With technological advancements taking shape at a rapid pace, Artificial Intelligence (AI) has become ubiquitous. AI projects have moved from exploratory and PoC stages to initial production deployment stage with actual return on investment being accrued by customers. Industries expect return on investments of more than 30 percent by leveraging AI ⁱⁱ. These benefits are now, no longer, limited to labor arbitrage between human agents and digital agents; it has now extended to risk reduction and improved speed in process execution, apart from several other value-adds.

What is different in AI projects

AI projects are different from traditional software implementation projects in the underlying intent. While software projects are primarily to enable human agents to execute complex tasks, AI projects are more for replicating the work human agents do. So unless the AI project is able to bring about some form of advantage to business, it becomes difficult to justify the project implementation. As a result, the levers mentioned in the subsequent section can be used to see if there is a business case.

Levers of business case determination

The two key costs needed for business case determination are:

1. Cost of AI solution: This total cost - pre-deployment and post-deployment - involves not only the solution or product cost but also the cost of customization, data labelling, model training, deployment, support and machine learning operationsⁱⁱⁱ. 
2. Cost of process execution: The human cost and the tool cost is limited to only the task the AI solution will perform. This will involve the fully loaded cost of the human agent (i.e. the salary of the agent, infrastructure to support the employee and other support costs), the number of hours the human agent works and the software used in the process.

 Using the above two expenses, the cost saving due to implementing the AI solution can be calculated as below:

Cost savings = Cost of process execution - Cost of AI solution

So the different levers for business case identification can be one or combination of the below:

1. Full Time Employee (FTE) takeout: As mentioned above, AI is about replicating human actions and intelligence. Once an AI project goes live, the key part is how many humans the bot will replace. Once the process where automation needs to be applied is identified, the first step is to list how many employees work in the process and the work they do. The second step is to figure the tasks that these employees do that can be done by the AI solution. The third step is estimating what percentage of the employee's tasks can be done by the AI solution. The next step is anticipating fallouts from the AI solution's output that needs to be manual intervention. The final step is to calculate the sum the above costs.

2. Widening the scope of work: Many times, stakeholders in a specific process are keen to implement an AI-led process but do not have a business case in the process, probably due to low volume or limited scope. In such a scenario, it makes business sense to see if the scope of the AI project can be widened to other processes. As the scale increase, two things happen:

a. The overall cost of current processes that the AI solution will impact increases, leading to making the business case stronger

b. As the scale increases, the total cost of the AI solution decreases because quite a lot of service providers like AWS, Azure provide hardware and software using slab-based pricing. As a result, as the volume increases, the cost decreases.

3. Software costs: There are processes where the human intervention is less but the process needs proprietary software for doing work which humans cannot do, one such instance is software being used for executing complex rules. These business rules have been set up in the proprietary software, adding to the software size and complexity. In an AI solution, the solution can itself create these rules and even customize the rules automatically by understanding the customer specific data. These rules can be configured such that machine learning enables these rules to be recalibrated as and when the data changes. In such a situation, the AI solution replacing the proprietary software creates a business case. The underlying assumption is that the complex rules that were built into the proprietary software is built in an automated manner by the AI solution.

4. User experience: AI has started enabling work that humans previously thought of as difficult to execute. For example, processing large number of documents in a short span of time or auditing a huge uptick of invoices during the quarter or financial year-end was considered difficult to be done by human agents. Even if it is possible to have enough human agents to process the large volume, it is still difficult to arrange a large number of human agents for a short span of time and then release them. Digital agents make business case in such scenarios. What the user experience is after the project is executed by the digital agents makes the business case. Since user experience cannot be quantified easily, the business case can be the cost that an organization incurs in case of, say, leakages in payments done or penalties due to lack of proper compliance.

Business case identification has become a key requirement in AI projects as AI projects have started getting into production implementation. The business case is important for both the business and process owners to understand their return on investment and decision-making ^iv. There are various ways to identify the business case as discussed above and it will vary from a case-to-case basis.

References:

i. https://www.horsesforsources.com/business-and-it-teams-collab_092019

ii. https://www.accenture.com/_acnmedia/PDF-77/Accenture-Workforce-Banking-Survey-Report#zoom=50

iii. https://towardsdatascience.com/rethinking-ai-machine-learning-model-management-8afeaa31d8f8

iv. https://www.horsesforsources.com/business-and-it-teams-collab_092019