Why many AI initiatives fail to deliver business value
Understanding where AI can create real business value requires a deep understanding of business processes and the technology behind them.
Across industries, AI adoption is growing but turning it into measurable results remains a challenge for many organisations. Pilot projects rarely scale, productivity gains are uneven, and expected business outcomes often fail to deliver.
Most enterprises run on a landscape of ERP, CRM, collaboration tools, analytics platforms and industry-specific systems. Each is designed to solve a defined task. In practice, business processes do not sit within a single system. They span across them. This is where inefficiencies may emerge.
Manual handovers, duplicated data, and fragmented decision-making remain common. As systems increase, so does the complexity of maintaining them. Integration points multiply, along with the risk of inconsistency, failure, and rising operational costs.
A significant share of IT capacity is spent servicing existing systems rather than delivering new value. At the same time, delivery cycles struggle to keep pace with changing business needs.
That’s why we’ve developed the Columbus Agentic Framework, a structured approach to support organisations in identifying where AI creates value and scaling it in a controlled way.
A structural shift
AI is not simply an additional layer of functionality. It reflects a broader shift in how work is executed.
Where applications require predefined steps, AI introduces systems that interpret intent and act across environments. Applications execute tasks. Agents coordinate processes.
This shift is visible across several dimensions. Organisations are moving from command-based interaction to intent-driven outcomes, from large transformation programmes to continuous delivery of value, and from siloed systems to more integrated intelligence layers.
“Enterprise technology is entering a new phase,” says Magnus Oxenwaldt, VP Group AI at Columbus. “We’re moving from systems where every step is predefined to systems that can interpret intent and act across processes. Introducing AI without addressing the underlying structure often results in limited impact.”
Why a different customer approach is required
The core problem is not lack of technology, but the approach. “Organisations are effectively playing 3D chess with a 2D strategy, trying to solve a three-dimensional problem, where applications, processes, and data interact simultaneously,” Oxenwaldt explains.
The average enterprise runs more than 200 SaaS applications. Each is designed to solve a specific problem in isolation. In practice, however, work happens between systems: manual handovers, duplicated data, and approvals managed through email.
As systems increase, complexity grows. Each connection introduces risk, from inconsistent data and process breakdowns to compliance issues.
The cost of maintaining this complexity is significant. A large share of IT capacity is spent servicing existing systems rather than delivering new value. In many organisations, most of the IT budget is allocated to keeping operations running, leaving limited room for innovation.
Oxenwaldt paints a picture: “You’re paying a premium just to stand still. And every year, standing still falls further behind what the market demands.”
At the same time, delivery timelines are out of sync with business reality. New capabilities often take months to implement, while priorities shift within the same period. By the time solutions are delivered, the need has already changed.
Meanwhile, AI adoption is already happening across the organisation, often without structure or oversight. Employees are using AI tools to improve productivity, but without governance, consistency, or visibility. This creates both risk and missed opportunities.
There is potential for meaningful productivity gains, but without a structured approach, those gains remain isolated and difficult to scale.
This shift makes the need for an AI agent framework increasingly clear, providing structure for how agents are governed and coordinated across systems and processes. Unlike traditional applications, agents can operate across systems, follow processes end to end, and adapt to changing inputs. They do not simply execute predefined steps, but they interpret intent and act within defined boundaries.
Without an agentic AI framework, a model for turning AI into action across applications and processes, organisations risk continuing to invest in transformation while most of their resources are tied up maintaining complexity. With it, they can systematically identify where AI creates value, scale what works, and build capability over time.
Where organisations face challenges today
Most organisations operate across multiple systems and in practice, work happens between them. This can create challenges. Data is copied between systems. Approvals take place in email rather than structured workflows. Decisions depend on information brought together from multiple sources.
At the same time, many systems are designed for users who understand how they work. This limits access. Frontline staff avoid complex applications, managers rely on prepared reports, and knowledge becomes concentrated among a few individuals.
Automation, where it exists, is often local. Finance automates finance. Supply chain automates supply chain. Sales automates sales. Yet end-to-end processes still depend on manual coordination, with unclear ownership and errors at handover points.
That’s where the shift towards AI agents becomes relevant. Agents can operate across systems, not just within them. They follow processes end to end and allow people to work in business terms rather than system logic.
The Columbus Agentic Framework
The Columbus Agentic Framework is designed to support organisations moving from AI experimentation to scalable, operational use, where each initiative delivers measurable business value and contributes to long-term capability building.
The approach begins with understanding where the organisation stands today and focuses on building value step by step. “AI maturity looks different in every organisation,” Oxenwaldt says. “Even within the same business, it can be at different stages. Some are curious but haven't started. Others are experimenting but struggling to show business value. A few are ready to scale.
We assess each part of the organisation individually and match our approach to their starting point, the language we use, the pace we set, and the complexity we introduce. A team that has never tried AI needs an initial hands-on experience and a quick win. A team that has been experimenting for months needs governance and measurable outcomes. Treating them the same way guarantees you lose one of them in the first meeting.”
The framework is built around four core elements:
1. Customer engagement model
A structured starting point based on self-assessment, mapping AI maturity and aligning initiatives with current capabilities.
2. Delivery methodology: iterative value creation
AI solutions are delivered in short cycles, progressing from MVP to scaled use.
-
- Sprint 1: MVP
- Sprint 2: Integration and feedback
- Sprint 3: Production-ready
- Sprint 4: Scale and optimise
Sprint 1: MVP - Map the actual business process, not the flowchart version. Identify the pain point and build a working AI agent that handles it. One process, one agent, measurable result. The goal is proof of value, not perfection.
Sprint 2: Integration and feedback - Connect the agent to enterprise systems and real data sources. Gather feedback from the people who will live with it daily. Iterate based on what they tell you, not what the project plan says.
Sprint 3: Production-ready - Validate through internal use, we call it Customer Zero. Eat your own cooking before serving it to customers. Add governance gates, monitoring, and the operational guardrails that separate a demo from a production capability.
Sprint 4: Scale and optimise - Expand more users, more regions and more volume. Measure ROI per deployment. Each success funds the next cycle — the model is deliberately self-funding, so value compounds rather than accumulating as cost.
“At Columbus, we deliver value iteratively,” Oxenwaldt says. “We come from a world of large ERP programmes, but that model no longer works. The pace of change is too fast.”
3. Team structure: combining business and technical expertise
Each initiative brings together a process owner, industry expert, AI specialist and IT representative to ensure practical, scalable outcomes.
4. Technology platform: Digital Coworker
A personal AI agent orchestrator that enables AI to operate across systems and workflows in a structured and scalable way.
This structure forms the foundation of an AI agent framework that enables scalable, governed automation across business processes.
Columbus Digital Coworker
At the centre of the framework is our Columbus Digital Coworker. It operates across business systems, connecting core enterprise tools and enabling AI agents to work across workflows. This allows AI not only to generate insights, but to support execution.
“It’s an intelligent application integrated into the business,” Oxenwaldt says. “It can operate across systems, support decisions, and act within defined boundaries.” By incorporating organisational context, including processes, roles and industry requirements, it enables AI to reflect how the business operates in practice.
Governance is built in. Organisations begin with full oversight and move towards exception-based control as reliability is proven. This approach enables organisations to move beyond isolated pilots.
One example illustrates how this works in practice. A global retailer moved from early AI experimentation to operational deployment by starting with a small number of high-value use cases, delivered in structured two-to-four-week cycles. Each cycle produced a working capability, not a presentation.
Over time, these capabilities compounded into something larger, as AI became embedded across key business areas, with each employee working alongside an AI partner tailored to their role and context. “At Columbus, we don’t start with technology,” Oxenwaldt says. “We start with the business problem and build from there.”
Get ready with the right approach to AI
The question is not whether AI will be used, but how. With the right agentic AI framework, organisations can turn incremental improvements into sustained business value.
Take the next step with the Columbus Agentic Framework.
