With the broad range of AI projects and applications, realizing a return on investment (ROI) can vary significantly. Some projects such as augmented intelligence or conversational projects can be implemented fairly quickly, and show immediate ROI. This rapid ROI is proving to be the case with generative AI. Almost immediately, any organization can augment their skills and tasks with the power of LLMs to help with content creation, image generation, social media posts, and similar tasks.
Other projects, such as predictive analytics or autonomous projects, take longer to implement and show returns. These projects take longer because reducing or eliminating the human from the loop requires greater levels of confidence, performance, and accuracy.
What is the return on investment for artificial intelligence?
While every AI project is different and different ROIs are determined for different business problems, generally AI projects tend to fall into different characteristics in terms of how long it takes for those AI projects to show ROI.
A recent AI Today podcast shared insights that projects that have a short time to ROI are ones where the human is not taken out of the loop. For project costs you need to determine if you need humans to fill in the gap for some of these AI systems and keep humans in the loop. If you are looking to have a system that has no humans at all in the process, then the projects will have much greater cost and risk. Having a fully autonomous system will take much longer to implement. It may be easier and faster to use an augmented AI solution than a fully autonomous system.
Augmented intelligence projects enhance human performance and can be integrated relatively swiftly into existing workflows, offering businesses a faster payoff on their AI investments. AI technologies that are meant to do things to help a human do their task better and provide some additional value are faster to implement and faster to realize value. This makes sense because the machine has less of a cognitive load. AI systems don’t have to do everything that humans have to do.
When determining a good use case for augmented intelligence solutions, consider the impact that the system will have on people, especially if things go wrong. If we don’t have a high tolerance for errors and problems, we should probably start with an augmented solution, because a human is in the loop. The second thing to consider is, will we have problems in how we deploy the system, how we operationalize it, how we use this in production? If there’s some challenge with getting this thing out, then maybe augmented intelligence solutions are where to start.
How do you measure the impact of AI?
Additionally, since AI projects are dependent on data quantity and quality, issues with data quality and availability dramatically impact the success and ROI of AI projects. This is why AI-centric project methodologies and frameworks such as CPMAI focus so intently on the data portion of AI projects.
Additionally, you need to consider the issue of labor cost versus the need and the return that you’re getting. You may have higher human labor costs with an augmented assisted system, if there’s humans in the loop. However the non-labor costs for other solutions may far exceed the return.
Examples of augmented intelligence solutions that provide short term to ROI include chatbots, conversational systems, unstructured data handling systems that improve existing processes or existing problems, and generative AI solutions. These are all augmented intelligence solutions where the human is kept in the loop and the system is just providing some help and guidance. These can provide some quick results and positive returns on the investment.
Take an AI enabled chatbot for example. There’s some aspects of this solution that can be autonomous, but at the end of the day if the chatbot can’t provide an answer or gets stuck there is a human that the conversation can be turned over to. Chatbots can improve user satisfaction, decrease costs, and provide 24/7 support.
AI projects with Longer time to ROI
Autonomous AI activities take the longest time to realize any sort of ROI. This is because the goal of autonomous systems is to fully replace humans. There is just no way to do this fast or shortcut your way to an end result without sacrificing safety and performance. Whether it’s physical autonomous systems like robots or autonomous cars, or autonomous software systems and software bots performing processes, these cost a lot to develop so it will take the time to recoup that investment will take a very long time. Autonomy requires intelligence systems to perform at near-perfect levels, so only embark if you have long time horizons for ROI.
As we move to other types of projects covered in the Seven Patterns of AI, we start increasing the time it takes to realize an ROI for the AI project. Predictive analytics systems can provide significant ROI, however, it takes time to validate the results of predictive analytics systems.Predictive analytics can provide significant financial benefit to organizations through cost savings, revenue, and discovery of business opportunities.
Recognition systems can have fairly quick ROI, but it entirely depends on the availability and quality of the training and inference data. Similarly, hyperpersonalization focused systems can have a wide range of ROI depending on the goals of the project and data dependency factors. Pattern and anomaly detection systems can likewise have short or long term ROI depending on the problem being solved and the quality and availability of data. Reinforcement-learning centric Goal-Driven Systems have yet to prove consistency in ROI in the short or long term.
Decision support systems have been shown to help reduce risk at organizations. Additionally, predictive analytics systems can keep an eye on markets and your competition. While it takes a bit longer to show results for predictive analytics and decision-support systems, the high value of ROI is significant.
If you’re trying to decide where to start for your AI project, and want to be able to really get a feel for ROI, start with a project that will be easy to implement and return an immediate value on your investment. If you can afford to wait, predictive analytics or autonomous projects may provide the return you’re looking for with an investment you can afford.
(disclosure: I’m a co-host of the AI Today podcast).
