Building your case: calculating ROI and TCO

When you are considering an AI workload in your business, there are many factors you must consider as part of the process. This section of the book will give you some helpful hints as you strive to deploy solutions to do the following:

• Recognize areas where you may be able to take advantage of AI easier than in other areas by understanding the complexity of deployment of each of the solutions.
• Calculate the ROI, because we all know that we have a boss who will want to know what it is going to take to recoup what is spent on this project.
• An overview of what you can expect your TCO to be when the solution is fully deployed, and calculating cost growth as your solution grows.
• How to pull all the necessary information together to ensure you have a complete solution that will bring value to your business and make you an AI hero!.

When we begin to discuss the areas where AI and cognitive services can be of benefit to an organization, we really need to put on our business analyst (BA) hat. Or, perhaps a BA has already started the process of identifying key areas where there are major inefficiencies in your organization. Each chapter of the book will include some helpful hints on how to be sure you are optimizing the services for cost and simplicity. That said, you will need to evaluate the value of implementing a cognitive service to help streamline a process, reduce human intervention, or be able to outline the benefit your organization will receive as a result.

Building your justification

There are costs associated with the implementation of these services, and without a full analysis of what the TCO and ROI are for the implementation, it may be very challenging to justify the out-of-pocket expenses required. Try to avoid deploying technology for the sake of technology. What I mean by this is that too many technologists really want to embrace new technology because it is "cool" or "exciting" to learn and build something new.

As an alternative scenario, you may have an executive or key stakeholder from your organization who really thinks that by implementing "x" technology, the business will be greatly enhanced. This is a dangerous scenario to have to deal with when it comes to your career at your organization, and you will have to be very careful with how you proceed. In reality, there may not be much value for the use case they are pursuing, or the data available for the use case isn't clean, accurate, or good enough. In situations such as these, you will have to decide whether it is worth your reputation to push back, or even go forward with, such an implementation.

To avoid the possible fallout from a failure, it is advisable that you build your TCO and ROI models with as little conjecture as possible. Break down the use case into simple math with parameters that can help you justify your position either way. Present your findings to the stakeholder and state your position. Let that person then make a decision on whether to proceed, hopefully reducing the pressure on you and the team you are working with.

In either case, you need to get a starting point for how to build these models. Here are some thoughts on ways to get started with the data you will need for doing so using a Form Recognizer scenario to automate the accounts payable process. Be sure to work with someone in finance or accounting to understand what the average overhead is of one of these individuals to build a baseline of your averages:

• How much, on average, does an accounts payable clerk cost the company per hour?
• How much time, on average, gets spent per week on manually entering the details of that accounts payable transaction into whatever system is being used for paying vendors?
• Is there already a process for capturing forms in a digital format, such as scanning to a repository? This could add additional costs to your calculation if there is not, or if equipment is required for such a process to be implemented.
• What are the costs required to insert data into the system, and with how much confidence can it be inserted? Are you getting handwritten slips, as in the seafood industry, making it harder to be accurate, or is everything typed?
• Is there already an electronic data interchange (EDI) process in place, and the only human intervention required is validation? Find your alternate scenarios and boil costs down in the same way so that you are sure to capture the full picture.
• What is the risk or cost if payment of an incorrect amount is made or it is made to the wrong vendor? Many of these systems can be built with very high accuracy; however, they are not completely infallible, so knowing what the downstream effects are as a result may need to be factored in.
• What is the cost of the development time required to build the solution? Can you put together a sample project plan with approximate timelines to evaluate development expenses?
• What will the solution cost the organization after it is implemented? There will be recurring costs, which we will cover in each chapter, that will need to be factored in.
• Do you have clean enough data to build an ML model to run against when processing invoices for payment? Good, clean data is critical for any deployment of AI in any scenario.

This list should provide a good starting point for where you can begin to evaluate what will be required when building out your cost models. Work through the costs and the savings to build the complete model and understand how long it will take to recoup the costs that are required to the point where you start to save the company money. Is it 3 months or 3 years? That break-even point can be a critical point that helps your solution float or sink, so be sure to have a trusted advisor look at your completed analysis.

Proving out your solution

After a complete evaluation, if you still feel it makes sense to push forward with building the solution, you will also likely have to build a proof of concept (PoC) for demonstration. Microsoft provides many tutorials for getting started with the technology through their documentation and GitHub, or you can use the solution we provide as your baseline with relevant documents for demonstration.

You have built your models, checked them with another trustworthy source, built a PoC, and tested your demonstration, so now, it is showtime. Be mindful—there is certainly no exact science to building a perfect presentation for management, as each team will have its own priorities. However, the goal is to try to make sure that you are as prepared as possible to be able to start one of these projects and build in some alternatives or answer those curveball questions that come at us when we are being challenged. The steps in the preceding list should hopefully prepare you for these challenges and will hopefully help you avoid chasing after a solution that won't provide as much value to the organization as you or the stakeholders originally thought.

Of course, there is always the possibility that the budget may not be available for any investment. If we believe there is still a compelling case to be made and it is worth pursuing, this is when we should flip the conversation to "this is how much revenue we believe can be expected as a result of implementing these solutions", if this is feasible. There is certainly a softer side of this conversation, as sometimes we need to evaluate best-case scenarios where maximum value can be achieved, without having the hard data to back it up. If "selling" your solution isn't one of your core skills, enlist a teammate, friend, advisor, or sponsor that you trust and see if they are willing to join your crusade of helping your company be more efficient. With time, the technology will improve and the costs will decrease for implementation, so maybe it is best, instead, to wait for the next budget cycle, possibly? Use your best judgment on how to have the most impact on your organization.