Click Here To Be Featured On The NEW CFOIntell.com

Will Company AI Investments Emulate ERP or Add Value?

By Mark Mishler
Chief Financial Officer, CFO Resource Management
Companies are applying generative artificial intelligence (AI) throughout their organization operationally to create advertising and design new products, and administratively to close the books and prepare financial reporting.

Underneath all the AI hype, CFOs are one of the few industry leaders expressing cautious optimism about this emerging technology. We have been down a parallel technology road before by riding the initial wave of inflated performance expectations of transformative new technology Return on Investment (ROI). This parallel path recalls oversold benefits and cost overruns from software and Enterprise Resources Planning (ERP) investments. CFOs are skeptically experiencing déjà vu.

Measuring benefits and calculating ROI (generally NPV) from adopting new technology has always been difficult. There are several explanations for this.

One reason is that, longer than the nearly two-year gestation of an elephant from conception to finished product, long-term benefits are not immediate. Worse, although project costs are mostly tangible, benefits are largely intangible. Vague benefit examples can be: increased employee productivity, lower operating costs, higher customer satisfaction, better strategic market positioning, and quicker management decision-making.

A second reason is technology complexity, which impairs the ability to estimate costs and project implementation time. Even though the costs are mostly upfront, past experience has resulted in significant direct cost overruns, not to mention indirect costs of management attention disruption during technology implementation periods which have also extended significantly beyond plans. Complexity also results because technology is highly intertwined with business operations, making its benefits difficult to isolate from other business initiatives.

What makes AI technology investments different this time is that ERP had centralized control (and still could not get it right), whereas AI is widely distributed and decentralized. Also different is that ERP had high total fixed costs and a low unit variable cost; whereas, AI’s cost structure is the opposite with a high unit* (per query) variable cost and relatively lower total fixed cost.

In common with earlier ERP investments, however, is that AI benefits are equally challenging to identify and measure because both are highly intertwined with the business. The difference is that while ERP benefits were skewed primarily to the future, AI technology benefits are more current.

CFOs are recalling prior bad experiences from technology investment projects because of these projects’ reputation for late completion, coming back to the money well too often, and not achieving ROI expectations. Thus, this time CFOs want to establish a more disciplined and measured approach to adopting AI technology. This includes capturing a compelling ROI to justify value created from AI costs incurred.

Make sure the AI investment delivers business value
How much value the AI investment creates is determined by the functional department sponsoring the project. Most challenging is establishing a dollar value to “soft” benefits which often are based on fuzzy metrics. Dangerously, by being cast in a quantitative way gives the false illusion of a ROI benefit from the project.

Selected examples of potentially misleading quantitative measurements and value-creating approaches are:
- Customer satisfaction improving from 87% to 90% can be measured; however, determining the business impact ROI requires measuring the incremental sales dollars from achieving this objective.

- System uptime increasing from 98% to 98.7% can be quantified; however to determine the business impact ROI requires measuring cash savings in lower headcount or reduced outside support costs.

- Employee productivity increases can be estimated; however, to determine the business impact ROI requires measuring cash savings in lower headcount or lower direct cash expenditures.

CFOs can take additional better approaches to evaluating AI technology investments.
Say the CEO says that our company needs AI. Before blindly jumping on the bandwagon, a CFO can redirect the conversation by clarifying that the CEO has described a technology and not expressed a business need. Otherwise everyone will have skipped the important step of tying AI to strategy, identifying an expected business benefit, and documenting its ROI. Only then, should the CIO perform a study and recommend a specific AI technology and develop an implementation plan

The CFO can perform a sensitivity analysis detailing how the expected ROI will be impacted if certain key elements of the implementation plan don't go according to plan. For example, what happens to NPV if the project implementation takes six-months longer than expected? Furthermore, the sensitivity analysis goes beyond direct costs as taking longer may end up competing for resources during the company’s peak revenue cycle.

A CFO can also require a robust technology implementation post-mortem and take the conclusion lessons learned and apply them to validating future technology cost, benefit, and ROI analysis.

Other value-creating measures
In addition to ROI using Net Present Value, other value-creating measurement tools include:
1. Balanced Scorecard – connects traditional financial performance objectives with strategic KPIs such as customer satisfaction, new product introductions, internal business process simplification, employee capabilities. This more clearly links business strategy and financial performance by connecting traditional “backward-looking” financial performance with “forward-looking” activities to achieve strategic objectives. Thus, it shortens the “forward-looking” connection between business strategy and technology investments versus a traditional IT focus on process execution and long-term technology cycles.

2. Economic Value Added (EVA): identifies shareholder value created by evaluating economic profit. EVA starts with accounting profit of net operating income and by subtracting the cost of capital assets (capital employed x WACC), calculates economic profit.

Operations may achieve its incentive plan goal of lowering unit product cost by producing greater quantities which lowers unit fixed cost; however, this measure does not include the economic cost of financing higher inventory and operating cost of managing more inventory.

CFOs have been down the technology investment road before with costly ERP investments that frequently under-deliver. This time, with costly AI spending looming, CFOs can lead with a more disciplined focus on ROI.

*Unit variable cost for AI queries is measured in “tokens.” AI tokens are the building blocks of data processed by AI large language models. Token cost is divided into:
1 - input tokens (AI query) and
2 - output tokens (what the AI generates) which cost 3-5 times more

Token costs are charged per 1 million tokens based on the AI model’s complexity and intelligence:
● Low-end: Gemini 2.0 Flash Lite or OpenAI Nano-Tier - $0.05-$0.60
● Mid-Range: GPT-5.4, GPT-4o, Claude 3.5 Sonnet - $2.50-$15.00
● Premium: OpenAI o1-Pro - $30.00-$600

0 Comments