Human-centered AI to build a trustful customer experience in retail

The rapid advancement of technology has changed the way businesses run, people buy, and the velocity at which these activities occur. These improvements continue to have a significant effect on business operations throughout different departments of a retail company. Furthermore, much of the existing retail research has focused on shop design, efficiency, and technology utilization. However, the Artificial Intelligence (AI) adoption in retail is being overlooked. This article looks at the challenges and prospects of AI in a large retailer’s business, and how we can ingest AI in a human-centered convergence, highlights AI obstacles, as well as opportunities, related to Human-centered design (HCD) as an approach to designing AI systems that serve people’s needs.

We want also to close the gap between human factors retailers and AI by giving real-world, practical knowledge about human factors from a business perspective by looking at retail innovation and system uptake. The parameters that determine AI readiness are a controversial topic in retailing nowadays. It’s also worth emphasizing that AI solutions may be tough to adopt due to their esoteric nature, which is tough for regular decision-makers to grasp.

While AI and machine learning are becoming more prevalent and visible on a daily basis through machine translation, voice recognition, picture categorization, and information retrieval, their implementation in businesses is beset by numerous problems. The taxonomy of AI difficulties is shown in Figure 1.

Human-centered design (HCD) is a method that can be used to address this challenge. Early on in the AI process, the data team should apply a human-centered design (HCD) approach to the technology developed.

In order to get started with applying HCD to AI system design, here are six stages. However, your industry, your resources, your company, and the people you’re trying to help will all play a role in what Human-Centered Design means to you.

1. Understand people’s needs to define the problem

Helping retail associates identify problems in their present experiences can help uncover unmet needs. A variety of strategies can be used, including monitoring individuals as they use current technologies and conducting interviews, forming focus groups, and evaluating user feedback. You should involve your entire team, including data scientists and engineers, to ensure that everyone has a clear picture of the people they hope to help with AI. The composition of your team should be diverse in terms of ethnicity, gender, and other identifiers. Think creatively and collaboratively about how to solve the challenge you’re facing.

A company wants to address the problem of sales forecasting and stock level predictions. The company starts by observing the supply chain team, marketers, planners, and other retail staff involved throughout the future sales forecasting process. It also interviews them about the current forecasting process — which relies on published guidelines and human judgment — and shares video clips from the interviews with the entire development team. The company also reviews research studies and assembles focus groups of former predictions. All team members participate in a freewheeling brainstorming session for potential solutions to improve the accuracy and best way to deploy the model.

2. Ask if AI adds value to any potential solution

When you know what problem you’re trying to solve and how you’re going to do it, think about if AI can help, by asking several questions :

  • Is it commonly accepted that the goal you’re pursuing is a worthwhile one?
  • There may be a considerable difference in effectiveness between an AI system and a rule-based solution that is easy to design and maintain.
  • Is the task you’re utilizing AI for, is tedious, repetitive, or difficult to focus on for humans?
  • Is there any evidence that AI solutions have been superior to previous solutions in the past?
    If you replied “no” to any of the following questions, an AI solution may not be appropriate or essential.

Using visual search to recommend the best product you find online or also in physical store. Everybody agrees that speeding up photo input analysis would be a good outcome, since faster embedding extraction, indexing and product search could be more faster than asking a human to recommend the most similar product in stock. The retail agency determines that an AI image recognition system would likely be more effective than a non-AI automated system for this task. It is also aware that AI-based image recognition tools have been applied successfully to review how attractive a product online. The agency therefore decides to further explore the possibility of an AI-based solution.

3. Consider the potential harms that the AI system could cause

In every step of the design process, from data collection and data labeling to model training and system deployment, consider the advantages and disadvantages of utilizing artificial intelligence (AI). Users and society should be considered. Your privacy team can assist you find hidden privacy issues and assess if privacy-preserving solutions like differential privacy or federated learning are acceptable for your organization. Human judgment should be included more effectively in the selection of data, the training of models, and the operation of the system to reduce damages. Don’t develop the system if you think the harms outweigh the advantages.

An online ecommorce company wants to use an AI system to ‘read’ and automatically assign scores to client reviews, and offer them a gift based on their review score, while redirecting company staff to double-check random reviews and to review text that the AI system has trouble with. The system would enable the company to quickly get scores back to clients. The company creates a harms review committee, which recommends that the system not be built. Some of the major harms flagged by the committee include: the potential for the AI system to pick up bias against certain patterns of language from training data and amplify it (harming people in the groups that use those patterns of language), to encourage clients to ‘game’ the algorithm rather than improve their purchase behaviour.

4. Prototype, starting with non-AI solutions

Create a non-AI prototype of your AI system as soon as possible to see how people interact with it. This simplifies, accelerates, and reduces the cost of prototyping. It also provides you with early insight into what consumers anticipate from your system and how to make their interactions more rewarding and meaningful.

Create a user interface for your prototype that makes it simple for users to learn how your system works, toggle settings, and provide feedback.

People providing feedback should come from a variety of backgrounds, including race, gender, expertise, and other qualities. They should also comprehend and agree to what and how they are assisting.

An movie streaming startup wants to use AI to recommend movies to users, based on their stated preferences and viewing history. The team first invites a diverse group of users to share their stated preferences and viewing history with a movie enthusiast, who then recommends movies that the users might like. Based on these conversations and on feedback about which recommended movies users enjoyed, the team changes its approach to how movies are categorized. Getting feedback from a diverse group of users early and iterating often allows the team to improve its product early, rather than making expensive corrections later.

5. Provide ways for people to challenge the system

Once your AI system is operational, users should be able to contest its recommendations or quickly opt out of utilizing it. Set up processes and tools to accept, monitor, and respond to problems.

Talk to users and consider yourself as a user: if you are inquisitive about or dissatisfied with the system’s recommendations, would you wish to dispute it by:

  • Do you want to know how it came up with its recommendation? an here XAI is very important part !
  • Do you want to make a modification to the information you entered? a what if simulation tool.
  • Disabling specific features?
  • Using social media to contact the product team?
  • Taking another course of action?

An online video conferencing company uses AI to automatically blur the background during video calls. The company has successfully tested its product with a diverse group of people from different ethnicities. Still, it knows that there could be instances in which the video may not properly focus on a person’s face. So, it makes the background blurring feature optional and adds a button for customers to report issues. The company also creates a customer service team to monitor social media and other online forums for user complaints.

6. Build in safety measures

Users are protected from damage by safety measures. They aim to reduce unwanted behavior and mishaps by guaranteeing that a system consistently produces high-quality results. This is only possible with rigorous and ongoing examination and testing. Create processes around your AI system to regularly assess performance, planned benefit delivery, harm reduction, fairness measurements, and any changes in how people are actually using it.

The type of safety measures your system need is determined by its purpose and the forms of harm it could do. Begin by studying the list of safety features included in comparable non-AI products or services. Then, go over your earlier examination of the potential risks of incorporating AI into your system (see Step 3).

Human supervision of your AI system is critical:

Create a human “red team” to act as someone attempting to trick your system into unwanted behavior. Then, fortify your system against such manipulation.
Determine how your organization’s personnel can effectively monitor the system’s safety once it is operational.
Investigate strategies for your AI system to immediately alert a human when it encounters a difficult case.
Make it possible for users and others to report potential safety hazards.

Conclusions

This article attempted to bridge the gap between AI and retail by recognizing human factors and addressing relevant, current concerns in retail activities and management. As a result, the study offers practical insights that are relevant to managers, their environments, and their talents. The results emphasize a variety of AI development prospects and potential hurdles in the retail sector. While significant expenditure in terms of capability and re-laying the network will be necessary, shift innovation and system adoption were identified as key hurdles of AI application. So, Even though the reasoning appears to be counterintuitive at first look, increased AI adoption will not decrease the human experience. Instead, it will aid in the scaling and harmonization of continuous monitoring of insights derived directly from hundreds of snippets of different customer feedback. Natural language understanding, together with chatbots, computer vision, deep learning, XAI and other assistive technology, will help to streamline and simplify increasingly fragmented client journeys.

Data Scientist at LVMH, Speaker and Kaggle Expert with interests in the fields of machine learning & Data science