Why a robust business case still matters for enabling AI

In this short point of view Michael presents a simple model showing the interaction between key business use cases, key business processes and core data components necessary to enable GenAI to drive business value.

The problem statement

Organizations are investing heavily in GenAI without first grounding their efforts in commercially viable use cases and the data foundations those use cases require. As a result, they create multiple pilots with unclear value, mounting costs, with greater exposure to risk. Without a robust business case and lack of poor data quality remediation efforts, GenAI initiatives rarely move beyond proof of concept, and when they do, they fail to deliver the expected returns.

The author’s perspective and expertise

Observations and learnings from recent projects

A recent PwC UK CIO survey (PwC, 2024) supports this view, revealing that nearly half of CIOs struggle to meet ROI expectations from their AI projects. This reinforces the notion that to successfully scale GenAI requires organizations to move beyond establishing whether the proposed technology is technically feasible and also establish whether the technology is commercially viable. Organizations must in parallel understand how GenAI supports revenue growth, cost reduction, operational efficiency, or risk mitigation. As an example:

• Revenue growth: hyper-personalized marketing and quicker time to market for new product and services launches.
• Cost reduction: decommissioning of legacy systems, applications, dashboards, and reports that can be either consolidated or entirely replaced.
• Operational efficiency: copilots and agents that can partly or fully automate key processes across key business functions like Finance, Supply Chain, and Procurement.
• Risk mitigation: evaluation of risk profiles, anomaly detection and continuous monitoring for security breaches.

Each of these value drivers needs to be referenced in the business case for the adoption of GenAI so it’s important to start with identifying the value, rather than getting side tracked with a demo. Putting it another way, reframe GenAI as a program enabled by data and technology, not a technology experiment hoping to find a business problem. More specifically:

• Build a value thesis: identify biggest pockets of value across key strategic initiatives, business functions and processes where GenAI can realise the greatest.
• Prioritize use cases: Evaluate business use cases in terms of impact vs effort focusing on ‘long-term investments’ (high impact and high effort) as well as ‘quick wins’ (high impact and low effort) creating a balanced portfolio that can demonstrate results within quarters, not years.
• Quantify ROI: Build business cases with clear baselines, benefits, and costs that can be validated and jointly owned by the CTO, the CFO and the respective heads of departments.
• Identify dependencies: For each use case, identify the critical data assets, domains, and components like data architecture, engineering and governance that can act as blockers to scaling GenAI.
• Pilot with purpose: Encourage your team to turn proof of concepts into proof of value so that there are clear gains and benefits.

Beyond this focus on building solid data foundations with a focus on the essential data components like:

• Data Quality: based on standardized, harmonized and enriched master records with supporting meta data that help models understand what data means, not just where it lives.
• Data Management: extraction, transformation and loading of data from multiple sources using well maintained pipelines and well written scripts.
• Data Governance: role-based access, policy enforcement, human-in-the-loop controls, audit trails, and monitoring for misallocation and misuse.

Business Use Cases and Core Data Components

To emphasize, solid data foundations is some generic data modernization wish list; each component needs to be directly traceable to the prioritized list of business use cases. For example if more accurate Finance forecasting and Commercial reporting is on the roadmap, you will need a consistent definition of Customer and Product across your organization with corresponding attributes, and hierarchies down to the lowest level of granularity. If you aim to reduce obsolete stock in your warehouses, you will need a repository of supporting documentation including goods receipts, invoices and contracts to identify instances of duplications and discrepancies.

Concluding point(s)

GenAI’s value does not come from the model alone, it comes from applying the right model to the right problem, enabled by sufficient data. Organizations that start with identifying pockets of value whilst building solid data foundations will reduce common points of failure.

Supporting perspectives

Ernest Orlowski’s value-first approach emphasizes the need to select use cases, define KPIs, and scale responsibly so your AI initiatives don’t stall after the initial PoC. Read Ernest's perspective.