Connect with us

Thought Leaders

Five Reasons Why AI Algorithms Can Be Difficult to Implement in Operational Management




By: Evgeniya Malina, Head of Business strategy and operations at Food Rocket.

According to recent research by the McKinsey Global Institute, AI is poised to boost global economic output by $13 trillion by 2030.

However, this comes with its own challenges and unintended consequences. Some of the most commonly identified risks and challenges of AI implementation include privacy concerns, the inability to generalize, and an overall lack of trust.

Below are three challenges with AI that specifically affect operational management for businesses.

1. Poor Data Quality

The number one thing that a robust AI algorithm needs is data. To properly train an algorithm, you must feed it vast quantities of precise, top-quality data. Unfortunately, it's not always easy to obtain this data, and a 2020 Gartner report notes that poor data quality can cost your company around $13 million every year.

For example, some processes may not have any digital footprint at all when you're first starting. There's no data for you to feed into an algorithm in these instances. Everything you give is just hypotheticals and educated guesses, which poses two problems.

First, this introduces significant human bias into your process straight from the start. Second, it means that any results from the algorithm are simply an extension of your best guesses. Ultimately, this leaves you with a sketchy data landscape and an unreliable, unstable decision-making process.

2. Navigating “Cold Starts” and Employee Engagement

Automation is excellent for streamlining existing processes, but the tradeoff is the “cold start.” This is when you must begin a process with no historical data on which the AI can base its routine. In every instance, AI will struggle to overcome this hurdle.

According to Harvard Business Review, 80 percent or more of an IT team's time is often spent trying to improve and refine inconsistent data for AI algorithms

It often takes a considerable investment of human effort to help the AI over this “cold start” hump and resume smooth operations.

In my experience, it can cause serious disruption in supply management, and it can cost companies considerable revenue too. We all know that AI isn't evolved enough yet to handle all aspects of an operational management system. This means that any AI solution your company utilizes will overlap with human decision-making processes.

While this can be a good thing, it can also lead to a disengagement of an employee's sense of personal responsibility. In some cases, employees feel like they can separate themselves from a decision because “the AI did it.”

In addition, it’s common for the introduction of a new algorithm to coincide with a significant drop in quality metrics. In my experience, this paradox is a result of a person who was formerly responsible for the metric feeling like they’re now simply an unimportant link in a chain of automated decision-making.

It's essential to manage this aspect of automation because of how easily it can lead your team down a path of disinterest and lowered commitment. It also has the potential to cause harm to your brand. If AI is left on its own to make decisions, it may unwittingly begin discriminating against customers in certain age, gender or geographical brackets.

3. Challenges With Transparency and Efficient Implementation

As every business owner knows, things can change in an instant. Companies don't always have time to build a complex AI solution for a new operation.

In fact, it's far more common for businesses to be on a time crunch and be forced to solve a problem without the help of automation because setting up a new process simply takes too long. Since there typically isn’t time to write complex models, one of two things happens.

First, the business might choose to implement a mostly complete process but insert a manual intermediate step until the process can be refined. In this case, businesses lose as much as 80 percent of the calculated efficiency of the process.

Alternatively, the business might deploy SaaS to speed up the implementation process. While time and money costs will be lower, the problem of efficiency loss remains. In this case, implementing SaaS algorithms that aren’t specifically adapted to the company’s needs can make a process less efficient than if it was done manually.

In addition to these problems, it’s important to understand that transparency in the AI process is incredibly difficult to communicate to management, even by experts. This is due to the complexity of the algorithms, but it can make your team feel reservations about transitioning to automated operations management.

Where Do We Go From Here?

Some researchers suggest that the challenges we currently face will lead to new human roles in a company: Trainers, explainers, and sustainers.

Trainers will help optimize AI performance; explainers will be tasked with breaking down AI decisions for non-professionals, and sustainers will work on making AI processes sustainable for the long term.

However, until then, businesses and founders must consider more than just the significant competitive advantage that AI can give. The advantage must be weighed against the ambiguity, time costs and growth hurdles that come with operational AI.

Artificial intelligence still has a long way to go in terms of growth, development and implementation. It can undoubtedly make a huge difference in operational management, but we cannot yet fully rely on it to always be the best option.

Evgeniya Malina is the Head of Business strategy and operations at Food Rocket. Evgeniya graduated from University College London and earned a master’s degree from the Queen Mary University of London. She is responsible for automating the company’s operational processes and building the foundations to scale. Thanks to her expertise, the IT team of 20 people has built a proprietary warehouse management system, which has now been fully rolled out across all dark stores and the recently launched Distribution center.