The retail sector is sizzling at the frontier of technological innovation. More than just enhancing operations, retail is using today’s most advanced technologies to fundamentally transform them.
Central to the next wave of this transformation has been the arrival of generative AI-powered systems, which have introduced levels of creativity across product design, marketing strategies and customer engagement that were unprecedented before and are now pushing the boundaries of what's possible in retail.
Where are we headed? What’s even working well right now? To help get some orientation around these questions, we recently spoke with three PwC leaders in consumer markets about the state of play around these latest developments and what your organization can start doing to help achieve better outcomes and work toward increasing business value.
Ian Kahn, principal, Salesforce practice and alliance leader, PwC U.S.: As we know, AI and machine learning have been transforming retail for years by analyzing customer data, providing tailored product recommendations, targeted marketing campaigns and much more. All that’s only getting better, but now GenAI is taking that to a new level.
One way I like to characterize it is by pointing to its creativity: GenAI is generating initial concepts for personalized marketing content, prototype designs and even new and customized products and services. Traditional AI couldn’t do this. Take packaging as an example: GenAI can prototype designs for me specifically tailored around the ESG criteria I choose, like decreasing material or reducing waste or recyclability.
Katrina Carrizales, consumer markets sector leader, cyber, risk and regulatory, PwC U.S.: Yes, exactly, and it’s certainly something you’re now seeing in the supply chain area too. Here, retailers have been using machine learning and complex data sets for years to do things like forecast demand, respond to trends in real-time and better manage stock levels.
One big boost you start to get with GenAI are the detailed simulations of supply chain scenarios. I’m talking about “what-if” analyses for extreme conditions or unprecedented market swings. This also means you’re testing the resiliency and efficiency of your supply chain strategies in a virtual environment that’s much more built out than ever before.
Bret Greenstein, principal and generative AI leader, PwC U.S.: Another area — and this is a big one — where you’re seeing retailers use AI is in customer service and support. I think most people have encountered an AI-powered chatbot or virtual assistant in some way.
The next step — which is already happening in some cases — is when these assistants can create initial drafts of human-like responses to customer questions, provide customized shopping tips and deal with complicated customer service issues. Before long they’re going to do this with precision and empathy, which can have a significant impact on the customer experience.
It’s essential to avoid the use-case trap when trying to scale GenAI. Instead, prioritize “patterns” that scale rather than those limited to isolated instances. GenAI’s capacity to unlock insights from unstructured data — such as text — can make knowledge workers better decision-makers.
Carrizales: First, you should establish clear guidelines and policies. This starts with a Responsible AI approach to help manage the risks associated with an AI-based solution and prepare for coming regulation. You also should have an AI governance framework aligned with the company's values, ethics and legal obligations. This should be a trust-first approach that accounts for data privacy, copyright issues and accuracy.
And you should bring everyone in on this: legal, compliance, IT and business teams. Cross-functional collaboration is what it takes. This is where you’ll define roles and responsibilities, set standards for cybersecurity and outline processes for AI development, deployment and monitoring.
Kahn: It’s worth noting that this really starts at the top. Since AI’s likely going to permeate the organization, CEOs and boards play a critical role. They’re setting the tone and defining the organization's vision and goals.
I think transparency’s especially important too. It’s about being clear around your objective, whether that’s improving customer experience, supply chain or personalization. What CEOs are saying is fundamental to building trust and confidence with stakeholders.
Greenstein: This might be a little more specific, but I’d really emphasize prioritizing data quality and integrity. These systems rely on the data they’re trained on. Strong data governance practices when it comes to data collection, cleaning, labeling and storage have never been more important.
Investment in data infrastructure and security measures is really a baseline to enter this space and to maintain customer trust.
PwC’s 2024 U.S. Responsible AI Survey states that only 11% of executives report having fully implemented these fundamental responsible AI capabilities — and we suspect many are overestimating progress.
Adopt a Responsible AI framework to help enhance governance across your company, promoting transparency, risk management and alignment with organizational values to foster trust and accountability. This approach should expand to cover diverse use cases and include a holistic risk taxonomy to manage everything from procurement and third-party risks to security and compliance.
Greenstein: We have to remember that these systems are still black boxes. The AI specialists are still spending an incredible amount of time getting better at understanding how these answers are determined and where they’re coming from. That’s what interpretability and explainability are about.
In retail, a key area is being able to explain to your customers how AI is being used to personalize their shopping experiences. If an AI system recommends a certain product, where is that coming from? How does it know you might like that? How much else might it know about you? The retailer should have clear explanations of why those recommendations were made, like being able to highlight specific customer preferences or past purchase history.
Carrizales: To build on what Bret said, our Responsible AI Toolkit is really a collection of solutions, processes and frameworks, all of which can be customized to how an organization wants to use AI. It emphasizes making informed, deliberate decisions when building and monitoring the ongoing use of your system — Who’s accountable? Are we safeguarding privacy? What’s our approach to identifying risk? — so you know what’s going into the AI when it gives its answer.
This is the sort of transparency that can help customers understand that their data is being used responsibly and that AI is enhancing their shopping experiences rather than manipulating their choices. Again, so much of this is about building and maintaining trust with customers.
Establish and lean on your business’s Responsible AI approach to enhance transparency and build trust by deploying solutions that promote the interpretability and explainability of AI systems in retail. This can help clarify how AI personalizes customer experiences, making AI operations more comprehensible and accountable to users.
Greenstein: The question here is where are these bias and fairness-related concerns coming from? In one sense, they’re coming from the data the model’s been trained on. Rooting out bias starts at the data collecting and preprocessing stage — making sure the data itself is diverse and representative to reduce risk.
Carrizales: Diverse, unbiased data is often critical. But bias can also creep in through the people designing AI models and figuring the weights that determine the particular results. And this can be a really subtle, unconscious sort of thing. They’re not intentionally introducing bias.
This is why an essential part of the responsible AI approach means diversifying your teams — in addition to testing rigorously and maintaining clear governance and controls.
Kahn: A significant part of this comes down to regularly monitoring, auditing and improving your models. AI systems are constantly evolving and so bias that wasn’t there yesterday could easily pop up today.
This is the other thing that’s hard to get right: It’s important that this vigilance isn’t an add-on that comes later. It should be built into your processes and frameworks from the start. Otherwise, it’s always out of sync and arriving too late.
One avenue for retailers is to seek feedback from users and customers. Build in mechanisms that allow users to challenge or appeal decisions made by AI systems. Also consider updating codes of conduct and acceptable use policies to create channels that facilitate the reporting of new risks.
Carrizales: Well, certainly it starts with having strong cybersecurity systems in place — firewalls, encryption, intrusion detection systems. These are the sorts of proactive measures we’ve had in our toolkits for years. But with GenAI it really feels like we’re taking some potentially new steps in security.
For example, GenAI models can simulate sophisticated cyberattacks that haven’t been seen before. This allows organizations to test their defenses against a broader range of hypothetical scenarios and it represents a significant advance in conventional AI's reliance on historical data to predict threats.
Greenstein: GenAI has an advantage over conventional AI tools in dealing with reputational risks, because it can comprehend language and context better and more deeply. The latest AI models can detect small differences in sentiment, like sarcasm and irony. That can help them produce more contextually appropriate insights.
Building on this will likely involve using GenAI to help create sentiment-conscious responses and content, allowing retailers to communicate with customers more empathetically and effectively.
Don’t start from scratch. You can build on established governance, privacy, cybersecurity and compliance programs as you roll out secure, responsible AI practices across your organization. Remember, even if you don’t have to reinvent the wheel, it’s critical to stay on the lookout for new risks.
Greenstein: Responsible AI requires educating employees at each level. This involves creating training programs to improve AI literacy, then transitioning mundane, manual work to chatbots and virtual assistants to allow employees to focus on more complex tasks.
But it also involves organizational values related to AI, possible biases and the need for human supervision, enabling employees to detect and report potential problems. And, as we discussed, these are systems that should be constantly monitored and adjusted.
Carrizales: You should also regularly train employees on the latest cybersecurity threats and industry-leading practices. This can include awareness of GenAI-powered phishing attempts, social engineering tactics and what it means to maintain high levels of vigilance for sophisticated fraud attempts.
Kahn: On the efficiency and scale part, it’s mind-blowing. We’re talking about personalization at a massive scale, with retailers effectively delivering bespoke content and experiences to a near-limitless customer base. Then extrapolate that to GenAI’s ability to help automate and streamline content creation processes — generating high-quality first drafts of product descriptions, ad copy, social media posts and email marketing content that can then be enhanced during human reviews — all this at an unprecedented speed and scale.
We know this, as we are “client zero.” Our agreement, between OpenAI and PwC US and UK, makes us the first reseller for ChatGPT Enterprise and the largest user of the product — enabling us to further scale AI capabilities across our businesses and help clients to do the same.
Unlocking value means having your people use GenAI to reimagine the way they work. Put your people first and provide incentives for innovation. When employees innovate their roles with AI, reward them with even greater opportunities.