Agentic AI vs Generative AI vs RPA: Which Automation Approach Is Right for Your Australian Business? product guide

The GenAI Paradox: Why Most Australian Businesses Are Getting Automation Wrong

Australian business leaders are spending more on AI than ever before — and getting less back than they expected. Nearly eight in ten companies report using generative AI. Nearly eight in ten report no significant bottom-line impact. McKinsey's AI unit QuantumBlack has a name for this: the "GenAI paradox." And it's not a technology problem. It's a category selection problem.

The organisations stuck in this paradox deployed the right tools for the wrong jobs. They rolled out enterprise copilots and chatbots — powerful assistive tools — into workflows that demand autonomous, end-to-end execution. The result? Individual productivity gains that evaporate before they hit the balance sheet, and a growing sense that the AI investment thesis is broken.

It's not. But Australian decision-makers need a clearer map of the automation options before they can navigate them. This article provides exactly that: a structured, side-by-side analysis of the three dominant automation approaches — Robotic Process Automation (RPA), Generative AI (copilots and chatbots), and Agentic AI — across the dimensions that matter most to Australian organisations: autonomy, integration depth, total cost of ownership (TCO), time-to-value, and regulatory risk under APRA CPS 230 and the Privacy Act 1988.

The goal isn't to declare a winner. It's to help you match the right technology tier to the right workflow — before you evaluate specific use cases or deployment pathways (see our guide on Agentic AI Use Cases Across Australian Industries).

Understanding the three automation approaches

Before comparing them, it's worth locking in precise definitions. These three categories get conflated in vendor marketing constantly — and that confusion is itself a source of costly decision-making error.

Robotic Process Automation (RPA)

RPA is enterprise automation software that uses bots to handle predefined, rule-based processes. In plain terms: it's rule-based task execution, best suited for structured data and predictable workflows.

RPA operates as a non-intrusive layer that integrates with existing applications, letting organisations modernise workflows without replacing their tech stack. By bridging legacy and modern systems, RPA accelerates digital transformation, improves process visibility, and helps organisations extract more value from existing technology investments.

The key constraint: when processes involve exceptions, require decision-making, or need to coordinate across multiple systems, RPA bots break or need human intervention. The technology works well for what it's designed to do — automate repetitive tasks — but it hits a wall fast when workflows get dynamic or complex.

Generative AI (copilots and chatbots)

Generative AI tools — Microsoft 365 Copilot, ChatGPT Enterprise, purpose-built chatbots — function as intelligent assistants. They augment human decision-making by generating content, summarising information, answering questions, and drafting outputs. Critically, they are reactive: they respond to human prompts but don't initiate action, execute multi-step workflows, or persist toward goals across sessions without human direction.

At the heart of the GenAI paradox is an imbalance between horizontal (enterprise-wide) copilots and chatbots — which have scaled quickly but deliver diffuse, hard-to-measure gains — and more transformative vertical (function-specific) use cases, about 90 percent of which remain stuck in pilot mode.

Agentic AI

Unlike generative AI tools that respond to prompts, agentic AI systems initiate action. These agents operate toward defined goals, interacting with APIs, databases, and sometimes humans, with limited oversight.

The core architectural difference: RPA says "do these steps exactly." Agentic AI says "achieve this goal however you can."

For a deeper treatment of how agentic systems are architecturally distinct from both generative AI and RPA, see our companion piece What Is Agentic AI? A Plain-English Explainer for Australian Business Leaders.

Side-by-side comparison: the dimensions that matter to Australian decision-makers

Autonomy level

In plain terms: RPA follows rules. Agentic AI pursues goals.

Integration depth

RPA integrates at the UI layer — it mimics mouse clicks and keystrokes. That makes it easy to deploy against legacy systems, but brittle the moment interfaces change. Licensing sits around AUD $5,000–$20,000 per bot per year depending on platform and tier. Stack on script maintenance engineers, downtime costs during bot failures, retraining after every UI update, and the workaround integrations your team builds to catch what bots miss — and total cost of ownership inflates 2–3x beyond the sticker price.

Generative AI tools integrate at the application layer, typically through APIs or native plugins (e.g., Microsoft 365, Salesforce). They can surface information across systems but can't write back to them autonomously or orchestrate cross-system workflows.

Agentic AI integrates at the process layer. At the core is an LLM "brain" that interprets instructions and reasons about approaches. This connects to various tools — APIs, databases, RPA bots, and enterprise applications. Memory stores retain context from previous interactions, allowing the system to build on past conversations. Planners break goals into steps and adjust when obstacles appear. Orchestration layers coordinate multiple agents and human handoffs.

Total cost of ownership: 2025 Australian market context

TCO comparisons in automation are routinely distorted by vendor pricing sheets that leave out maintenance, governance, and failure-recovery costs. The benchmarks below reflect realistic 24-month TCO for Australian mid-market enterprises (200–2,000 employees), adjusted for Australian labour rates and data-residency requirements.

RPA (24-month TCO estimate)

• Licensing: AUD $8,000–$28,000 per bot per year (UiPath, Automation Anywhere, Blue Prism tiers)
• Implementation and process mapping: AUD $30,000–$80,000 per deployment
• Ongoing maintenance (script updates, UI changes): 20–40% of implementation cost annually
• True TCO inflator: stack on script maintenance engineers, downtime costs during bot failures, retraining after every UI update, and workaround integrations — and total cost of ownership inflates 2–3x beyond the sticker price.
• Best-fit ROI window: 6–18 months for high-volume, stable processes

Generative AI / Copilots (24-month TCO estimate)

• Licensing: Microsoft 365 Copilot at AUD $45–$55 per user per month; enterprise LLM API costs vary by volume
• Implementation: Low — typically weeks, not months
• Productivity capture: Highly variable. Gartner's Developer Productivity Analysis (2025) shows that while 25% of time is theoretically saved, 20–80% leaks through coordination and task-switching, and 30–70% of value goes unharvested through lack of cost conversion — yielding 1–14% actual gain in real-world deployments.
• True TCO inflator: the real total cost of ownership, including prompt engineering time, compliance reviews, monitoring infrastructure, and human-in-the-loop QA, inflates initial vendor quotes by 200–400%.

Agentic AI (24-month TCO estimate)

• Platform/build cost: while a basic single-agent deployment might start at USD $15,000, enterprise-grade multi-agent systems routinely exceed USD $150,000. In AUD terms, mid-market deployments typically range from AUD $80,000–$350,000 for the initial build, depending on integration complexity.
• Complex enterprise environments with multiple data sources, APIs, and legacy systems may see integration costs reach 30% of the total project.
• Governance overhead: safety and governance requirements add 20–35% to total agentic AI costs but are non-negotiable for applications where autonomous agents make decisions affecting output, safety, and regulatory compliance.
• ROI profile: companies report average returns on investment of 171%, with U.S. enterprises achieving around 192% — exceeding traditional automation ROI by 3 times. Australian deployments in financial services and logistics have demonstrated comparable trajectories, with payback periods of 12–24 months for targeted, high-value workflows (see our guide on Measuring Agentic AI ROI).

Time-to-value

The time-to-value picture for generative AI is deceptive. Gartner identified "a productivity paradox" in early GenAI deployments: while its use has enhanced individual productivity for desk-based roles, these gains are not cascading through the rest of the function and are actually making the overall working environment worse for many employees.

The average supply chain employee now uses 3.6 GenAI tools. Higher anxiety among employees correlates with lower levels of overall productivity. More tools, more friction — that's not the outcome anyone signed up for.

Regulatory risk under APRA CPS 230 and the Privacy Act

This dimension is uniquely consequential for Australian organisations in financial services, insurance, and superannuation. On 1 July 2025, APRA's Prudential Standard CPS 230 Operational Risk Management came into force. The aim of CPS 230 is to ensure that an APRA-regulated entity is resilient to operational risks and disruptions. Key requirements include: identify, assess and manage operational risks, with effective internal controls, monitoring and remediation.

Entities must also effectively manage the risks associated with service providers, with a comprehensive service provider management policy, formal agreements, and robust monitoring.

How does each automation approach interact with these obligations?

RPA under CPS 230: Relatively low regulatory risk. RPA bots execute deterministic, auditable sequences. Process logs are straightforward. The primary risk is operational resilience — a bot failure in a critical process (e.g., payments processing for an ADI) must be captured within the entity's business continuity plan. APRA mandates that core business operations — including payments, deposit-taking, and customer functions for ADIs; claims processing for insurers; and investment management for RSE licensees — must be classified as "critical." Any RPA bot embedded in these processes must be covered by a tested BCP.

Generative AI under CPS 230: Moderate regulatory risk, driven primarily by hallucination risk and accountability gaps. A shift from long-established techniques to complex and opaque AI techniques creates the risk of unexplainable decisions that may include issues of fairness, bias, and discrimination. There is a need to balance competing risks — such as automated decisions against partly automated and non-automated decisions — business efficiency against consumer risks and harms.

Agentic AI under CPS 230: Highest regulatory complexity, but manageable with the right governance architecture. APRA's CPS 230 sits within Australia's complex and overlapping general legislative, regulatory, and common law obligations that address the use of AI. These include the Corporations Act 2001's directors' duties and a financial services licensee's obligation to provide services "efficiently, honestly and fairly," and the Privacy Act 1988's obligations when collecting, using, and disclosing personal information.

The critical point: an APRA-regulated entity must not rely on a service provider unless it can ensure that in doing so it can continue to meet its prudential obligations in full and effectively manage the associated risks. When an agentic AI system from a third-party vendor is embedded in a critical operation, this obligation applies directly to that vendor relationship.

For a comprehensive treatment of governance architecture for agentic deployments, see our guide on Agentic AI Governance and Compliance for Australian Businesses.

Why the GenAI paradox is an architecture problem, not an AI problem

The evidence is consistent across multiple sources. The enterprise AI situation reveals a stark gap: organisations have committed unprecedented capital to generative AI adoption — between AUD $30 and $40 billion by conservative estimates — yet the transformation promised by this technology remains confined to a remarkably small subset of implementers.

Sixty percent of organisations evaluated enterprise-grade AI systems, but only 20 percent reached pilot stage and just 5 percent reached production.

The root cause isn't the technology. The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. Most GenAI systems do not retain feedback, adapt to context, or improve over time.

This is precisely the architectural gap that agentic systems address. The focus on ROI is pushing enterprises toward more targeted AI applications — particularly autonomous AI in the form of AI agents: software programs designed to collect data and perform self-determined tasks with minimal human oversight. These agents show "more solid potential for productivity and efficiency gains compared to many current GenAI initiatives."

The transition is already visible in adoption data: 25% of companies using generative AI launched agentic pilots in 2025, with that figure projected to double to 50% by 2027. Organisations are recognising that autonomous execution — not just content generation — drives real business value.

The decision framework: which approach is right for your workflow?

Use the following decision logic to match your specific workflow to the right automation tier. This isn't a permanent classification — workflows evolve, and the right answer today may shift as your data estate matures.

Step 1: Characterise the workflow

Ask these diagnostic questions:

1. Is the process fully rule-based with structured inputs? RPA is a strong candidate.
2. Does the process require human judgment, natural language understanding, or content generation? Generative AI (copilot) may suffice.
3. Does the process span multiple systems, involve unstructured data, require exception handling, or need to run end-to-end without human initiation? Agentic AI is required.
4. Does the process involve consequential decisions (credit, claims, patient triage)? Agentic AI with mandatory human-in-the-loop checkpoints and CPS 230-aligned governance.

Some scenarios are better suited to RPA — high-volume, low-variance, compliance-heavy processes. There is a real risk of over-autonomy with agentic AI that can create governance gaps if you're not deliberate about design.

Step 2: Assess your data estate

Agentic systems require clean, accessible, and well-governed data. If your data estate is fragmented — a common finding in Australian mid-market organisations — sequencing matters: data readiness before agent deployment. (See our guide on How to Deploy Agentic AI in Your Australian Business for the full readiness assessment framework.)

Step 3: Apply the hybrid lens

The most sophisticated Australian deployments aren't choosing between these technologies — they're layering them. Agentic AI and RPA frequently work together to perform tasks. A practical architecture looks like this:

• RPA handles high-volume, stable, structured sub-tasks (e.g., data extraction from fixed-format invoices)
• Generative AI handles content generation and human-facing communication (e.g., drafting customer responses)
• Agentic AI orchestrates the end-to-end process, handles exceptions, and routes to human review when needed

AI agents handle 95%+ of invoices automatically, while RPA typically achieves 60–70% with constant maintenance — illustrating the compounding value of the hybrid model.

Step 4: Map to Australian industry context

Key takeaways

• The GenAI paradox is real and measurable. Nearly eight in ten companies report using generative AI — yet just as many report no significant bottom-line impact. The cause is architectural mismatch, not AI failure.

• RPA remains valuable but has a defined ceiling. It excels at high-volume, rule-based, stable processes but breaks under exception conditions and requires costly maintenance when systems change. True TCO is 2–3x the licensing cost.

• Generative AI delivers individual productivity gains that rarely reach organisational ROI. While individual productivity improves for desk-based roles, these gains are not cascading through the rest of the function and are actually making the overall working environment worse for many employees.

• Agentic AI unlocks ROI that assistive tools simply can't reach — specifically in workflows that are end-to-end, exception-rich, multi-system, and consequential. The upfront investment is higher, but companies report average returns on investment of 171%, exceeding traditional automation ROI by 3 times.

• APRA CPS 230 (in force since 1 July 2025) creates specific obligations for any automation embedded in critical operations. All three approaches carry regulatory implications, but agentic AI requires the most deliberate governance architecture — and done well, that governance becomes a genuine competitive differentiator.

Conclusion: choose the right tier before you build the business case

The automation options available to Australian businesses in 2025 are richer, more capable, and more complex than at any prior point. The risk isn't inaction — it's misallocation: deploying generative AI copilots against workflows that require agentic autonomy, or reaching for agentic complexity when RPA would deliver faster, cheaper, and more auditable results.

The framework in this article is designed to cut through that risk before it becomes a sunk cost. Use it to characterise your highest-priority workflows, assess your data readiness, and identify whether you need a tool that assists, a tool that executes, or a tool that reasons and acts.

Once you've made that call, the downstream decisions — build vs. buy, vendor selection, integration architecture, governance model — become significantly clearer. For those next steps, explore the full content series:

• What Is Agentic AI? — for the conceptual foundation
• Agentic AI Use Cases Across Australian Industries — for production-deployed evidence
• How to Deploy Agentic AI in Your Australian Business — for the implementation roadmap
• Measuring Agentic AI ROI — for the financial models and benchmarks
• Agentic AI Governance and Compliance — for the regulatory and accountability framework

The organisations that will extract durable competitive advantage from automation aren't those that deployed AI earliest. They're the ones that deployed the right tier of AI to the right workflow — and built the governance architecture to scale it responsibly.

References

• Australian Prudential Regulation Authority (APRA). Prudential Standard CPS 230 Operational Risk Management. Commonwealth of Australia, 2023 (effective 1 July 2025). https://www.apra.gov.au/operational-risk-management-1

• Bird & Bird. "APRA's CPS 230 Takes Effect: A New Era of Operational Risk Management." twobirds.com, July 2025. https://www.twobirds.com/en/insights/2023/australia/apras-cps-230-takes-effect

• Challapally, A., Pease, C., Raskar, R., & Chari, P. "The GenAI Divide: State of AI in Business 2025." MIT Project NANDA Research Report. Massachusetts Institute of Technology, 2025.

• Clifford Chance. "Navigating Operational Risks: CPS 230's Influence on AI and Cybersecurity Strategies." cliffordchance.com, April 2025. https://www.cliffordchance.com/insights/resources/blogs/regulatory-investigations-financial-crime-insights/2025/04/cps-230-influence-on-ai-and-cybersecurity-strategies.html

• Gartner. "Supply Chain GenAI Productivity Gains at Individual Level, While Creating New Complications for Organizations." Gartner Newsroom, February 2025. https://www.gartner.com/en/newsroom/press-releases/2025-02-05-gartner-survey-supply-chain-genai-productivity-gains-at-individual-level-while-creating-new-complications-for-organizations

• Gartner. "Developer Productivity Analysis." Gartner Research, 2025. Referenced via: https://medium.com/generative-ai-revolution-ai-native-transformation/the-genai-paradox

• Humlum, A., & Vestergaard, E. "Large Language Models, Small Labor Market Effects." National Bureau of Economic Research Working Paper, 2024. Referenced via Penn Wharton Budget Model, September 2025. https://budgetmodel.wharton.upenn.edu/issues/2025/9/8/projected-impact-of-generative-ai-on-future-productivity-growth

• McKinsey & Company (QuantumBlack). "Seizing the Agentic AI Advantage." McKinsey Global Institute, June 2025. Referenced via: https://blog.irvingwb.com/blog/2025/10/the-ai-productivity-paradox.html

• MinterEllison. "CPS 230: Your Roadmap to Compliance." minterellison.com, September 2024. https://www.minterellison.com/articles/cps-230-your-roadmap-to-compliance

• Mordor Intelligence. "Robotic Process Automation Market Size, Trends, Forecast 2026–2031." mordorintelligence.com, January 2026. https://www.mordorintelligence.com/industry-reports/robotic-process-automation-market

• SS&C Blue Prism. "Agentic AI vs RPA — Comparing AI Agents and RPA Bots." blueprism.com, January 2026. https://www.blueprism.com/resources/blog/agentic-ai-vs-rpa-vs-ai-agents-comparing/

• 6W Research. "Australia RPA and Hyperautomation Market (2025–2031)." 6wresearch.com, 2025. https://www.6wresearch.com/industry-report/australia-rpa-and-hyperautomation-market

Frequently asked questions

What is the GenAI paradox? Deploying AI tools for the wrong workflows, yielding no bottom-line impact.

Who coined the term "GenAI paradox"? McKinsey's AI unit QuantumBlack.

What percentage of Australian businesses use generative AI? Nearly 80%.

What percentage of Australian businesses report no significant bottom-line impact from AI? Nearly 80%.

Is the GenAI paradox a technology problem? No.

What type of problem is the GenAI paradox? A category selection problem.

What are the three automation approaches compared in this analysis? RPA, Generative AI, and Agentic AI.

What does RPA stand for? Robotic Process Automation.

How does RPA make decisions? It doesn't — it executes predefined rules only.

What type of data does RPA work best with? Structured data.

What type of workflows suit RPA best? Predictable, rule-based, repetitive workflows.

Does RPA handle exceptions well? No — it fails or requires human intervention.

How does RPA integrate with existing systems? At the UI layer, mimicking mouse clicks and keystrokes.

Is RPA brittle when interfaces change? Yes.

What is the annual licensing cost range for RPA per bot (AUD)? AUD $8,000–$28,000.

What is the implementation cost range for RPA per deployment (AUD)? AUD $30,000–$80,000.

What is the annual ongoing maintenance cost for RPA? 20–40% of implementation cost.

By how much does RPA's true TCO exceed its sticker price? 2–3x.

What is the best-fit ROI window for RPA? 6–18 months for high-volume, stable processes.

How long does RPA deployment typically take per bot? 4–12 weeks.

How long does RPA take to deliver measurable ROI? 3–9 months.

Does RPA scale exponentially? No — each new process requires a new bot.

What is Generative AI in this context? Intelligent assistants like copilots and chatbots.

Does Generative AI initiate actions autonomously? No — it is reactive to human prompts.

Can Generative AI execute multi-step workflows autonomously? No.

Can Generative AI write back to systems autonomously? No.

What is the monthly licensing cost for Microsoft 365 Copilot (AUD)? AUD $45–$55 per user.

How quickly can Generative AI tools be implemented? Typically weeks, not months.

Does Generative AI deliver immediate individual productivity gains? Yes.

Do Generative AI individual productivity gains cascade to organisational ROI? No.

By how much does GenAI's true TCO inflate beyond vendor quotes? 200–400%.

What is the actual real-world productivity gain range from GenAI deployments? 1–14%.

Why do GenAI productivity gains leak? Through coordination overhead and task-switching.

What is the average number of GenAI tools used per supply chain employee? 3.6.

Does using more GenAI tools improve overall productivity? No — it increases anxiety and friction.

What is Agentic AI? AI systems that initiate action to achieve defined goals autonomously.

How does Agentic AI differ from RPA in its core logic? RPA follows rules; Agentic AI pursues goals.

Can Agentic AI handle exceptions? Yes — it self-corrects or escalates with context.

Does Agentic AI have persistent memory across sessions? Yes.

Can Agentic AI orchestrate multiple tools? Yes — APIs, databases, RPA bots, and email.

Can Agentic AI self-improve? Yes — through feedback loops.

What is the initial build cost range for Agentic AI mid-market deployments (AUD)? AUD $80,000–$350,000.

What percentage of total Agentic AI project cost can integration reach in complex environments? 30%.

What percentage does governance overhead add to Agentic AI costs? 20–35%.

What is the average reported ROI for Agentic AI? 171%.

By how much does Agentic AI ROI exceed traditional automation ROI? 3 times.

What is the typical payback period for targeted Agentic AI deployments? 12–24 months.

How long does the first Agentic AI agent typically take to deploy? 8–24 weeks.

How long does Agentic AI take to deliver measurable ROI? 6–18 months.

Does Agentic AI scale exponentially? Yes — agents can orchestrate other agents.

How does Agentic AI integrate with existing systems? At the process layer.

What is APRA CPS 230? Prudential Standard for Operational Risk Management.

When did APRA CPS 230 come into force? 1 July 2025.

Which industries are primarily affected by APRA CPS 230? Financial services, insurance, and superannuation.

What is the regulatory risk level of RPA under CPS 230? Relatively low.

Why is RPA's regulatory risk low? It executes deterministic, auditable sequences.

Must RPA bots in critical processes be covered by a tested BCP? Yes.

What is the regulatory risk level of Generative AI under CPS 230? Moderate.

What drives Generative AI's regulatory risk under CPS 230? Hallucination risk and accountability gaps.

What is the regulatory risk level of Agentic AI under CPS 230? Highest regulatory complexity.

Is Agentic AI's regulatory complexity manageable? Yes — with the right governance architecture.

What percentage of organisations reached pilot stage after evaluating enterprise AI? 20%.

What percentage of organisations reached production stage with enterprise AI? 5%.

What is the core barrier to scaling AI according to the article? Learning — most GenAI systems don't retain feedback or adapt.

What percentage of companies using GenAI launched agentic pilots in 2025? 25%.

What percentage of companies are projected to run agentic pilots by 2027? 50%.

What is the recommended first step in the decision framework? Characterise the workflow.

What is required before deploying Agentic AI if data is fragmented? Data readiness work first.

Can RPA and Agentic AI work together? Yes — in a hybrid architecture.

What percentage of invoices do AI agents handle automatically in hybrid models? 95%+.

What percentage of invoices does RPA typically handle? 60–70%.

What does RPA handle in a hybrid model? High-volume, stable, structured sub-tasks.

What does Generative AI handle in a hybrid model? Content generation and human-facing communication.

What does Agentic AI handle in a hybrid model? End-to-end orchestration and exception handling.

Which Australian industry fits an Agentic AI plus RPA hybrid best? Financial services.

Which automation approach suits healthcare administrative workflows? RPA.

Which automation approach suits professional services knowledge work? Generative AI.

Is Agentic AI suitable for consequential decisions like credit or claims? Yes — with mandatory human-in-the-loop checkpoints.

Does the article declare one automation approach the winner? No.

What is the article's primary goal? Help organisations match the right technology to the right workflow.

Label facts summary

Disclaimer: The information below is extracted from published analytical content, regulatory documents, and market research citations — not product packaging. No physical product label exists for this content. Figures should be independently verified against primary sources before informing procurement or investment decisions.

Verified cited figures

No physical product label, packaging data, ingredient list, nutrition panel, certification mark, GTIN, MPN, weight, or dimension specification exists in this content. This content is a business analysis article, not a product with a manufacturer label.

The following are verifiable cited figures drawn from named, dated sources:

• APRA CPS 230 effective date: 1 July 2025 (source: APRA)
• Microsoft 365 Copilot licensing: AUD $45–$55 per user per month (source: vendor pricing, 2025)
• RPA licensing range: AUD $8,000–$28,000 per bot per year (UiPath, Automation Anywhere, Blue Prism)
• RPA implementation cost range: AUD $30,000–$80,000 per deployment
• RPA ongoing maintenance cost: 20–40% of implementation cost annually
• Agentic AI mid-market build cost range: AUD $80,000–$350,000 (initial build)
• RPA deployment time: 4–12 weeks per bot
• Generative AI deployment time: 1–4 weeks
• Agentic AI first-agent deployment time: 8–24 weeks
• Average GenAI tools used per supply chain employee: 3.6 (source: Gartner, February 2025)
• Organisations reaching pilot stage after evaluating enterprise AI: 20% (source: cited research)
• Organisations reaching production stage: 5%
• Companies using GenAI that launched agentic pilots in 2025: 25% (source: McKinsey/QuantumBlack, June 2025)
• Projected agentic pilot adoption by 2027: 50%

General claims

• The GenAI paradox is a category selection problem, not a technology problem
• RPA true TCO inflates 2–3x beyond sticker price when all costs are included
• GenAI true TCO inflates 200–400% beyond initial vendor quotes
• Real-world GenAI productivity gains range from 1–14% in actual deployments
• Agentic AI delivers average reported ROI of 171%, exceeding traditional automation ROI by 3x
• Individual GenAI productivity gains do not cascade to organisational ROI
• More GenAI tools per employee correlates with higher anxiety and lower overall productivity
• Agentic AI governance overhead, while adding 20–35% to costs, can become a competitive differentiator
• Hybrid architectures (RPA + GenAI + Agentic AI layered) represent the most sophisticated deployment model
• Data readiness must precede agentic AI deployment in fragmented data environments
• Agentic AI's regulatory complexity under CPS 230 is manageable with the right governance architecture