Part 2 of a 5 part series of suplimental commentary blogs following the parent “position paper” recently published titled “Why AI Must Be Measured as Software and Systems not Anthropomorphised Apps” – please read this parent article as a precursor to this 5 part series..
If we are to firmly discard the notion of artificial intelligence as a human resource or a novelty application, we must completely overhaul how we calculate its return on investment. As of 2026, despite nearly universal adoption across enterprise environments, an estimated 80 to 95 per cent of AI projects fail to deliver their promised business value. Furthermore, recent industry research has highlighted that the vast majority of generative AI deployments yield no measurable profit and loss impact. This staggeringly high failure rate is rarely due to the core algorithmic models themselves. Instead, initiatives routinely collapse because of poor data pipelines, inadequate systems integration, and a fundamental misunderstanding of workflow economics. When organisations base their financial models on a simplistic comparison between the cost of API tokens for an anthropomorphised app and an employee hourly wage, they are destined to fail.
Evaluating AI as a system means recognising the total cost of ownership inherent in complex software deployments. The true price of an AI solution is never merely the inference cost or the monthly cloud subscription fee for a consumer facing interface. It includes the capital expenditure for system implementation, ongoing code maintenance, stringent data governance, and the crucial human in the loop review processes. For instance, even if a generative AI tool drastically reduces the time required to draft technical reports or process insurance claims, the necessity for a human expert to thoroughly review the output for hallucinations or errors represents a significant, ongoing operational expense. A seemingly cheap artificial intelligence model can ultimately yield a deeply negative return on investment if the software integration is clumsy, governance is weak, or end user adoption fails because the system was treated like a standalone gadget rather than core infrastructure.
Consequently, true business value is generated at the workflow level, not the individual task level. Corporate leaders must analyse how an AI system alters the entire operational pipeline rather than just timing a single automated output from an app. Does the software integration reduce downstream rework? Does it protect existing revenue streams? Does it systematically mitigate material risk? In many business scenarios, traditional, deterministic software, which strictly follows fixed, predictable rules, might actually be a cheaper, safer, and more reliable solution than a generative model. Treating AI strictly as software forces decision makers to conduct rigorous technical evaluations, asking whether the probabilistic nature of machine learning is genuinely suited to the problem at hand, or if it is simply being deployed for the sake of following a trend. A systems first approach to ROI also demands dynamic, continuous operational tracking to gain a complete view of whether value is being sustained.
Comprehensive Cost Modelling
Boardroom Conversation Starter: What is our exact budget for the ongoing data governance, human review processes, and system maintenance required to keep this AI deployment secure and accurate, beyond the interface level? It is vital to surface these hidden costs early, as the initial procurement price of artificial intelligence software often represents a mere fraction of the true financial commitment required to sustain it in a live enterprise environment.
Impact on Success Measurement: If success is measured solely on the initial deployment speed and front end conversational capabilities, leadership will be blindsided by the ongoing operational backend costs, viewing the long term integration as a budgetary failure. Ignoring the continuous requirement for data cleansing, model fine tuning, and cloud computing resources ensures that the project will eventually bleed capital, completely destroying any perceived initial victories.
Actionable Path Forward: Define success through a continuous total cost of ownership framework. Allocate specific budgets for data pipeline management and human in the loop reviews, measuring financial viability on a rolling quarterly basis rather than relying on a single launch metric. This ensures the business commits to the technology as a permanent infrastructure capability rather than a one off experiment.
The Workflow Integration Imperative
Boardroom Conversation Starter: How does this new software seamlessly connect with our existing legacy systems, and where does it potentially create new operational friction for our staff if treated just as a standalone app? The most sophisticated machine learning model is entirely useless if it forces employees to manually copy and paste data between disconnected platforms just to complete their daily tasks.
Impact on Success Measurement: Measuring the capability of a standalone language model in isolation provides a false positive. If the software cannot integrate smoothly into the wider enterprise architecture, the overall operational success metric will plummet due to process bottlenecks. A tool that produces brilliant analysis but traps that insight within an isolated silo ultimately degrades corporate velocity rather than accelerating it.
Actionable Path Forward: Audit the entire operational pipeline before deployment. Measure success by the reduction of friction between different business units and the smooth transfer of data across the integrated systems, rather than the isolated performance of the anthropomorphised interface. True workflow economics demand that the AI acts as a digital lubricant for the entire business process.
Evaluating Deterministic Alternatives
Boardroom Conversation Starter: For this specific business problem, do we actually require a probabilistic machine learning model with a conversational overlay, or could a traditional rules based software system solve it more reliably and cheaply? The hype surrounding generative models often blinds executives to the reality that a simple database script can sometimes achieve the exact same operational outcome with a zero per cent chance of hallucination.
Impact on Success Measurement: Implementing generative AI where simple automation would suffice leads to severe over engineering. Success metrics will be skewed by the unnecessary complexity, high error rates, and increased compute costs of using the wrong tool for the job. Pushing probabilistic software into a deterministic workflow guarantees endless debugging cycles and massive frustration for the operational teams.
Actionable Path Forward: Create a strict technical evaluation matrix before any procurement. Measure the success of the IT strategy by its ability to match the correct software architecture to the right problem, deploying generative models only when strict deterministic rules cannot apply. Discipline in software architecture is the ultimate driver of sustainable return on investment.
Navigating the immense complexities of any successful adoption, integration, or development of AI in any way, shape, or form, be it in house or via third party systems and platforms, can be exceptionally challenging. Because these capabilities are not normally part of an organisation’s core business, ensuring your enterprise is fully prepared requires strategic foresight and expert guidance.
If you, the reader, or your business find yourselves in a similar situation and need help responding to an unexpected technical event, or if you want to proactively plan to ensure you deliver successful outcomes for any AI project or initiative, please reach out. In my role as Chief Executive Officer of Sociaall Inc., a leading consulting and advisory business, my team and I are available to provide support to develop a robust roadmap, actionable strategic plans, and capable internal teams. As artificial intelligence consulting and advisory experts, we bring decades of hands on practitioner experience to every engagement we take on. Our professionals have proven track records in enterprise data strategy, complex systems integration, and corporate leadership, meaning we understand the technology both as code and as a business driver. We would love to work with you and support your organisation with new projects or any initiatives currently in progress, regardless of type, size, scope, or scale.
