Evaluating the Impact of Intelligent Data Processing for Corporate Finance with the Use of Real Options Analysis

Abstract

Technological innovation is changing virtually every aspect of business practices and operational procedures. The introduction of large language models and various types of intelligent processing, commonly referred to as artificial intelligence, presents significant change to cope with. In this paper, we suggest an estimation method, based on real options analysis (ROA), that improves the assessment and valuation of intelligent data processing’s impact on organizations. The presented approach can reflect direct and indirect effects from introducing artificial intelligence methods and is therefore better suited than traditional financial metrics for the assessment of contemporary intelligent tools and solutions. Using Monte Carlo simulation and American-style real options, we have estimated two sample use cases to compare the ROA results against other common valuation methods. Numerical experiments indicate that the suggested approach is capable of capturing both the direct and indirect impact of new technologies, which improves relevant financial and management decisions.

1. Introduction

While digitalization can take many paths and forms, it can be summarized and narrowed down to the various technologies that replace manual labor and automate various tasks, speeding them up and reducing the time required to process inputs and transform them into reasonable decisions. Even though such a process is appealing in many ways—being simple to understand and explain—there are many long-term side effects that may go unaccounted for. Therefore, it is necessary to use a more solid approach when analyzing the impact of innovative data processing and decision support tools. In their bibliometric study, Rihab et al. (2025) highlight that research is often focused around specific technologies and their influence on financial practices. While specific tools and analytical algorithms are very important for the success of such innovations, we can argue that their true impact has to be understood by considering the broader changes they will eventually trigger in organizations. In addition to the most popular ways of approaching intelligent method applications in corporate finance, we suggest another criterion based on how they enhance or limit the flexibility of companies to adjust to ever-changing market conditions and new restrictions (which could be in the form of new regulations or even novel competitive advantages—as in the control of AI usage).

Table 1. Common ways to address the impact of intelligent data processing for corporate finance problems.

Approach	Remarks and Special Characteristics
Technology-based	This is one of the most frequently used approaches, which was used, for example, in Liu et al. (2025), Adorm-Takyi and Asare (2026), Chaplygina et al. (2025) and Bhati et al. (2025). Such an approach may be very well justified when a special case or individual implementation is considered, where technology plays a crucial role and the major source of uncertainty and risk can be attributed to such implementation details. However, focusing on instruments is rarely sufficient to understand the game changing innovations that influence the various aspects of organizational processes.
Algorithm and/or area of application	Due to the large number of possible applications, the algorithmic or implementation area approach is very useful when addressing highly specialized improvements—as in Qidi et al. (2026), Ouyang (2025) and Gary et al. (2025). As the name of the approach suggests, these research efforts are focused mostly on the implementation and improvement of existing solutions; thus, their relevance is highest with regard to the specific algorithm or area use cases. When it comes to estimating the overall impact of intelligent methods on organizational behavior and company development, assessments consider the importance of the respective application and method.
Market and competitiveness	The impact on market share and competitiveness approach uses sound financial logic to estimate the impact of innovative technologies and intelligent data processing—as shown in Qiang et al. (2026) and Barry and Haitham (2026). As the focus is the overall results, it is less likely that you will get caught in addressing a local maximization and miss the whole picture. Yet, the same focus on the big picture makes it hard to link to specific decisions, as multiple factors influence market positioning and competitiveness. The application of intelligent data processing for financial data is certainly contributing to future company growth, but such estimates are typically forward-looking and very uncertain. In addition, due to the fact that new methods and technologies are changing fast, the assessment of their influence may be highly speculative. For example, forecasts on artificial intelligence applications vary from “total transformation of workforce and economy” to much more modest estimates for “increasing productivity by 1.5% by 2035”.

summarizes why the flexibility-centric approach is different and how it compares to other methods of addressing financial data processing. Through this paper, we refer to innovative and machine intelligence-supported algorithms as “intelligent data processing”, instead of using the term “AI”, since artificial intelligence itself can refer to many applications, with some of them not relevant to our research.

In order to combine the advantages of the various approaches discussed in Table 1, without affecting the specific implementation context, we suggest a different approach toward the use of intelligent data processing. If one considers every new technology and its respective tools as a new opportunity for companies and their employees, then adoption and long-term impact can be modeled as if there is a specific option for the use of the methods and tools.

The aim of this paper is to close an existing gap between various impact assessment methods, shown in Table 1, and the practitioners’ requirements for a clear and easy-to-understand metric that can help to introduce new technologies that can disrupt and even fully reshape existing business and analytical processes. Applications that rely on contemporary AI methods (ranging from a pure LLM that supports individuals in extending their expertise to intelligent personal assistants that automate some repeating tasks) are clearly an example of this, since their use does not have a long history that can be studied and analyzed and because they are also subject to rapid change with the introduction of new algorithms and specific tools. The combination of high uncertainty, technological complexity and company-wide impact, where almost every aspect of standard activities has to be updated, creates a huge challenge for traditional risk estimation and financial evaluation approaches. There are various studies (as in Lee, 2011) that focus on the advantages of real options analysis in addressing technical uncertainty and evaluating the flexibility to switch between various tools, modes and use scenarios. However, ROA is rarely applied in the context of AI impact assessments, even though real options can effectively account for both the strategic effects and the evolving risks of introducing intelligent data processing into various business processes. Our study aims to help close this gap by highlighting specific use cases of these analytical tools. Experiments are also carried out that show how typical estimation challenges can be resolved with the use of Monte Carlo simulations.

2. Literature Review

The advantages of real options analysis have long been used in the valuation of new technological solutions in renewable energy, as in Lee (2011) and Papadimitriou et al. (2023) for example. There are various industries where the application of ROA is highly beneficial (Li et al., 2022; Androniceanu & Sabie, 2022), yet most of the use cases follow a particular project and focus on several areas of improvement:

-

Improvements in valuation accuracy over standard methods like net present value and simple sensitivity analysis.

Antikarov (Antikarov, 2025) discusses the potential pitfalls of net present value (NPV) and how flexibility and timing alignment can be improved significantly by the use of real options for valuation of investment opportunities. At the same time, ROA is far from being perfect as applications can suffer from increased complexity (Janney & Dess, 2004) and lack of integration with other business processes in the organization (Miller & Waller, 2003). A common limitation of studies that focus on ROA advantages over traditional metrics is that they are very often limited by analyzing a specific project or activity, thus ignoring the indirect impact that some decisions may have. More sophisticated scenarios are often represented as a portfolio of options (Petri et al., 2007; Smith & Thompson, 2008), but this approach is not only more complex, but also with limited capabilities to analyze indirect effects.

-

Improvements in understanding and estimating risks and building adequate decision paths.

ROA has been used as a tool for extending scenario analysis (Miller & Waller, 2003) and formulating proper risk counter measures, as well as a means to estimate new types of specific risks (Stroombergen & Lawrence, 2022). In both cases, we can see that there are clearly defined sources of uncertainty. Therefore, a major obstacle is to estimate the strength of the impact, while its nature and target are mostly known in advance. AI-based tools, on the other hand, tend to combine not only risks from introducing wrong technology, but also the danger from biased decisions due to misunderstanding and misusing them. This requires ROA to be used with extra care and it has to be kept in mind that the uncertainty organizations are facing is complex and due to the simultaneous presence of multiple factors.

-

Highlighting strategic options and new opportunities.

Supporting strategic decision-making with ROA is one of the possible use scenarios, where option valuation advantages are most relevant to the specific agenda and issues that intelligent data processing and AI algorithms are imposing on organizations. This is true because moving toward intelligent processing is in fact a decision with strategic impact and consequences. We should explicitly note that there is a big difference between the decision of using AI support and AI-based tools and the application of intelligent methods to support strategic decisions (as in Stone et al., 2020). The former focuses on the impact of new technologies on various aspects of organization operations, while the latter is structured around a specific narrower use case. This research focuses on estimating the impact of intelligent processing uses, so that organizations can better justify investments in such tools and prepare for the various consequences that will follow their decision.

Use of Real Options Analysis with Intelligent Data Processing

Unlike their financial market counterparts, real options considered in this paper cannot be separated from the normal business processes of the respective economic agents. This is due to the fact that opportunities, risks and uncertainties that accompany introduction of intelligent tools are related to the improvement in organization current activities. When introducing various intelligent data processing methods and tools, we need to consider the organization’s characteristics and specific context. Therefore, taking the decision to go (or not to go) over a specific way, when to implement an intelligent tool and how to integrate it are all examples of exercising a real option. Taking such a point of view has a significant advantage—it allows us to map our opportunities exactly to the specifics of analyzed organization (or individual process), while at the same time it requires us to have a clear definition of what are the driving factors of success. Because of the existing internal dependencies, it also raises the level of awareness on potential side effects—which could be either positive or negative. We believe that this last part is an essential benefit from using real options analysis, as it gives an opportunity not only to assess a specific intelligent solution, but also to estimate its impact throughout the whole organization. Recent studies, like, for example, in Füller et al. (2024), indicate that artificial intelligence is triggering changes that go far beyond improvements in data gathering and processing. We argue that with properly adapted and applied real options, we can achieve an accurate numerical estimate of these effects.

Real options analysis (ROA) has already been used to support strategic decisions in Kabaivanov et al. (2013) and evaluate the benefits of flexibility in de Mello-Sampayo (2023). Advantages of the ROA can also fit well in the assessment of intelligent data processing for corporate finance, due to the fact that real options favor flexibility, which is a central part of the suggested assessment approach.

Figure 1 highlights the typical application of areas of AI and intelligent data processing in general, which is used to support existing corporate finance tasks. It is worth noting that as we move to the higher application levels, the invisible side impact of intelligent solutions increase, as they influence a larger number of corporate finance tasks and their use affects the organization for longer periods. For example, a strategic decision that was suggested and resulted from AI support system reasoning will have much bigger effect on the whole organization, compared to a mere next period forecast. While both can be good or bad, the impact of the first one will be left for much longer.

With regard to technical complexity, it is not necessary that higher-level solutions are more complex or technically demanding. Even though they intend to solve tasks that are more impactful, this does not imply they need to be more sophisticated. Layering on the figure mainly focuses on the scale of application consequences and time effects, and not on technical challenges.

With regard to this argument, we can assume that more elaborate applications will also bring with them higher uncertainty for the future. While this is true for any strategic decision, our claim is linked to the use of AI and intelligent data analysis, since their application can most likely trigger a different outcome, compared to traditionally used methods.

Our paper extends previous ROA implementations in several ways, that are relevant to the impact of intelligent data procession on the way organizations work and do business:

-

The layering of AI and intelligent data processing applications, seen in Figure 1, allows us to address each level with different types of options, to ensure that they are most suitable to its immediate and long-term effects.

This method reduces the complexity of the matter and applies the most relevant intelligent processing method for the respective task. Considering the large variety of contemporary solutions that are labeled as “AI-assisted” or “AI-driven”, such an approach can split complex implementation scenarios into several smaller problems.

-

Special attention is paid to side effects and the indirect impact of intelligent data processing, which widens as the scope of introduced solutions.

The benefits from keeping track of side effects can be split into two major categories. The first one being that indirect impacts reflect the broader changes triggered by intelligent data processing such as changing individual performance, highlighting steps and procedures that can be fully or partially automated and allowing for the fast introduction of new practices. It should be noted that side effects do not need to be all positive, as intelligent algorithms can also lead to the loss of specific habits and promote close-to-average behavior. Further discussion on how AI can impact decision-making in a limiting manner can be found in Crowther and Hamdan (2024).

The second major category refers to how identified real options can be properly defined and scoped in order to prevent scope creep, when introducing new intelligent processing methods and other AI tools.

-

The indirect impact of intelligent data processing can span over long periods, often surpassing the life time or planning period for a particular solution or project.

ROA in this case can be of help to identify the “remainder” effect, which standard evaluation methods will miss or ignore. As intelligent processing methods are expected to influence more than one aspect of how organizations operate, such additional effects will accumulate and, if not properly accounted for, will bias our estimates and distort future decisions.

3. Methods and Analytical Tools

Estimating the impact of a new technology or data processing solution is a complex task that relies on various assumptions. This is even more relevant, when the technology itself is yet to be proven and under very active development—as is the case of large language models and intelligent data processing. To address this problem, we suggest a three-step approach, presented in Figure 2.

The first step in the process is to classify the type of intelligent data processing solution. The purpose is twofold—the first is to identify what are the appropriate estimation techniques and the second one is to evaluate if there is an indirect impact from introducing the respective solution to other business processes. For example, a simple solution that automates financial document scanning and verification is expected to have an important but highly localized impact on the way a typical company would continue to operate. On the other hand, a decision support system that deploys a large language model with business-specific reasoning will influence not only its direct users, but virtually everyone in the company.

It would be naïve to consider that the impact of both presented cases can be assessed with the same analytical methods, sharing the same assumptions and limitations. Therefore, the initial classification step aims at putting a specific solution under a predefined category and then selecting an appropriate valuation method.

Table 2. Common estimation methods and typical problems when used with intelligent data processing solutions.

Estimation Methods	Special Remarks
Cash flow and return-based methods	Cash flow and return-based estimation relies on common metrics like return of investment, calculated as net benefits divided by total investment. Despite the solid logical and economic background of such metric, it is not well-suited for applications that may result in significant side effects, spanning over a very large period and carrying the promise to completely disrupt important processes in the organization. The core reason for such an argument is that ROI and its derivatives are simply not able to capture precisely the full impact of intelligent solutions because they do not account for indirect impacts and the long-term competitiveness/efficiency improvements that may go beyond the time scope of initial new intelligent methods implementation.
Performance-based methods	Performance-driven analysis often relies on changes in speed when carrying out specific operations, or improvements in the total output from a given activity. While such an approach does not require explicit monetary estimate of the impact, it is still vulnerable to an incorrect assessment of the side effects and a misinterpretation of the potential of new technologies.
Combination of methods	Combinations of performance and return-based methods can solve some of the issues highlighted in the previous sections. However, the main obstacle with the valuation of flexibility and overall impact still remains, as none of the methods discussed in this table is designed to account for the potential big side effects that can lead, in time, to strategic shifts in organization behavior.

provides a summary of the suggested assessment techniques based on the inherent features of the analyzed solution.

The common tools used to estimate the impact of new technologies, like return on investment (ROI) or various key performance indicators (KPIs), are not perfectly suited for the case of intelligent data processing due to the significance of the side effects, the fast evolution of the deployed tools and the very fast “depreciation” of deployed solutions. Table 2 summarizes some of the deficiencies. As seen from Table 2, we have two major problems with the estimation—how to account for flexibility and changes in implementation as new information comes in, and how to consider effects from new technologies that are beyond the immediate implementation scope. Both issues are very important and if not properly included in the analysis can lead to wrong estimates. In particular, miscalculating the benefits from flexibility may delay introducing new algorithms and intelligent analysis methods, thus also harming the long-term competitiveness of the organization.

As seen from

Table 3. Approaches to assess intelligent data processing solutions for corporate finance.

Solution Characteristics	Suggested Techniques	Remarks
Limited in scope solutions that target an individual step or analytical operation. Typically, such solutions would map to either forecasting and/or automation level in Figure 1.	ROI, KPI and physical time savings (speed improvements)	Due to their localized impact, such solutions can be evaluated and monitored with standard financial success indicators and performance metrics.
Intelligent data processing that affects multiple aspects of corporate finance practices. Such solutions can aim at risk analysis support as well as at the automation of complex tasks and workflows.	ROI and ROA analysis	As the number of affected processes grows, so does the number of possible outcomes.
Data processing solutions that have long-term impact and side effects across organizations. Typically, these belong to strategic decision support systems and complex risk analysis.	ROA with limited application of standard metrics.	With the increased uncertainty and spreading of additional effects on the whole organization, the ROA is better able to capture uncertainty and results from future decisions.

, effects on the flexibility of the analyzed organization are crucial, as they highlight the benefits of using ROA over other assessment techniques like simple ROI, efficiency and speed KPI to name a few. Option-based assessment has two additional advantages, which are important for the application of intelligent solutions:

-

Real option valuations allow organizations to have more than a source of uncertainty. When corporate finance analysis relies on intelligent processing methods, it is especially valuable since one needs to account for market uncertainty in addition to technological complexity and potential risks related to the use of new solutions.

-

Option analysis fits very well into the step-by-step introduction of new methods and support tools. Even though reports (KPMG, 2024) indicate that the adoption rate for AI tools, for example, is high, it is worth pointing out that covered use cases are different and of various maturities.

As applications of intelligent data handling progress in terms of process coverage and complexity, real options are better capable of encapsulating different stages and decisions that follow them, based on the success of failure of the experiments.

Quantification of the impact is the third step and includes the evaluation of already identified opportunities. While it is sufficient to have just one real option in many situations, there are cases where a portfolio of options can be built—depending on the complexity of the implementation scenario.

Table 4. Suggested techniques to assess intelligent data processing solutions for corporate finance.

Option Parameter	AI/Intelligent Processing Parameter	Financial Option Counterpart
Underlying asset value	Present value of the cash flows generated by the implementation or estimation of the savings generated with it.	Value of the underlying asset—for example a security.
Exercise/strike price	Financial resources required to back up the implementation of the respective solution.	Exercise price
Time to maturity	Planned testing period (in case a decision is expected whether to keep the solution or not) or time span till next step in the implementation.	Time to maturity of the option
Volatility of the underlying asset value/price	Uncertainty about the overall impact of the solution and the cash flows/savings it will lead to.	Volatility of the underlying asset
Additional cash flows	Additional in/out cash flows associated with implementation lifetime.	Dividends paid

summarizes the inputs required to complete the evaluation.

Actual valuation of the real options can be done with one of the commonly used methods:

-

Various solutions for the Black–Scholes equation, as in (1),

$${\displaystyle \frac{\partial V}{\partial t}}+{\displaystyle \frac{1}{2}}{\sigma }^2{S}^2{\displaystyle \frac{{\partial }^{2}V}{\partial S^2}}+rS{\displaystyle \frac{\partial V}{\partial S}}-rV=0,$$

(1)

where V(t, S) is the option price as a function of time and underlying asset value, $\sigma$ is the volatility of that same asset value and $r$ represents the risk-free return, while S denotes the spot price of the underlying asset.

As the typical model application is for European calls that are not paying dividends, one can use well-known extensions and approximations for American options (better suited to map decisions that could be taken at any point in time) (Barone-Adesi, 2005; Alghalith, 2018) and for situations with intermediate benefits or cash flows (resembling the dividend payment model extensions as in Lioui (2006)).

-

Variations in binomial option pricing models, as in (2),

$$\begin{gathered} C_0=V_0\left[\sum _{n=a}^t{\displaystyle \frac{t!}{(t-n)!n!}}{p}^n(1-{\mathrm{p})}^{\mathit{tn}}{\displaystyle \frac{u^n{d}^{t-n}}{(1+r_f{)}^t}}\right] \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}-{\displaystyle \frac{E}{(1+r_f{)}^t}}\left[\sum _{n=a}^t{\displaystyle \frac{t!}{(t-n)!n!}}{p}^n(1-{\mathrm{p})}^{t-n}\right], \end{gathered}$$

(2)

where $C_0$ is the call option price for an underlying asset that may go up in value (with coefficient u) or down (with d). Parameter a that we see in (2) comes from (3).

$$\begin{array}{c}\mathrm{a}-\hfill \\ >{\displaystyle \frac{\mathit{ln}(\frac{E}{V_0{d}^n})}{\mathit{ln}(\frac{u}{d})}}, smallest non\hfill \\ -negative integer value, derived from u^a{d}^{n-a}{V}_o\ge E,\hfill \end{array}$$

(3)

As (1) and (2) suggest the use of either the closed-formula solution or the binomial tree variant, it would allow us to model the uncertainty and consider that introducing intelligent data processing is not a one-way street with possibilities to adapt to intermediate results. Thus, if the solution proves to be more promising than the initial expectations, its application can be extended or scaled up (thus exercising an expansion option). On the contrary—if the output is not satisfactory, then it could be either completely abandoned (e.g., exercising termination option), replaced by a different method (e.g., a switching mode/type option), temporarily put on hold, or scaled down.

-

The use of higher-order trees to model complex scenarios, which means that organiziations typically have to deal with more than one major source of uncertainty (Kabaivanov et al., 2013).

Typically, these sources of uncertainty may be technological risks, associated with new intelligent solutions, as well as the economic impact from their use—which we can expect could be improved competitiveness.

-

The use of Monte Carlo simulations for the evaluation of complex options.

Binomial trees offer a powerful and relatively easy-to-use way to evaluate real options. Yet there are situations where we need to account for more than two important factors, in order to present complex scenarios. In this case, Monte Carlo can support our evaluation in two distinct ways:

○

Estimation of variance: This is needed for closed-form solutions, like Black–Scholes, for example.

This use case assumes that we can create a supplementary model for the changes in the underlying asset value (in this case, this would be a company value change as a result of strategic-level applications or savings/costs for a process, in case of more localized applications). This estimate can then be fed into one of the assessment methods mentioned before—in order to arrive at a valuation of the respective real option.

○

The estimation of option value directly, with the use of certain company development projections.

This use case is built upon the assumption that once there is a model with sufficient precision, the underlying asset value changes can be derived from it. Then, with such knowledge in hand, it is possible to generate a large number of scenarios and estimate the impact of the respective intelligent data processing solutions. With a large number of simulated scenarios, it then becomes possible to assess the most probable outcome, as well as in what range it will most likely be.

Table 5. Estimation of option parameters in the case of AI and intelligent data processing applications.

Option Parameter	Remarks on Estimation/Calculation
Underlying asset value	Underlying asset value is derived from the expected gains and benefits relevant to introducing the new technology. It represents the expected value of savings in terms of time, effort, manpower and other resources that intelligent data processing brings. Underlying asset value can be estimated in a number of ways, but, essentially, one can follow the good practices associated with other common financial indicators, since this parameter is also the mean/expected value in effect.
Exercise / Strike price	The strike price is estimated based on the need of financing in order to introduce and put in operation the respective intelligent solution. While it is often considered to be a constant value, there is no strict requirement for this and the exercise price may vary, depending on the real option context.
Time to maturity	Time to maturity is determined by the implementation and use time of the respective intelligent data processing solution. Based on the rapid development of AI technologies, we can expect that the lifetime of individual tools will be relatively short, which makes it easier to split implementation into separate parts and figure out the time taken for each one to mature.
Volatility of the underlying asset	The volatility of the underlying asset can be estimated with regard to how impact of the used tools evolves. This can be done either with the use of expert analysis or by running dedicated Monte Carlo simulation. Proper estimation of this parameter is crucial in order to make sure that further analysis is well justified and not biased by too high expectations.
Associated cash flows: Indirect impact Side effects Additional investment spending	Associated cash flows are one of the most important inputs for the real option valuation when applied to AI and intelligent data processing algorithms. The reason is that with these extra costs and revenues, one can represent the indirect impact of analyzed solutions. Positive side effects will result in extra savings or the higher efficiency of other processes, thus resulting in positive associated cash flows. On the other hand, negative indirect impact can be represented with value outflows, thus reducing the estimated real option value.
Option type	Although listed last in this table, option type is the very first input that has to be clarified in order to continue with the analysis. Many of the well-known real option types can be applied for intelligent data processing implementations, like the following: - Options to abandon can be used to terminate implementations that do not meet expectations, or have triggering significant negative side effects. A lot of situations are very well-suited for this type of real options—especially for tools that are service- and subscription-based, since the direct termination costs are not very high. For internally implemented intelligent solutions, the expenses may be higher though. - Expansion options are particularly valuable in case of considerable positive effects that evolve with the implementation. While it is hard to plan for positive impact with high precision, the importance of flexibility to expand may be tracked through monitoring the associated indirect effects and potential changes in the “dividend-like” benefits. - Real options to switching between various use modes and tools can be used to represent the quick changes and updates in the available intelligent data processing tools. It takes time to adapt and train intelligent algorithms for a specific task. However, we have witnessed a rapid development of new tools, frameworks and analytical methods. Thus, organizations need to consider their freedom to choose new variants and apply intelligent solutions for new tasks. - Delay options are useful for solutions that have noticeable side effects, but it requires time to gather feedback and estimate them. No matter how sophisticated and well-thought an intelligent data processing method is, eventually, it will have to fit into the established organization processes and gradually change them. This involves also changing human behavior and educating people, which requires time. Real options to delay and wait are very useful, as they allow organizations to better estimate the full impact of new solutions.

provides an overview of the possible methods to estimate real option parameters, based on how intelligent data processing is introduced in existing organization processes. With other valuation models, correct estimates of these parameters are very important in order to be able to obtain an adequate value for the respective real option. For some of the inputs like underlying asset value, one can reuse existing approaches that are well-known from common methods like net present value and scenario analysis. Others, in particular associated cash flows and volatility, require in-depth study of the context in which real options exist. This is necessary as potential positive or negative side effects may differ based on the organization context, even if the introduced intelligent solution characteristics are well-known.

We argue that the use of associated cash flows and real options analysis models that include dividend-like payments is very important for estimation and the following decision-making steps. Simpler valuation models (for example, the initial Black–Scholes model (Black & Scholes, 1973)) that do not allow for dividend payments are more appealing in terms of complexity and ease of use. But they do not offer the capability to account for side effects and organization-wide changes that new methods are expected to trigger. Therefore, allowing for dividend-like payments during option life time is not just a technical detail, but a very important feature that allows us to better include real-world effects into our estimates.

Real options analysis should also be considered with regard to its specific reasons and limitations. In particular, if there is very little uncertainty about the outcome of the introduced new tools and technologies, then simpler and more common estimation methods can be applied. We can assume that the lack of uncertainty would also result in a stable estimate of the associated cash flows (e.g., indirect impact), so these can also be integrated into other common valuation techniques of preference.

ROA benefits will be best visible when there is a high uncertainty, both from a technical and business point of view, which is a characteristic that we believe is inherent in the recent development of AI and intelligent data processing.

4. Applications and Results

We discuss two typical ROA valuation applications, based on the assumption that the organization wants to automate a single process with the use of an intelligent data processing tool—a learning agent. While this may seem like a very specific and narrow-scoped scenario, we consider it suitable for demonstrating the benefits of ROA due to the fact that it is pretty common and fits well into the “start small” strategy that prudent organizations take when considering technology that is brand new and complex.

To demonstrate the importance of side effects, we first carry out an experiment where it is assumed that the use of the intelligent agent does not cause significant side effects. Then, we repeat the valuation, but this time with the presence of positive indirect impact, speculating that intelligent processing improves the efficiency of employees involved in the process by a fixed percentage. Costs and subscription rates have been averaged from publicly available data from major AI and service provider rates. For the volatility estimate, we have carried out a simulation on the possible outcomes and calculated the standard deviation of the results. Depending on the complexity of other scenarios, this part can be extended to include more factors and involve other means of volatility reckoning.

4.1. Numerical Example 1

For a simple demonstration of real option, let us assume a company wants to experiment by introducing a new intelligent agent to scan through financial documents and give a simple summary feedback. To avoid large investment costs, a subscription model is chosen with a total yearly cost of 2748 EUR. The agent implementation is expected to introduce savings in terms of freeing up to at least 10% of the time of a typical junior department employee, which, at the time of planning, amounts to 2400 EUR per year. Should the experiment results prove to be not satisfactory, the company expects that it can switch the application to simple invoice scanning that is expected to give a solid saving of half of the initial amount (e.g., 1200 EUR).

It is expected that the agent introduction will also have a positive impact on other processes in the financial department, leading to faster decision-making and better understanding of the ongoing processes, which may actually boost the benefits further and the estimate is that the actual savings may be different with their volatility estimated at 35% per year. Taking out a sample risk-free rate of 2.222%, using the recent 12-month EURIBOR rates as reference, this can lead to a simple valuation of the available option.

Table 6. Simple valuation of new technology option—introduction of intelligent agent.

Option Parameter	Parameter Value
Underlying asset value	The underlying asset value is the estimated savings from introducing the technology, initially set at 2400 EUR.
Exercise/strike price	With an initial investment of 2748 EUR (which we treat as a sunk cost) and a termination option for the experiment (switching to another use), we have an exercise price of 1200 EUR.
Time to maturity	Assuming one-year test period.
Volatility of the underlying asset value/price	Volatility set at 35% (0.35) at the initial experiment. For the Monte Carlo simulation, we allow the volatility to change with log-normal distribution~Lognormal (0.35, 0.1).
Additional cash flows	No additional cash flows have been included in the example.

contains a summary of the numeric example inputs. Being able to cancel the test and switch to an alternative with a fixed result (even if the outcome there is not fixed, we can still evaluate the option but a constant value makes it easier) is in effect exercising a put option. Since this decision can be taken at any time, we evaluate an American put with the Crank–Nicolson method at 4.408327 EUR and with the Barone–Adesi–Whaley method at 4.540619 EUR.

If we allow for the volatility to change, then the Monte Carlo simulation can be done with volatility fluctuating over the estimated value of 35%. In this case (assuming the volatility itself follows log-normal distribution, as it cannot be negative), the output of the sample simulation is presented at Figure 3. The mean estimate from the 10,000 samples is 6.405283 EUR with the Crank–Nicolson method (left) and 6.405209 EUR with Barone–Adesi–Whaley method (right). This assessment gives a more accurate view of the positive impact that agent implementation has, compared to traditional net present value (which will result in a negative outcome as expected value of the benefits (2400) is less than the investment costs (2780)) and ROI (which in this case is also negative, based on the initial expectations or −13.67%).

Depending on the specific business case, a numerical example can be further extended to account for any intermediate positive (or negative) effects of integrating the new intelligent solution. This may be done by introducing a dividend yield in the valuation that represents these effects.

4.2. Estimation of Indirect Impact with the Use of Real Options

Real options have much in common with their financial counterparts. However, there are some subtle variations that should be taken into consideration, even if they are not explicitly visible from the closed-form valuation methods (as these are frequently adapted from financial option valuations). We focus on two important differences that affect the impact assessment of new technologies and, in particular, intelligent data processing:

-

With option valuation methods, as we have talked about, the option value is considered to be $max(S-E, 0)$ for calls and $max(E-S, 0)$ for put variants, which implies the option intrinsic value should be zero or positive.

With real options used for the new technology assessment, there is a significant investment (which we also denoted as sunk cost). Its expected payoff does not simply constitute the immediate cash flows that follow the project start, but also includes improvements in the competitiveness of the business. Therefore, it is possible that we observe situations where the costs of obtaining and maintaining a new technology are higher (and potentially not only as initial investment but also as subsequent maintenance spending) than the immediate benefits but the solution is still introduced and tested.

As a result, we can see options being held and the reasons behind it is that decision makers are either trying to minimize a loss, or they want to maintain a specific technology in light of its future benefits. It should also be noted that in some cases, a fear of falling behind the competition (which uses similar technology) can be a strong driving factor for using it—even if the calculations are not fully supporting the economic benefits of such a decision.

With traditional valuation methods, such arguments are not easy to quantify and justify, but we think that ROA can help in improving the situation. In situations where we are facing a seemingly “non-rational” behavior, like maintaining and not abandoning solutions that are at first glance not profitable, we can use the calculated negative value as a (rough but still more objective) estimate of these concerns. It was also used to measure the indirect impact (provided that financial analysts are rational) as maintaining such a solution would mean that the negative value is at least offset by these positive side effects.

-

In line with traditional option use, an opportunity that is not paying off is not considered to be of use, when its maturity is reached.

The same thoughts apply for situations where options are kept open and specific implementations are not abandoned, even though they seem to indicate a negative value. In this situation, the negative value calculated can also be used as an estimate (lower end) of the external benefits and positive side effects that the management observes.

In situations where it is possible to estimate the additional (indirect) impact more clearly, it can be incorporated in the existing analysis through the dividend yield factor. Going back to the numerical example, if we assume that introducing the intelligent technology will help with other activities, these benefits can be represented with a positive dividend yield. Figure 4 demonstrates the same simulation scenario as before, but this time with 2% dividend yield. This yield represents the potential positive side effects from introducing the intelligent agent. Its own value estimation can be done with the help of experts, or by carrying out additional simulation/scenario analysis on what is the influence of this new tool on the other processes in the organization (for example, being able to detect errors in a faster way that would reduce the time to completion of other processes).

In this case, the mean option values from the simulation are 7.168668 EUR (Crank–Nicolson, left) and 7.168093 EUR (Barone–Adesi–Whaley, right). Clearly the positive side effects will increase the option value, thus also bringing additional arguments for the introduction of new intelligent processing methods. As expected, the value of the option increases and that would also explain the hope and stimuli that new technologies hold. The larger the positive side effects are, the higher the option value, thus containing more incentives to increase investment in it. This fits well into what we observe, where companies are betting on the use of artificial intelligence tools and intelligent data processing in order not only to cut immediate costs, but also introduce positive side effects through the organization. We should note that common estimates like net present value and ROI would still give a negative assessment of this opportunity, even if we add the 2% dividend yield, assuming this percentage applies to the lump sum of sunk costs (in this case, the ROI will be better, but still negative −10.66%). The calculated option value is not just useful as a numeric estimate, it also highlights the importance of flexibility and the need to consider all side effects that introducing new technologies can bring.

5. Discussion

Intelligent data processing methods and large language models aim to be not just a new technology but to present a radical change in the way we interact with algorithms and analyze available data. Their applications have important consequences not only for the immediately affected processes but also for the respective organization as a whole. We have suggested an approach to evaluate both the immediate and indirect impact of intelligent data processing that is based on real options analysis. While there are other means to estimate the importance of new technologies for organizational development, we argue that ROA analysis can provide easy understanding and comparison assessment. By yielding a result that is directly comparable with other monetary indicators, company management can better plan for future development and proactively consider changes in the organizational competitiveness and market position.

We have carried out two numeric experiments in order to compare the valuation of intelligent data processing impact with real options and standard financial metrics like ROI. Assessment results indicate that real options are better capable of handling both flexibility in implementation and potential side effects (expressed as indirect impact).

Accounting for flexibility can improve our understanding and assessment of the true consequences from intelligent methods and tools. While this is a known advantage of the real options, our research has shown that ROA can also capture the indirect impact of new technologies. This is an important result, as it can bring us closer to valuating complex situations, where intelligent data processing influence grows beyond its immediate application and context. Financial analysts and decision makers can benefit from our findings as they can estimate the full impact of intelligent tools and plan for their implementation. On the other hand, service providers can use the suggested valuation to plan for their deployment variants and service plans, using typical use cases as a mean to demonstrate the full benefits of new solutions.

Studies in this direction can be extended further, by accounting for and quantifying the following effects:

-

While there are many appealing characteristics of automated and intelligent data processing, there are also drawbacks related to transparency in argumentation and reasoning.

This is particularly important for decision support systems and complex analytical flows, where small errors or inconsistencies can be crucial for future success. To properly include this effect in our analysis, it would be possible to consider the “loss of skills” or “lack of transparency” penalty factor, that will also influence the option value.

-

With the suggested ROA approach, indirect impact assessment is estimated as a whole, which does not give full feedback where the changes are most relevant and empowering.

To assess in detail the indirect impact, it would be necessary to consider the changes as a portfolio of options, which will increase the complexity of the solved problem, but also provide more specific information that is relevant to individual processes in the organization. Depending on the requirements and goals of the actual evaluation, either a single option or portfolio of option approaches may be applied, though we have limited this paper to demonstrating only the first one.

6. Conclusions

Intelligent data processing and native language interaction with complex models is reasonably expected to revolutionize the way we do business and take decisions. Beyond the appeal of the technology itself, we still need a way to estimate the impact of its application to our existing processes and way of work. As a step that claims to be a revolutionary one, intelligent processing needs time to reveal its full potential and also triggers changes that are strongly affecting business organizations—far beyond the scope of immediate application.

Under these conditions, traditionally used KPIs and success indicators, like, for example, simple ROI, may not be able to completely capture the impact of the new technology. We suggest improving existing assessment by introducing real options analysis in the process. ROA has a long and successful history of correctly evaluating complex investment scenarios that span over large time periods and involve important intermediate decisions.

We have demonstrated that these strengths can be deployed to estimate not only the immediate impact of modern artificial intelligence solutions, but also to capture the side effects. Building upon the existing option valuation models, we have demonstrated how new technology impact can be assessed with either closed-form solutions, binomial (or multinomial) trees, Monte Carlo or combination of these. The ROA approach gives us an opportunity to evaluate, or in the worst case, estimate a lower boundary of the indirect impact of new technology under investigation. This is not strictly bound to intelligent data processing, as the same idea could be applied to any novel technological solution. But it is particularly useful when analyzing artificial intelligence and autonomous agents, as they are expected to trigger a more significant overall change—thus generating positive side effects of large magnitude. The second part of the numerical example that was made contained a demonstration of how side effects can be included in the analysis. If there has been no estimate of the indirect impact, the ROA approach can be used to implicitly evaluate it based on options kept open and exercised.

The ROA approach is not limited to estimating the impact of AI and intelligent data processing solutions, as it can enhance our corporate finance toolset to study various other new technologies. But we believe that these recent developments are especially good examples on the strengths of option-based approach, because intelligent tools are expected to trigger very large positive side effects. This makes valuation with traditional corporate finance methods hard and subject to very strict assumptions. Real options, on the other hand, can better fit in and provide more adequate and precise assessments.