While we are now seeing organisations investing heavily in AI, 70% report minimal or no tangible returns. Interestingly, 70% seems to be the common ratio for failure. Digital transformation suffers a similar chance of failure – most often due to poor planning and lack of a clear vision. 

Indeed, many IT initiatives struggle to progress beyond pilot stages, often remaining isolated from core operations and lacking a clear, measurable impact.

The introduction of Agentic AI, which features autonomous goal-setting and decision-making capabilities, signals an imminent major transformation in organisational structures. But like the transition to cloud, the implementation of an ERP, or any other common digital transformation exercise, vision and sound strategy is crucial for maximising the chances of success.

But because this isn’t our first rodeo, dear reader, let’s acknowledge the mistakes of the past and prepare for a well-oiled transformation to enterprise Agentic AI.

What Is Agentic AI? 

Agentic AI goes beyond task-specific tools and passive assistants. Agentic systems achieve objectives through reasoning processes and autonomous actions while improving their responses based on feedback loops. Think of them not as tools, but as new forms of digital co-workers that operate across functions: Agentic AI systems may perform tasks such as contract reviews, content generation, customer sentiment analysis, ticket triaging and team coaching along with other functions.

Agentic systems bridge multiple departments and workflows by working collaboratively and adapting continuously, whereas traditional AI systems function in isolation. These systems demonstrate intelligence in a practical way that allows them to perform meaningful actions.

As an example of such a platform, Decidr.ai is an Australian-made Agentic AI company offering such a solution. Piquing my interest is their approach to providing interconnected Agents that can be deployed across department knowledge stores and systems: I see this as being the most significant and fundamental change in business processes and efficiency.

Imagine prompting your enterprise Agent like so…

• Find the contract documents relating to client ‘A’ (Agent integration with enterprise file storage)
• How much did client ‘A’ spend last FY? (Agent integration with finance accounting system)
• Are there any outstanding orders for client ‘A’ that have not yet been dispatched? (Agent integration with ERP)
• List the last 3 customer tickets (Agent integration with customer service system)
• Who is our main contact at client ‘A’ (Agent integration with CRM)
• Call ‘main contact’ and record the call (Agent invokes Teams call and records conversation)
• Summarise the call, generate call notes and any follow up actions (Agent saves call recording, notes and creates activities/tasks in CRM)

… With all interactions available from a single Agent prompt, the efficiencies are obvious as in many organisations this type of information is siloed, rarely centrally available, and may require significant training for staff to translate operational processes to systems usage. We could take this a step further and set the above as a standard customer service process with the Agent prompting the user with the next best action. 

In the near future, I also envisage Agentic UIs where the above may be dynamically and intelligently rendered into a single pop-up interface so that users can view the retrieved data in a structured interface rather than simply listed in a prompt. The UI may dynamically change depending on the type of data being viewed. 

Anyway, back to reality. 

Why Now?

Three key drivers establish Agentic AI as both practical and essential for modern organisations.

Explosion in AI capabilities

Massive investments in LLMs and reinforcement learning have enabled rapid maturation of autonomous AI’s technical foundation.

Horizontal inefficiencies

Digital transformation efforts have primarily targeted vertical organisational silos. Agentic AI provides leaders with a method to link data and functions through systems.

Economic necessity

Organisations face unprecedented pressure to cut operating expenses while maximising strategic benefits from limited resources. Agentic AI makes it possible. 

The Risks of Getting It Wrong 

Signs of fragmentation are already evident in the enterprise AI landscape. The proliferation of single-task AI tools leads to the same outcome we’ve seen in SaaS: Organisations face hidden maintenance costs and integration challenges because AI subscriptions overlap and workflows become disconnected.

AI systems hold significant risk when their decisions carry actual commercial implications. Inadequate AI implementation results in hallucinations along with data governance issues that require expensive rework.

Data privacy and security are the other common risks. 

But we need to start somewhere. Transformation and innovation cannot occur without experimentation while appreciating the known (or unknown) risks.

A Roadmap to Horizontal Agentic AI 

Introducing Agentic AI throughout the organisation requires more than IT expertise because it represents a fundamental transformation of business operations. It’s an operating model transformation. CIOs can follow this structured method to achieve scalable and sustainable results.

1. Start with Real Business Outcomes 

Identify the particular business functions that require enhancement through Agentic AI before choosing a tool or vendor. These often include: 

• Sales: prospect triage, pipeline enrichment 
• Marketing: content generation, competitor analysis 
• Operations: task orchestration, forecasting 
• HR: onboarding automation, candidate evaluation
• Customer service: autonomous Agents for triage and resolution 

2. Design for Horizontal Intelligence 

Avoid deploying isolated vertical Agents. Instead, focus on systems that: 

• Share context across departments 
• Learn continuously from a unified data layer 
• The system allows for extensions without needing to reconstruct the primary logic for each new use case.

Horizontal AI enables a smaller number of Agents to perform superior work through increased capabilities.

3. Build a Data-First Foundation 

The quality of Agentic AI depends entirely on the access it has to data. Establish: 

• Develop a unified data warehouse that supports AI operations and maintains vector data storage.
• Operational systems such as CRM, ERP, and CMS need to maintain real-time (or near real-time)synchronisation.
• Effective data governance with role-based access combined with audit trails

Your business needs this structure to function as its AI nervous system. Consider Microsoft Fabric as a possible solution to building a centralised source of truth.

4. Enable Human + AI Collaboration 

Agentic AI should serve to enhance human capabilities rather than replace them. Ensure: 

• Human-in-the-loop controls for critical decisions 
• Interfaces that are usable by non-technical staff 
• Role-specific oversight of Agent behaviour and outcomes 

This prevents the problem of decisions being made in an opaque process.

5. Run a Pilot—But Think Scale 

This is the most important part. Begin by selecting a use case that yields high-value results within a short timeframe. For example: 

• An AI recruiting system to screen candidates and generate offers
• A marketing assistant conducts competitor research to deliver campaign recommendations.
• An automated customer service Agent handles basic customer questions without human intervention.

Assess impact levels before improving governance, which allows planning expansion into nearby departments.

Case in Point: Klarna’s AI Assistant 

As an example of Agentic AI in practice, Klarna, a Swedish fintech company, launched an AI-powered customer service assistant through its partnership with OpenAI. During its initial month online, the AI assistant successfully handled two-thirds of Klarna’s customer service interactions, totalling 2.3 million chats. The performance matched what 700 full-time Agents would handle and resulted in major operational efficiencies.

Key outcomes included: 

Enhanced Efficiency

The AI assistant improved resolution times from 11 minutes to under 2 minutes, which resulted in higher customer satisfaction and enabled human Agents to handle more complex assignments.

Improved Accuracy

The accuracy of errand resolution increased, which resulted in a 25% decline in repeat inquiries and demonstrated better first-contact resolution rates.

Global Reach

The AI assistant supported operations in 23 markets while managing more than 35 languages to better connect with diverse customer groups, including immigrant and expat communities.

Financial Impact

The AI assistant’s efficiency gains will lead to a $40 million profit improvement for Klarna in 2024.

Strategically applied Agentic AI demonstrates its capacity to transform business operations. Interestingly, Klarna’s CEO – Sebastian Siemiatkowski – stated last year that they were replacing their Salesforce SaaS with Agentic AI. While his position has since been clarified, it begs the question as to the approach IT departments will need to take towards enterprise-wide technology enablement.

What to Look for in a Platform 

When evaluating an Agentic platform, several key features are essential for long-term success. First and foremost, the platform should integrate seamlessly with your existing system tools, avoiding the need for a complete software overhaul. It should support low-code configuration to easily customise Agent behaviour according to your business needs. A strong platform also functions horizontally, unifying business tasks and data into a cohesive system rather than operating in silos. Equally important is the ability to provide oversight, feedback mechanisms, and opportunities for continuous improvement of Agent performance. Ultimately, the goal should be to create an interconnected “AI mesh” within your organisation, enhancing functionality across departments—not just adding another standalone tool.

What CIOs Should Monitor 

As organisations expand their adoption of these solutions, leaders need to monitor these metrics to validate their return on investment:

• Reductions in role and task duplication
• Speed to competency for new hires (AI as embedded trainer)
• Time-to-decision improvements across teams
• Reduction in manual interventions or rework
• Integration fatigue or cost creep from AI subscriptions

Agentic AI will not solve all problems, but it represents a substantial improvement. CIOs need to shift from experimental AI applications to developing a strategic approach that prioritises data and horizontal integration.

To realise real value, AI must be operationalised at scale—driven by aligned use cases, reliable data foundations, and robust governance frameworks.

Successful organisations will do more than implement new tools by fundamentally changing their thinking processes and decision-making actions. As a result, more businesses are appointing Chief AI Officers to lead enterprise-wide strategy, oversee execution, and ensure accountability for AI outcomes.

Contact Newpath if you wish to explore your first Agentic AI solution.