Why AI Isn’t Working for Everyone

*This post was co-authored by Alex Barnes, Senior Director of Product Management, antuit.ai, part of Zebra. 

In our last post, we explored some of the transformative capabilities of AI, specifically how they relate to retail and CPG industries: AI can rapidly analyze large datasets, automate mundane tasks, and improve business decision-making processes. We also discussed AI-powered demand forecasting and planning and how it can be used to streamline the buying process, manage in-season markdowns more effectively, and provide accurate demand forecasts. And we noted how all of these AI-powered efficiency gains may ultimately translate into stronger margins. 

Next, it’s important we look at the data science behind AI because, at some point, you’ve probably asked, “How does AI work?” You may still wonder today—mainly because everyone is talking not only about what AI can do, but what exactly is the “techno magic” behind it.

Before we talk about what happens behind the scenes, let’s address the issue at the forefront of nearly every AI discussion these days: ethics.

In light of many recent ominous headlines, there’s growing trepidation surrounding AI – so much so that people are calling for government intervention. By understanding the concerns and using what we can leverage from the emerging data science, we can develop both confidence and discernment when utilizing AI.

So, let’s delve in. 

The Issue of AI Ethics

Recently, there have been several articles that call into question the ethics of AI. Take this article, describing a situation in which ChatGPT (wrongly) named a law professor among a list of colleagues who had allegedly sexually harassed a student, citing a supposed March 2018 article in the Washington Post as its source. The problem is that the original Washington Post article about the law professor and this supposed list didn’t exist. The AI had grossly misperceived its own automated research—potentially putting that professor (and likely several others) at immeasurable risk to their professional and personal reputations. 

In this scenario, who do you hold accountable? How could an AI-based application cite something that never existed? Is the risk of misinformation being spread —allegedly by you or about you—reason to avoid AI?

These are all important questions to ask and answer. But before we do, let’s consider a more subtle example of the same AI’s struggle with presenting accurate facts. 

It’s actually from a colleague who prompted ChatGPT to find an optimal reorder point. The answer the generative AI gave was close, and to a non-retailer could have appeared correct, but our colleague recognized it was slightly off. 

Does that mean that AI should not be trusted? 

Not at all. But it does mean that you should understand that there are different types of AI models and there are things that each AI model will excel (and fail) at. While generative AI may be good for some things, adaptive AI might be best for business applications. Even within these broader AI categories, there are types of AI that may be more beneficial when you need background information compared to when you want certain decisions or actions to be automated. 

We know that ChatGPT is not ideal for calculations but seemed to have been able to generate reasonable reasoning for this problem given the likelihood that it had encountered similar problems in its training data. Therefore, it could potentially be consulted with the understanding that its logic is limited to its training data set.

With this in mind, how would you ensure that you’re…

• Choosing the right type of AI for the task at hand?

• Getting accurate answers from the AI that don’t cause you, at minimum, to reveal a lack of expertise, or at worst, falsely accuse someone of a crime?

The first question is best answered by consulting with AI experts, though Zebra’s Senior Director of Artificial Intelligence and Advanced Development, Stuart Hubbard, does a good job explaining why adaptive AI (versus a generative AI like ChatGPT) will probably be the type of AI you’re going to need in this blog post:

The Truth About AI: What Mainstream Media Coverage is Missing

Regarding the second question, just know that the data you put in is the key to accuracy and confidence in the tool that you’re working with. As they say, ‘garbage in, garbage out.’ If you don’t give it the right data, or clean data, to work with, the end product will be skewed and provide inaccurate guidance. 

’Know Before You Go’—Can AI Work for You?

To understand AI or machine learning (ML), a form of AI, first consider that it’s not a “one-size-fits-all” tool.  As noted in this article in Dataconomy, there are several types of AI models, including supervised learning, unsupervised learning, and reinforcement learning. These models enable AI systems to recognize patterns, classify information, and make decisions based on data inputs. 

Another article, from DZone, reviews the top 10 most popular AI models— the algorithms and mathematical representations that allow machines to learn from data and make predictions or decisions. As DZone explains, "by selecting the appropriate model for a particular task or application, developers and data scientists can create powerful AI systems that can automate tasks, improve efficiency, and provide insights that would be difficult or impossible for humans to obtain through manual analysis."

To select the right model, there are several key topics to understand:

1. Data collection and preprocessing: Before we can build AI models, we need to gather relevant data and clean it. This may involve handling missing values, dealing with outliers, and normalizing data to ensure that the input features are on a similar scale.

2. Exploratory data analysis (EDA): This is the process of visualizing and summarizing the data to gain insights, identify patterns, and formulate hypotheses for further analysis. EDA helps us understand the underlying structure of the data and informs the selection of appropriate machine learning algorithms.

3. Feature engineering: This involves the process of selecting and transforming the most relevant variables, or "features," from the raw data to improve model performance. Feature engineering may include techniques such as dimensionality reduction, feature scaling, and feature extraction.

4. Machine learning algorithms. These are the core of AI systems and can be divided into three main categories – supervised learning, unsupervised learning, and reinforcement learning. Some popular algorithms include linear regression, decision trees, support vector machines, clustering algorithms, and neural networks.

5. Model selection and evaluation: Choosing the right model and evaluating its performance is crucial. This involves splitting the data into training and testing sets, selecting the best model based on performance metrics (e.g., accuracy, precision, recall), and fine-tuning the model to achieve optimal results.

6. Model deployment: Once a suitable model has been developed, it must be deployed for real-world use. This may involve integrating the model into existing systems, monitoring its performance, and updating it as new data becomes available.

7. Ethical considerations: AI systems have the potential to impact society in significant ways. It is important for data scientists to consider the ethical implications of their work, such as fairness, transparency, and privacy.

However, if your team doesn’t ask the right business questions or you focus on inaccurate or irrelevant data, the results may likely be frustrating, —and AI won’t perform to its full capabilities. Here are some hypothetical examples for CPG and retail concerns:

CPG:

1. Incorrect data inputs: Your team may try to use AI to predict consumer demand for every location and SKU, but if the data inputs are incorrect or incomplete, the predictions may be inaccurate. For example, if you don’t consider local events or promotions that may affect demand, the AI system may not be able to accurately predict demand in that location.

2. Overreliance on legacy solutions: You may be using legacy solutions that aren't able to keep up with the demands of modern business. If you rely too heavily on these solutions, you may miss out on the benefits of AI. For example, if you are using a legacy demand forecasting solution that can't handle large datasets or provide real-time updates, you may miss out on opportunities to improve your margins.

3. Lack of understanding of AI: We speak with leaders of many CPG companies who do not fully understand how AI works or how to use it effectively. They may not understand the importance of data quality or how to select the right AI model for a particular task. So, make sure you and your team do your due diligence, which can be as simple as calling my team. We’re happy to walk you through anything AI related and answer any questions. 

Retail:

1. Failure to ask the right business questions: If you’re not asking the right business questions, you’ll end up with inaccurate or irrelevant data inputs. As a retailer, if you’re trying to predict omnichannel demand for all products across channels and time, but you don't consider changing market conditions or customer preferences, the predictions may not be useful.

2. Inaccurate data inputs: If you’re trying to ensure that you have the right product, at the right place, time, and price, but you’re relying on inaccurate or incomplete data, you may miss opportunities to improve their margins. In fact, many retailers do – mainly because they are working with incorrect data from the start.

3. Lack of integration with legacy systems: If you are using an outdated inventory management system, you may not be able to take advantage of the benefits of AI-powered demand forecasting or pricing optimization because integration with the AI system will be difficult, if not impossible.

To avoid these pitfalls, whether you’re running a retail or CPG operation, it's important to approach AI strategically and with a clear understanding of your business needs. Take steps to ensure your data inputs are accurate, that you’re using the right AI models for their specific tasks, and that you’re integrating AI effectively with their legacy systems. If you don’t know what those right steps are, call someone who does. Again, our team is happy to help guide you through this process and share any information you may need to feel comfortable even starting down this path to an AI-informed operation.

###

Editor’s Note:

Stay tuned for upcoming blog posts from Nick and Alex, as they plan to explore practical case studies in the retail and CPG sectors to highlight AI's significant impact on a company's profits. They’ll analyze the essential elements that businesses, specifically CPG and retailers, require to implement, adopt, and execute AI technologies effectively.

In the meantime, if you want to learn more about whether AI and ML solutions could potentially help you meet your business goals, contact the antuit.ai team here. 

You can also read their last post here:

AI 101: Automating the Mundane