We are not yet to the point where super intelligent machines can act completely on their own, replacing the need for people. However, there are already quite a few real-world use cases where AI technology matches or outperforms humans. As of now, AI lacks the ability to strategize, build complex business models, shift goals or come up with many of the other tactical, value-driving ideas that humans are capable of.
That said, AI already carries with it a considerable amount of business value, as it can significantly save workers’ time by doing mundane, repetitive tasks. Such tasks can be more efficiently and accurately carried out by AI, allowing employees to invest their energy in more creative and thought-provoking assignments.
But how can we measure the results and the value of an AI solution? It can be complicated for businesses to project and quantify the expected outcome of an AI solution, as such software is often unique and special, learning automatically to solve specific issues. In this article, the Exadel AI Practice shares the best ways to measure the ROI of AI, including the metrics of returns and costs.
How to Get ROI out of AI
If you want to get the maximum ROI out of an AI project, first single out the value it can bring to users, determine its competitive advantage and define the financial result expected. The discovery, estimation and development of a proof of concept can be helpful to determine if adding AI is valuable.
Generally speaking, metrics on returns for an AI project may be similar to the metrics of any IT project: it is important to cut costs, improve certain business processes, increase sales or save money. Before an AI project can start, take into consideration certain risks and expected costs as it may be hard to know when and where AI is going to bring revenue.
Costs Associated with AI Projects
These are the expected costs for an AI project:
- Data preparation
Data is fuel for AI. Data plays a significant role in AI projects and there is often a lot that needs to be done with data prior to deploying AI. Companies can easily underestimate the amount of time they will need to dedicate to data preparation. Without clear and structured data it is impossible to make an AI solution work. Businesses need to be aware that many of the costs associated with an AI project go to preparing data: data gathering, data validation and data cleaning. Even with the necessary data in place, an organization may spend 60 percent of its time working to understand, analyze, filter, and clean the data, leaving only 30-40 percent of the allotted time to create the actual AI solution. If the company begins with zero data for their AI project, even more time may be spent to accumulate the necessary information and write a system.
This one is simple: if AI implementation changes the processes in the company, it may take more effort to educate the team to work in the new environment.
- Cost of tools
The hardware, software and third-party services necessary for an AI project are often more expensive than other innovation-based projects. These expenses may go toward cloud solutions such as AWS Sagemaker, Azure Machine Learning Studio, tools (Alteryx, Dataiku) and hardware (GPUs and TPUs). These help to integrate AI solutions but they don’t come cheap.
- AI implementation
AI implementation or integration is not an easy task. Implementation of an AI solution requires skilled and educated employees who may be in scarce supply. Furthermore, developers with these skills tend to cost more than those without AI implementation knowledge.
With these costs in mind, you can begin to look at ROI. When developing an AI project it is important to take a look at the metrics that will best determine value.
Metrics for ROI
- Employee hours → decrease
As AI replaces some human tasks it should help with quicker analysis, categorization, and decision making, reducing employees’ time spent on certain processes. For example, in healthcare, it may expedite a diagnosis or analyze DNA according to certain markers and symptoms that may not even be visible to a human. In e-commerce, AI can provide recommendations on what to buy based on the user’s tastes and chosen filters. This significantly helps retail organizations save employees time who otherwise could not easily provide recommendations to thousands of visitors each day. Essentially, AI should significantly save time on analysis and decisions.
- Employee productivity → increase
If employee time spent on tasks decreases with the help of AI, it does not always mean that the headcount should decrease as well. As AI helps with decision making, employees can take on new, more complex tasks during the same time period.
- Number of routine tasks → decrease
To ensure a ROI, the number of routine tasks done by employees should significantly decrease. For example the AI technology Robotic Process Automation (RPA) is widely used to help organizations cut down on routine operations. RPA is widely used in banking, financial organizations, manufacturing companies, etc. It helps to get rid of bureaucratic hurdles and improve the flow of operations. It also helps to handle defects in manufacturing.
For example, in banking, fraud analysis in large scale operations is still done manually in some instances. Employees check every operation and determine whether or not it can be considered fraud. AI has the ability to check all operations in several seconds, freeing workers from the time consuming routine of these manual tasks.
The Bottom Line
All in all, to ensure you are getting value out of AI, start with something simple. Take for example, a recommendation system on a streaming platform. It may first simply recommend the most popular films of the day or films with the most clicks (which is actually statistics, not AI). However, based on this data an AI solution can be made to provide recommendations on what somebody might like, based on other users’ preferences. With the statistics gathered you can analyze if the number of users increases or not and if there is value being generated. Then, by accumulating more data and making the process more streamlined it is possible to create an automated AI solution that recommends films based on individual interests.
It is great to start with simple algorithms that may not be connected with AI, but are still able to show if the idea is viable, whether the users really need an AI solution, and if there is real value in it. Only then should you start gathering data for AI, because starting with a more simple approach can help you achieve ROI for an AI project.
Exadel has extensive expertise in developing AI solutions, ensuring ROI, and providing business value for our clients. We’ve developed an AI solution for classifying poker chips, an automated product management solution with AI that helps to find info on labels, created Face recognition services for office security that detects intruders, implemented a chatbot platform, and developed a technical support task classification plugin that uses NLP.