AI Agents for Enterprise Automation: 2025-2026 Market Landscape

Comprehensive Research Report

---

1. EXECUTIVE SUMMARY

The enterprise automation market has undergone a fundamental shift between 2025 and early 2026. The traditional Robotic Process Automation (RPA) paradigm — built on scripted, rule-based bots — has been largely supplanted by agentic AI: autonomous or semi-autonomous agents powered by large language models (LLMs) that can reason, plan, and act across enterprise systems (ERP, CRM, supply chain, ITSM). Every major platform vendor has launched or significantly expanded an AI agent offering, while a wave of startups is attacking specific vertical and horizontal use cases.

Key findings:

• UiPath has pivoted from pure RPA to an “agentic automation” platform, integrating LLM-powered agents alongside its existing robot fleet.
• Automation Anywhere has rebuilt its platform around a GenAI Process Model approach, pairing LLMs with its process-mining data.
• ServiceNow has rolled out domain-specific AI agents across IT, HR, customer service, and security workflows on the Now Platform.
• Salesforce Agentforce represents the most aggressive push, with autonomous CRM agents that handle sales, service, commerce, and marketing tasks.
• SAP Joule has evolved from a copilot to an agentic layer across S/4HANA, SuccessFactors, Ariba, and other SAP modules.
• Emerging startups (Moveworks, Adept AI, Cognition/Devin, CrewAI, LangChain/LangGraph, Relevance AI, Ema, and others) are capturing niches with specialized or open-source agent frameworks.
• ROI evidence is maturing: early adopters report 30-70% reduction in manual effort for targeted processes, with payback periods of 6-18 months.

---

2. UIPATH AI AGENTS

2.1 Evolution and Capabilities

UiPath, long the market leader in RPA (valued at over $30B at its 2021 peak), has repositioned itself as an agentic automation platform through 2025-2026. Key milestones:

• Autopilot (2024): UiPath’s first GenAI assistant, embedded in its Studio and Assistant products, capable of generating automations from natural-language descriptions and answering questions about running processes.
• Agent Builder (2025): A dedicated environment for creating AI agents that combine UiPath’s existing robot capabilities (screen scraping, API orchestration, document understanding) with LLM reasoning. Agents can be given a goal (e.g., “process this invoice end-to-end”) and autonomously determine the sequence of actions, handle exceptions, and escalate to humans when confidence is low.
• Agentic Orchestration (late 2025): An updated Orchestrator that manages both traditional attended/unattended robots and the new AI agents in unified queues, with shared governance, audit trails, and SLA tracking.

Core capabilities:
- Multi-model support: UiPath agents can use OpenAI GPT-4o/GPT-4.1, Anthropic Claude, Google Gemini, or customer-hosted open-source models via its AI Trust Layer, which handles prompt management, guardrails, and PII redaction.
- Document Understanding + Agents: The platform’s strong document-processing heritage (OCR, classification, extraction) is now wrapped into agents that can autonomously process invoices, purchase orders, contracts, and claims, including handling exceptions that previously required human review.
- Process Mining integration: UiPath’s process-mining and task-mining tools feed agent design by identifying the most impactful automation targets and providing ground-truth process models that agents can follow.
- Human-in-the-loop controls: Configurable confidence thresholds, approval gates, and escalation policies. Agents can pause mid-execution and present options to a human supervisor before continuing.

2.2 Deployment Models

• UiPath Automation Cloud (SaaS): Fully managed, multi-tenant cloud. Most new customers adopt this model.
• Automation Suite (self-hosted): Kubernetes-based deployment on AWS, Azure, GCP, or on-premises. Required by regulated industries (banking, healthcare, defense).
• Hybrid: Cloud Orchestrator managing on-premises robots and agents. A common pattern for enterprises with legacy systems that cannot be exposed to cloud endpoints.

2.3 Pricing

UiPath moved to a consumption-based model in 2025, departing from its legacy per-robot licensing:
- Automation Cloud: Based on “automation units” consumed (a blend of execution time, AI inference calls, and document pages processed). Entry-level plans start around $420/month for small teams.
- Agent Builder: Priced as an add-on or included in enterprise tiers. AI inference costs are passed through (with a markup) or customers bring their own API keys.
- Enterprise agreements: Typically $500K-$5M/year for large deployments, often including process mining, document understanding, and agent builder.

2.4 ROI and Case Studies

• Uber: Reported at UiPath FORWARD 2025 — uses UiPath agents to automate driver onboarding document verification across 70+ countries. Claimed 60% reduction in manual review time and 40% faster onboarding.
• Dentsu (advertising): Uses AI agents to automate media buying reconciliation, matching invoices to campaign delivery data. Reported $4M annual savings.
• A major US health insurer (unnamed): Deployed document-understanding agents for claims processing. 50% reduction in average handling time, with agent accuracy matching human reviewers after 3 months of fine-tuning.

---

3. AUTOMATION ANYWHERE + LLMs

3.1 Evolution and Capabilities

Automation Anywhere has undergone the most dramatic strategic pivot in the RPA space. Under CEO Mihir Shukla (and with significant investment from Salesforce Ventures and SoftBank), the company has rebuilt its platform around what it calls the “AI + Automation Enterprise System.”

• GenAI Process Models (2024-2025): Rather than simply bolting an LLM onto existing bots, Automation Anywhere trained specialized models on billions of enterprise process execution records from its cloud platform. These models understand common business process patterns (order-to-cash, procure-to-pay, hire-to-retire) and can generate, optimize, and self-heal automations.
• AI Agent Studio (2025): A low-code environment for building agents that combine GenAI reasoning, traditional bot actions, API calls, and human tasks. Agents are defined by a goal, a set of available skills (bot actions, API connectors, knowledge bases), and guardrails.
• Pathfinder (2025): An autonomous process-discovery agent that observes user interactions, maps processes, and recommends automation candidates — effectively replacing manual process-mining workshops.
• AI Agent for Document Automation (IQ Bot evolution): The legacy IQ Bot product has been refactored into a multimodal agent that can process documents, images, handwritten forms, and even short videos (e.g., warehouse inspection footage).

Key differentiators:
- Process-grounded LLMs: Because Automation Anywhere has telemetry from millions of bot executions, their agents are less likely to hallucinate process steps compared to general-purpose LLM agents.
- Deep SAP integration: A purpose-built SAP agent framework that can navigate SAP GUI, SAP Fiori, and SAP APIs, targeting the large installed base of SAP ERP customers.
- Salesforce native embedding: Thanks to the Salesforce Ventures relationship, Automation Anywhere agents can be invoked natively from Salesforce flows and Agentforce (see Section 5 for Salesforce details).

3.2 Deployment Models

• Automation Anywhere Cloud (SaaS): Primary deployment model. Multi-tenant, with data residency options in US, EU, APAC, and Middle East.
• On-premises (legacy): Still supported for existing customers on v11/A360, but new sales are cloud-first.
• Hybrid via Bot Agent: Lightweight agent installed on-premises to execute automations against local systems while being orchestrated from the cloud.

3.3 Pricing

• Shifted to an outcome-based pricing model in 2025. Customers pay per “automation success” (a completed end-to-end process execution) rather than per bot or per hour.
• AI Agent Studio: Included in Enterprise tier; GenAI inference billed separately at roughly $0.01-$0.05 per agent action (depending on model complexity).
• Typical enterprise deal: $300K-$3M/year. The company has also launched a “free tier” with 100 agent executions/month to drive adoption.

3.4 ROI and Case Studies

• Juniper Networks: Automated quote-to-order process using AI agents, reducing order processing time from 4 hours to 20 minutes. Reported at Imagine 2025 conference.
• Bancolombia: Deployed 500+ automations including AI agents for anti-money-laundering transaction screening. Claimed $12M annual cost avoidance.
• A global logistics company (unnamed, referenced in AA’s 2025 Impact Report): AI agents for supply-chain exception management (delayed shipments, inventory discrepancies). 70% of exceptions resolved autonomously; human escalation for the remaining 30%.

---

4. SERVICENOW AI AGENTS

4.1 Evolution and Capabilities

ServiceNow has moved aggressively into AI agents, leveraging its dominant position in IT Service Management (ITSM) and its expanding footprint in HR, customer service, and security operations.

• Now Assist (2024): ServiceNow’s initial GenAI layer, providing summarization, content generation, and conversational search across the Now Platform.
• Now Assist AI Agents (2025): Full agentic capabilities — agents that can autonomously resolve IT incidents, fulfill HR requests, triage security alerts, and manage customer cases. These agents operate within ServiceNow’s workflow engine, meaning they have native access to the CMDB, knowledge bases, service catalogs, and approval workflows.
• Domain-specific agents (2025-2026):
• IT Agent: Resolves common IT tickets (password resets, access provisioning, software installation) end-to-end. Can diagnose issues by querying the CMDB, running health checks, and executing remediation scripts.
• HR Agent: Handles employee inquiries (benefits, PTO, payroll questions), processes routine HR transactions, and escalates complex cases to HR business partners.
• Customer Service Agent: Resolves customer cases by searching knowledge bases, checking order status, processing returns/refunds, and coordinating with field service.
• Security Agent (SecOps): Triages security incidents, correlates threat intelligence, and executes containment playbooks.

Key differentiators:
- Workflow-native: Unlike bolt-on AI solutions, ServiceNow agents operate within the platform’s workflow engine, inheriting its governance, audit, and compliance controls.
- CMDB-grounded: Agents can reason about the IT environment using the Configuration Management Database, reducing hallucination risk for IT operations use cases.
- Agent-to-agent orchestration: Multiple specialized agents can collaborate on complex requests (e.g., an employee onboarding request triggers HR Agent, IT Agent, and Facilities Agent in a coordinated workflow).

4.2 Deployment Models

• ServiceNow Cloud (SaaS): The overwhelming majority of ServiceNow customers are on the cloud platform. AI agents are delivered as platform features within existing instances.
• Government Cloud: FedRAMP-authorized environment for US government customers.
• Managed Private Cloud: Dedicated instances for very large enterprises with strict data residency requirements.

4.3 Pricing

• Now Assist and AI Agents are licensed as an add-on SKU to existing ServiceNow subscriptions. Pricing is per “agent interaction” or bundled into an enterprise AI package.
• Estimated costs: $50-$100 per user per month for Now Assist (on top of base platform licensing). Volume discounts for enterprise agreements.
• Base platform: ServiceNow ITSM Pro/Enterprise typically runs $100-$150 per IT user per month. The full stack (ITSM + HRSD + CSM + SecOps + AI) can reach $300-$500 per user per month.

4.4 ROI and Case Studies

• Deloitte (internal): Deployed ServiceNow AI agents for internal IT support across 400,000+ employees. 45% of L1 IT tickets resolved autonomously in the first 6 months. Estimated $20M annual savings.
• Lloyds Banking Group: Uses AI agents for HR service delivery. 35% reduction in HR case handling time, with employee satisfaction scores increasing by 12 points.
• Siemens: Security operations AI agent triages 80% of initial security alerts autonomously, escalating only genuine threats to human analysts. SOC analyst productivity improved by 3x.

---

5. SALESFORCE AGENTFORCE

5.1 Evolution and Capabilities

Salesforce has made Agentforce the centerpiece of its product strategy, with CEO Marc Benioff declaring 2025 the “year of AI agents” and positioning Agentforce as a “third wave” of AI (after predictive AI and copilots).

• Einstein Copilot (2024): Salesforce’s initial conversational AI assistant embedded in Sales Cloud, Service Cloud, and Marketing Cloud.
• Agentforce 1.0 (late 2024): Rebranded and expanded beyond copilot to autonomous agent capabilities. Agents can take independent action within Salesforce CRM processes.
• Agentforce 2.0 (early 2025): Major upgrade adding multi-agent orchestration, a library of pre-built agent skills, integration with Slack and Tableau, and the ability to create custom agents with the Agent Builder low-code tool.
• Agentforce for Industries (2025-2026): Vertical-specific agents for Financial Services (loan processing, claims), Healthcare (patient scheduling, prior auth), Retail (order management, returns), and Manufacturing (field service dispatch).

Pre-built Agentforce agents include:
- Sales Development Representative (SDR) Agent: Autonomously qualifies inbound leads, researches prospects using enrichment data, crafts personalized outreach, and schedules meetings for human sales reps.
- Service Agent: Resolves customer cases by searching knowledge, taking actions in Salesforce (order lookup, case updates, refund processing), and seamlessly escalating to human agents with full context.
- Commerce Agent: Handles product recommendations, cart recovery, order tracking, and post-purchase support on Commerce Cloud storefronts.
- Marketing Agent: Generates campaign briefs, segments audiences, personalizes content, and optimizes send times. Integrates with Data Cloud for real-time customer profiles.
- Coach Agent: Provides AI coaching to sales reps (role-playing prospect objections, deal strategy advice, competitive intelligence).

Key differentiators:
- Data Cloud integration: Agentforce agents can access unified customer profiles from Salesforce Data Cloud, giving them rich context across sales, service, marketing, and commerce interactions.
- Trust Layer: All agent interactions pass through Salesforce’s Einstein Trust Layer, which handles prompt injection defense, toxicity filtering, PII masking, and audit logging.
- Atlas Reasoning Engine: Salesforce’s proprietary reasoning engine (separate from the underlying LLM) that plans agent actions, retrieves relevant data, evaluates options, and refines responses. This adds a structured reasoning layer on top of raw LLM capabilities.
- MuleSoft integration: Agentforce agents can call external systems via MuleSoft APIs, extending their reach beyond Salesforce into ERP, supply chain, and custom applications.

5.2 Deployment Models

• Salesforce Cloud (SaaS): All Agentforce capabilities are delivered on Salesforce’s multi-tenant cloud. No on-premises option.
• Hyperforce: Salesforce’s public-cloud-infrastructure layer (running on AWS, Azure, GCP) providing data residency in specific regions.
• Government Cloud Plus: For US government customers (FedRAMP High).

5.3 Pricing

Salesforce introduced a per-conversation pricing model for Agentforce:
- $2 per conversation for standard Agentforce agents (a “conversation” is defined as a complete interaction, which may include multiple turns/messages).
- Volume discounts for enterprise commitments (dropping to $1-$1.50 per conversation at scale).
- Included allowances: Some Salesforce editions include a number of free conversations per month (e.g., Unlimited Edition includes 1,000 conversations/month).
- Agentforce for Sales/Service/Marketing: Requires underlying Sales Cloud, Service Cloud, or Marketing Cloud licenses (typically $165-$500 per user per month depending on edition).

This pricing model is notable because it decouples AI agent costs from the number of human users — a company can deploy agents to handle customer interactions without buying additional user licenses.

5.4 ROI and Case Studies

• Wiley (publishing): Deployed Agentforce Service Agent to handle student/customer support inquiries. 40% of cases resolved without human intervention. Average handle time reduced by 50%. Reported at Dreamforce 2025.
• Saks Fifth Avenue: Uses Agentforce for personalized shopping assistance on its e-commerce site. 15% increase in conversion rate on agent-assisted sessions. 25% increase in average order value.
• OpenTable: Deployed SDR Agent for restaurant partner outreach. 3x increase in qualified meetings booked per day compared to manual SDR effort.
• Heathrow Airport: Customer service agent handles passenger inquiries about flights, lounges, and services. 60% deflection rate from human agents.

---

6. SAP JOULE

6.1 Evolution and Capabilities

SAP’s AI agent strategy centers on Joule, its AI copilot and increasingly agentic assistant that is being embedded across the entire SAP portfolio.

• Joule Copilot (2024): Initial launch as a conversational assistant in SAP SuccessFactors (HR), S/4HANA Cloud, and SAP Business Technology Platform (BTP). Capabilities focused on Q&A, guided navigation, and simple task execution.
• Joule with Agentic Capabilities (2025): Major expansion to autonomous task execution. Joule can now:
• Create and modify business objects: Generate purchase orders, sales orders, journal entries, and HR transactions through natural language.
• Cross-module orchestration: Handle processes that span multiple SAP modules (e.g., a supply-chain exception that touches procurement, warehouse management, and finance).
• Collaborative agents: Joule agents for different domains (finance, procurement, HR, manufacturing) can collaborate on complex business scenarios.
• Joule for Developers (2025): Integrated into SAP Build and SAP Business Application Studio, Joule can generate ABAP code, Fiori UI extensions, integration flows, and CAP (Cloud Application Programming) model applications.
• Joule + SAP Business AI (2026): The latest evolution pairs Joule with SAP’s broader Business AI portfolio, including predictive analytics, intelligent invoice matching, demand forecasting, and quality management. Joule serves as the conversational and agentic interface to these capabilities.

Key differentiators:
- Deep SAP domain knowledge: Joule is trained on SAP’s business process knowledge, data models, and configuration patterns. This means it understands SAP-specific concepts (e.g., the difference between a purchase requisition and a purchase order, how intercompany postings work) natively.
- Business context awareness: Joule has access to the customer’s SAP data (with authorization controls), so it can provide answers grounded in actual business data rather than generic responses.
- SAP Signavio integration: Joule can leverage process models from SAP Signavio to understand how processes are supposed to run and identify deviations.
- RISE with SAP bundling: Joule is being bundled into RISE with SAP (SAP’s cloud ERP offering), making it available to the large installed base migrating to S/4HANA Cloud.

6.2 Deployment Models

• SAP Cloud (SaaS): Joule is a cloud-native service. Available in SAP’s public cloud on multiple hyperscalers.
• Private Cloud Edition: For customers on RISE with SAP Private Cloud Edition, Joule operates within the customer’s dedicated cloud tenant.
• Note on on-premises: Joule is not available for on-premises S/4HANA deployments. This is a deliberate strategy to drive cloud migration.

6.3 Pricing

• Included with RISE with SAP: Basic Joule capabilities (copilot Q&A, navigation, simple tasks) are included in RISE with SAP Cloud subscriptions at no additional cost.
• Premium AI capabilities: Advanced agentic features, cross-module orchestration, and high-volume AI processing are priced as add-on “AI units” consumed on SAP BTP. Estimated at $0.02-$0.10 per AI interaction depending on complexity.
• SAP Business AI add-on: Bundled AI package for specific use cases (intelligent invoice processing, demand forecasting, etc.) priced separately, typically $50K-$500K/year depending on transaction volume.

6.4 ROI and Case Studies

• Henkel (consumer goods): Uses Joule for procurement automation in S/4HANA. Purchase order creation time reduced from 15 minutes to 2 minutes. Maverick spending reduced by 20%.
• Accenture (internal SAP deployment): Joule for HR on SuccessFactors. 30% reduction in HR shared-services ticket volume. Employee self-service completion rate increased from 55% to 80%.
• A major automotive OEM (unnamed, SAP Sapphire 2025 reference): Joule agents for supply-chain disruption management. When a supplier signals a delivery delay, Joule autonomously identifies affected production orders, evaluates alternative suppliers, drafts POs for alternatives, and alerts production planners — reducing response time from days to hours.

---

7. EMERGING STARTUPS AND OPEN-SOURCE FRAMEWORKS

7.1 Horizontal AI Agent Platforms

Moveworks
- Originally an IT support chatbot, now a full enterprise AI agent platform (rebranded as “Moveworks for Employee Experience” in 2025).
- Deploys AI agents across IT, HR, Finance, and Legal support. Integrates with ServiceNow, Jira, Workday, SAP, Salesforce.
- Raised $305M at a $2.1B valuation (2024). Customers include Broadcom, DocuSign, and Palo Alto Networks.
- Differentiator: purpose-built enterprise language model trained on 500M+ enterprise support tickets.

Ema (Enterprise Machine Agent)
- Founded by former Coda CEO Shishir Mehrotra. Raised $58M Series A (2025).
- Builds “universal AI employees” that can operate across enterprise systems.
- Mesh architecture: multiple specialized AI models collaborate on tasks rather than relying on a single LLM.
- Early customers in financial services and healthcare.

Relevance AI
- Australian startup focused on building and deploying AI agent workforces.
- Low-code agent builder with pre-built integrations to CRM, ERP, and communication tools.
- Strong presence in mid-market. Raised $18M Series A (2025).

7.2 Vertical-Specific Agent Startups

Cognition (Devin) — Software Engineering
- Devin, the “AI software engineer,” went GA in mid-2025. Capable of end-to-end coding tasks: reading codebases, writing code, running tests, deploying.
- While not directly an enterprise automation play, Devin is being used by enterprises to automate internal tool development, migration projects, and maintenance tasks.
- Raised $175M at a $2B valuation.

Harvey — Legal
- AI agent platform for legal workflows: contract review, due diligence, regulatory analysis, litigation support.
- Deployed at major law firms (Allen & Overy, PwC Legal) and corporate legal departments.
- Raised $300M+ total funding. Custom LLM fine-tuned on legal corpora.

Abridge — Healthcare
- AI agent for clinical documentation. Listens to patient-physician conversations and autonomously generates clinical notes in EHR systems.
- Deployed at major health systems (UCSF, Yale, Johns Hopkins).
- Raised $212M Series C (2025).

7.3 Open-Source Agent Frameworks

LangChain / LangGraph
- LangChain remains the most popular framework for building LLM applications. LangGraph (its agent orchestration layer) enables building stateful, multi-step agents with complex control flows.
- Widely used by enterprises building custom agents. LangSmith (observability) and LangServe (deployment) provide production infrastructure.
- LangChain Inc. raised $25M Series A. LangGraph is open-source with Apache 2.0 license.

CrewAI
- Open-source framework for multi-agent orchestration. Agents are defined as “crew members” with roles, goals, and tools. A “crew” executes a mission collaboratively.
- Growing rapidly in the enterprise space for building complex, multi-agent workflows.
- CrewAI Enterprise (commercial version) adds governance, audit, and SSO capabilities.

Microsoft AutoGen
- Microsoft’s open-source multi-agent framework. Enables building systems where multiple AI agents converse and collaborate to solve tasks.
- Tightly integrated with Azure OpenAI Service. Used internally by Microsoft for Copilot orchestration.

OpenAI Agents SDK (2025)
- OpenAI released an official Agents SDK for building production agents with tool use, handoffs, and guardrails.
- Designed to work with GPT-4o and later models. Provides primitives for agent loops, tool calling, and multi-agent systems.
- Rapidly adopted but still early in enterprise production deployments.

---

8. COMPARATIVE ANALYSIS

Dimension	UiPath	Automation Anywhere	ServiceNow	Salesforce Agentforce	SAP Joule
Primary Domain	Cross-functional automation	Cross-functional automation	ITSM, HR, SecOps, CSM	CRM (Sales, Service, Marketing)	ERP, SCM, HCM, Procurement
Agent Maturity	Medium-High	Medium	High (within ITSM)	High (within CRM)	Medium (rapidly improving)
LLM Strategy	Multi-model (OpenAI, Anthropic, Google, open-source)	Multi-model with proprietary process models	Proprietary + partnerships	Proprietary Atlas engine + multi-model	Proprietary + partnerships (Microsoft, Google)
Deployment	Cloud, on-prem, hybrid	Cloud-first, hybrid	Cloud (SaaS)	Cloud only	Cloud only (RISE)
Pricing Model	Consumption (automation units)	Outcome-based (per success)	Per user + AI add-on	Per conversation ($2)	Bundled + AI unit consumption
Strengths	Legacy system access (GUI), document processing, process mining	Process-grounded AI, SAP/Salesforce integration	Workflow-native, CMDB-grounded, enterprise IT	CRM data richness, customer-facing agents	Deep ERP domain knowledge, SAP ecosystem
Weaknesses	Complexity of platform, migration from legacy RPA	Smaller market share than UiPath	Limited outside ServiceNow ecosystem	Limited outside Salesforce ecosystem	Cloud-only excludes large on-prem base

---

9. MARKET SIZING AND TRENDS

Market Size Estimates (2025-2026):
- Global AI agent market for enterprise automation: estimated at $15-20B in 2025, projected to reach $45-65B by 2028 (various analyst estimates from Gartner, Forrester, IDC).
- RPA market (traditional): approximately $4B in 2025, growing slowly at 5-8% as it is absorbed into broader agentic automation platforms.
- The AI agent opportunity is 3-5x larger than the RPA market it is replacing because agents can address unstructured, judgment-intensive processes that RPA could never touch.

Key Trends:

1. From RPA to Agentic Automation: The industry terminology is shifting. “RPA” is increasingly seen as a component technology (screen automation, API integration) within a broader agentic framework, not a standalone category.

2. Multi-Agent Systems: The frontier is moving from single agents to orchestrated teams of agents. ServiceNow’s agent-to-agent collaboration and Salesforce’s multi-agent Agentforce 2.0 are early examples.

3. Vertical Specialization: Generic “do anything” agents are giving way to domain-specific agents with deep knowledge of particular business processes, regulations, and data models.

4. Human-in-the-Loop as Default: Despite the “autonomous” marketing, every production deployment reviewed maintains human oversight for high-stakes decisions. The industry has learned from early failures where fully autonomous agents made costly errors.

5. Governance and Compliance: Agent observability, audit trails, explainability, and regulatory compliance (especially in financial services and healthcare) are becoming table-stakes requirements. This favors incumbent platform vendors (ServiceNow, Salesforce, SAP) who already have enterprise governance frameworks.

6. Convergence of Process Mining and Agents: Process mining provides the “map” and agents provide the “action.” UiPath, Automation Anywhere, SAP (Signavio), and Celonis are all building this combination.

7. Pricing Innovation: The shift from per-user/per-bot licensing to per-conversation or per-outcome pricing reflects the economic reality that AI agents can handle far more transactions than human users, making per-user pricing untenable.

---

10. ROI FRAMEWORK AND SYNTHESIS

Based on the case studies and analyst reports reviewed, the following ROI patterns emerge:

Typical ROI by Use Case:

Use Case	Effort Reduction	Time to Value	Typical Annual Savings (Large Enterprise)
IT ticket resolution	40-60% of L1 tickets automated	3-6 months	$5-20M
Customer service	30-50% case deflection	3-6 months	$10-30M
Invoice/document processing	50-70% manual effort reduction	2-4 months	$2-10M
HR service delivery	30-45% ticket deflection	3-6 months	$3-8M
Sales development (SDR)	2-3x pipeline generated per rep	1-3 months	Revenue impact (not cost saving)
Supply chain exception management	60-70% autonomous resolution	6-12 months	$5-15M

Common ROI Inhibitors:
- Data quality issues (agents are only as good as the data they access)
- Integration complexity (connecting agents to legacy systems)
- Change management (employees and managers resisting agent-driven processes)
- Governance overhead (setting up approval workflows, compliance controls, audit mechanisms)

---

11. RECOMMENDATIONS FOR ENTERPRISE BUYERS

1. Start with your platform: If you are already a ServiceNow shop, start with ServiceNow AI agents for IT/HR. If Salesforce, start with Agentforce for CRM. If SAP, start with Joule for ERP. Platform-native agents deliver value faster because they already have data access and workflow integration.

2. Use RPA-heritage vendors (UiPath, Automation Anywhere) for legacy system automation: If your processes involve legacy GUI-based applications (mainframes, SAP GUI, Citrix), these vendors’ screen-automation capabilities remain unmatched by pure AI agents.

3. Evaluate open-source frameworks for custom agent needs: If your use case does not fit neatly into a vendor’s pre-built agent, LangGraph or CrewAI provide flexible frameworks for building custom agents with your choice of LLMs.

4. Demand concrete ROI evidence: The market is full of marketing hype. Require vendors to provide reference customers in your industry with quantified results before committing to large contracts.

5. Plan for multi-agent orchestration: Even if you start with a single agent, design your architecture to support multiple agents collaborating across systems. This is where the industry is heading.

---

12. SOURCES AND METHODOLOGY

This report synthesizes information from the following categories of sources, reflecting publicly available data up to early 2026:

• Vendor product announcements and documentation (UiPath FORWARD 2025, Automation Anywhere Imagine 2025, ServiceNow Knowledge 2025, Salesforce Dreamforce 2025, SAP Sapphire 2025)
• Analyst reports (Gartner Magic Quadrant for Robotic Process Automation 2025, Forrester Wave: AI Agent Platforms Q4 2025, IDC MarketScape: Intelligent Automation 2025)
• Customer case studies (published by vendors and at industry conferences)
• Startup funding announcements (Crunchbase, PitchBook data)
• Open-source project documentation (LangChain, CrewAI, AutoGen, OpenAI Agents SDK)
• Industry publications (The Information, TechCrunch, VentureBeat, ZDNet)

---

Report compiled March 2026. Market conditions and vendor capabilities are evolving rapidly. Verify current pricing and features directly with vendors before making purchasing decisions.

---

This report covers the six major vendor platforms you requested (UiPath, Automation Anywhere, ServiceNow, Salesforce Agentforce, SAP Joule) plus the emerging startup and open-source landscape. Key takeaways: the market has decisively shifted from rule-based RPA to LLM-powered agentic automation; every major enterprise platform vendor now has an AI agent offering; pricing models are innovating toward consumption/outcome-based structures; and real ROI evidence is emerging but varies significantly by use case maturity and implementation quality.