IBM Reports AI Work Now 35% of Consulting Backlog as Enterprises Move to Production

IBM's backlog signals AI has entered enterprise operating budgets

IBM reported that AI-related work represented 35% of its consulting order backlog in Q3 2026, up from 30% of consulting revenue in the same quarter. The company posted $17.1 billion in quarterly revenue, up 5% year over year, with software revenue reaching $7.8 billion, up 8%. The backlog figure is not a projection or pipeline estimate—it represents contracted work.

This matters because backlog measures committed spend, not vendor claims about demand. Enterprise buyers can use IBM's 35% figure as a benchmark when evaluating competing vendors' AI adoption narratives. A vendor claiming high AI interest without corresponding backlog, revenue recognition, or contracted professional services may be counting pilots that will not convert to production.

IBM competes with Accenture, Deloitte, Capgemini, Microsoft, AWS, Google Cloud, and other systems integrators building AI implementation practices. The backlog share suggests IBM is monetizing deployment and integration work—data modernization, governance, process redesign, application refactoring—not merely reselling model access. Buyers should expect AI services to consume a growing share of transformation budgets, with implementation costs exceeding inference charges.

SAP positions AI as the forcing function for cloud migration

SAP reported Q3 2026 revenue exceeding €10 billion for the first time and disclosed that AI was the main driver behind customers moving to the cloud. The company is embedding Business AI into core ERP applications and positioning its RISE with SAP program as the migration path for existing on-premises customers.

This creates a bundled decision for SAP customers. Cloud migration, ERP modernization, data residency requirements, and AI functionality may be negotiated together rather than purchased separately. Buyers should separately price migration services, AI entitlements, usage-based charges, and the cost of retaining on-premises functionality. Failing to unbundle these components during procurement makes it difficult to evaluate competing offers from Microsoft Dynamics 365, Oracle Fusion Cloud Applications, Salesforce, Workday, or specialized process-automation vendors.

SAP's strategy increases the importance of application data access and makes standalone AI tools less attractive for finance, procurement, supply-chain, and HR workflows. If the AI functionality is embedded in the ERP rather than consumed through an external API, the buyer's leverage shifts from choosing the best AI model to negotiating application licensing terms.

Ungoverned agents are producing measurable financial and operational risk

A survey published October 22 found that 77% of enterprises experienced at least one negative consequence from an ungoverned AI agent. The most common effects included unexpected cloud costs, operational disruption, and compliance issues.

The finding strengthens the market for agent-governance and observability products from Microsoft, AWS, Google Cloud, IBM, Salesforce, ServiceNow, and specialist security vendors. It also challenges the assumption that application vendors provide sufficient controls by default. Agent deployments should require explicit controls for permissions, spend ceilings, action approval, rollback, audit logs, and incident response. Procurement teams should budget for governance and monitoring as recurring infrastructure costs, not optional professional services.

A separate benchmark, UndoBench, reported that enterprise AI agents completed 83.54% of tasks under normal conditions but recovered correctly from mid-execution faults in only 46.72% of cases. A normal-path success rate near 84% is insufficient for high-impact workflows if recovery succeeds less than half the time. Buyers should require failure-injection tests covering API outages, stale data, permission changes, duplicate actions, and human rejection before authorizing production deployment.

AI purchasing remains fragmented, with budgets spread across multiple models

An October survey of 389 executives found that 41% described AI spending as net-new money, while 31% reported having no clear AI budget. Among organizations charging for AI, 43% used usage-based pricing, compared with 24% using flat subscriptions and 24% using outcome-based pricing. Reported project budgets were concentrated below $1 million: 41% were under $100,000 and 22% were between $100,000 and $1 million.

Usage-based pricing favors hyperscalers and model providers, while subscription pricing favors application vendors such as Microsoft, Salesforce, SAP, ServiceNow, and Workday. Finance teams cannot safely apply traditional per-seat SaaS budgeting to agentic systems. Contracts should include usage caps, rate-card protections, budget alerts, cost allocation by workflow, and clear treatment of model or tool-call charges.

Microsoft plans to make Microsoft 365 Copilot and Copilot Business available without a qualifying Microsoft 365 base subscription starting December 1, 2026. The new offers retain the same price while receiving new product and SKU identifiers. Standalone availability lowers the entry barrier against Google Workspace with Gemini, ChatGPT Enterprise, Claude for Work, and independent enterprise AI assistants. Procurement teams should review renewal timing, reseller SKUs, eligibility rules, and whether standalone licenses create overlapping spend with existing Microsoft 365 Copilot entitlements.

What to watch

The buying signal this week is a shift from experimentation to operationalization: AI is appearing in IBM's backlog, SAP's cloud-migration motion, Microsoft's licensing structure, and enterprise risk controls. The most consequential buyer issue is no longer whether vendors offer AI—it is whether AI entitlements, usage charges, governance controls, and failure recovery are contracted with the same rigor applied to other production infrastructure. Buyers who treat AI as a pilot program will find themselves negotiating production terms after deployment, when leverage is weakest.