including Coupa, SAP Ariba, ServiceNow, Ivalua, and Oracle Procurement. The differentiator is access to external benchmark data that can influence negotiated prices, not simply conversational assistance or document drafting.

This development could shift enterprise procurement AI from workflow routing toward automated negotiation support and real-time price validation. Buyers will need to evaluate the provenance, freshness, coverage, and contractual rights associated with benchmark data. An autonomous agent making pricing recommendations can affect supplier relationships and spend commitments if approval thresholds are poorly configured, creating a control risk that procurement leaders must address before deployment.

Inference Pricing Falls 41%, Improving Deployment Economics

The Ramp AI Index reported that effective pricing for AI inference fell 41% to $0.68 per million tokens from a 2026 peak of $1.15 per million tokens in March. Lower effective prices intensify competition among model and cloud providers, including OpenAI, Anthropic, Google, Meta, Amazon, Microsoft, and specialized inference providers. Price competition also makes smaller or open-weight models more viable for high-volume enterprise workloads.

The decline improves the economics of internal search, customer support, coding assistance, document processing, and agentic workflows. However, token price alone is not the full cost. Buyers must also account for retrieval infrastructure, data egress, observability, human review, security, and integration. The pricing trend makes volume deployments easier to justify, but it does not by itself demonstrate business value.

What to Watch

The widest gap is between AI budget growth and measured production value. Buyers are likely to prioritize platforms that combine models with workflow integration, governance, pricing intelligence, and ROI measurement. The strongest near-term competitive pressure is shifting from model benchmark scores to which vendor can deploy, control, and economically operate AI across existing enterprise processes.

Organizations funding AI pilots without clear ownership, approval processes, or return metrics are at risk of accumulating spend without operational savings or revenue. The maturity data suggests that many enterprises are still in the experimentation phase while committing to production-scale budgets. Buyers should require vendors to specify how their tools integrate with existing finance systems, how they measure usage and outcomes, and what controls prevent autonomous agents from making unchecked spending or operational decisions.