Printable version here

ChatGPT constantly made the headlines for the past 9 months and its adoption was massive (it reached 100m monthly active users in January 2023, 2 months after its launch. For comparison, it took 9 months for TikTok and 2,5 years for Instagram to reach the same amount of MAU). However, ChatGPT is just the tip of the iceberg, more precisely a ‘demo’ product showing to the general public the capabilities of Generative AI (a system able to answer queries, or ‘prompts’, after having been trained on large amounts of data, and therefore capable to bring answers which are going further than simply mimicking the data it learnt).

Generative AI has already started to disrupt many industries and retail is no exception. This is the reason why IADS invited Cyrille Vincey, Partner in Advanced Analytics at Bain & Company, to provide IADS member CEOs with more information and give a few examples of use cases in retail.

Cyrille Vincey has been in data science for 25 years in various industries, always obsessed with bridging business matters on one side, with technology and data science on the other.

After having started his career at a software publisher, applying operational research and graph theory approaches to supply chain optimization, he founded a data science startup, later acquired by an e-commerce player. He then took on a CDO role and shipped a large-scale audience-sharing platform used by 8,000 e-retailers.

As a consultant, Cyrille has been working primarily in the retail industry and assisted on most of the tech M&A deals in Europe, including ad tech, digital trust and enterprise software. At Bain, he fosters the next generation of data-enabled consultants and works closely with Open AI in the framework of the Bain x Open AI alliance.

He defines himself as a ‘data nerd’, a ‘tech guy’ who has been in tech entrepreneurship for 25 years and half-jokingly opened his presentation by mentioning that, now that AI is a commercially available product, people like him are not needed anymore, as businesses and organizations now only need software engineers to plug pipes and coordinate the new AI-powered tools available at hand.

Generative AI: landscape and perspectives

What are we talking about?

Generative AI, and large learning language models (LLMs), represent a new paradigm of artificial intelligence, which unlocks advanced capabilities to replicate human capabilities (perceive & understand, communicate & create, reason & plan, act & use tools). For instance, ChatGPT can complete a poem when fed with the beginning of it, based on the probability of the words’ occurrences.

The 2 root causes for this breakthrough are a 2017 research paper from Google describing a new architecture for deep learning models and the fact that IT costs for training this new type of model have been consistently falling for the past 3 years, by a factor of 100 (even though a training iteration for a large company would still cost EUR 20m today, that’s still a bargain when compared to the cost 3 years ago).

Vincey remarked that while OpenAI made a big splash with ChatGPT, the winners of this new arms race might be very well Google and Microsoft, who have the pipelines needed to integrate and scale this tech, make it available to the general public, and disrupt the way companies operate by providing them with new tools.

He also stressed the fact that ChatGPT is just a demo layer of a larger model, GPT, a text-to-text model. There are also other OpenAI products: Co-Pilot (text-to-code), Whisper (speech-to-text), Dall-E2 (text-to-image), Clip (image-to-text). This product portfolio will be questioned in Q4 2023 with the release of a general model proposing a multi-modal convergence (users will be able to use whatever form of input, text, image, speech or code, and to choose the form of the output they want as well). Vincey is convinced that this first general model will encourage the appearance of new applications still unheard of today. For instance, an entire PDF can be analyzed by this new model, and relevant parts can be extracted for use in a specific given context.

An impactful breakthrough in all industries

These new-gen AI-powered tools are fundamentally built with a general purpose in mind, while the previous models (IBM Watson, Einstein…) were built with a specific purpose. Generative AI models can be used out of the box by companies and applied to every internal use case. Therefore, AI innovation projects within organizations now take weeks, if not days, and not 6 months anymore like they used to. This allows industrializing innovation and experimentation with very limited time and costs. Companies who are late in AI-powered decision-making process investments, use Gen AI to leapfrog and do quite advanced things.

Generative AI does not, however, replace the existing quantitative AI-based decision models relying on machine learning and operation research, as it is not, by no means, a one-size-fits-all solution. Instead, it completes them by providing the possibility to bridge the decision taken thanks to the quantitative model, to stakeholders (partners, suppliers, clients, employees…), with adapted context. Gen AI is the ‘last mile’ of the decision-making process. For instance, if a company has invested in a quantitative AI tool to monitor prices and define the best price to be offered, Gen AI will generate ready-to-use pitches for negotiation with suppliers supporting the commercial team in charge.

How does it change business?

The shifts created by the rise of Gen AI have impacted all aspects of businesses and organizations. From a tech perspective, there are no more barriers. Now that AI is a commercial product, in-house development

does not make sense anymore and is not a strategic advantage. The focus now is on the use case play, rather than ensuring that the use cases can be powered. Companies are rushing to find applications for this new capability.

From a general business perspective, all industries are impacted. AI is a mature and business-ready tech, and finding the right way to use it is a competitive advantage, especially in low-margin sectors such are retail. Vincey mentioned that only 1/3 of the use cases developed at Bain are customer-facing solutions, while the remaining 2/3 are backstage efforts on productivity.

Now that tech is just a product that can be bought off the shelf, companies have only one strategic asset: their proprietary data. The urgency is to define the best opportunity how to combine AI products with company data. In that game, companies can either decide to be the first mover or the fast follower. All options are valid.

The impacts in terms of new needed mindset, capability to test, risk and encourage innovation within organizations, are heavily felt and visible. This is why phasing is key:

• The first phase is to develop and deploy Gen AI tools in employee assistance, to save time, limit errors and increase productivity (for instance, to help employees better recommend products to their customers),
• The second phase is to see AI as a co-pilot, proactively making propositions. For instance, AI can manage real-time customer engagement.
• The third phase is to consider full automation. Carrefour’s new chatbot takes over the relationship between the e-commerce website and the customer.

A tectonic shift in retail and for department stores

There are 4 current archetypes across industries on how Gen AI is used:

• Content generation, including images and assets. Coca-Cola now uses a platform based on GPT-4 and DALL-E to create artwork based on company archives and A/B test them at scale (millions of copies are created and tested automatically, vs. 2 in the past, which were developed by a third party). Carrefour uses Gen AI to deal with product data management (product descriptions).
• Advanced analytics, such as a predictive NPS based on conversations. Morgan Stanley has developed a model to help their advisors with AI ‘listening’ to their conversations and making product suggestions in real-time.
• Personalized chatbots, such as Carrefour which has developed a conversational grocery shopping chatbot, are able to make recipe proposals according to customers’ needs, adjust them and propose a full cart of

products needed to make such meals.

• Information retrieval, where syntheses are generated from unstructured data. Salesforce is developing a tool which generates AI-powered content for CRM across the data available in all Salesforce clouds. The Carrefour case perfectly illustrates the capability offered by Gen AI to leapfrog and quickly implement new services and usages: the chatbot, addressing a website with 15m MAUs, has been developed in 6 weeks between kickoff and going live. It can propose a list of menus according to the dietary requirements of the customer, adjust them, and suggest a shopping cart with the right quantity of ingredients needed, in a given budget. Only 10% of the whole development needed specific information from Carrefour, the rest came from the learning capabilities of GPT.

Where in the value chain Gen AI can impact retail?

The answer is simple: everywhere, both in customer-facing operations and backstage.

• In the outreach part, marketing is enhanced with tailor-made propositions customized at scale, which in turn provide the company with a smart view of the customer (for instance, Carrefour has now access to the

customer’s decision-making process easily as it can monitor it in real-time thanks to the client’s interactions with the chatbot), while, in terms of internal processes, Gen AI can help in RFP creation, vendors communications and negotiations, as well as category recommendations.

• In the “decide & buy” part, Gen AI impacts personalized targeting, product information (which is dynamically managed), checkout & payment, and customer service. It also impacts operations (employee enablement,

report generation, product design…) and HR.

• In the “receive & return” part, AI can help with the delivery scheduling and tracking, and provide help with returns, which are also more easily processed internally.
• Finally, in the” use” part, AI impacts customer connectivity and loyalty, and efficiency.

For Vincey, the impact of Gen AI is virtually unlimited. Some tests have been made on the supply part of Carrefour’s business, and GPT proved able to analyze and score RFPs on a quantitative and qualitative basis and recommend modifications that proved correct to buyers.

Bain noted from their own experiences with their customers that the productivity gain is on average +30% in the back office with the use of Gen AI.

Reimagining product categories and customer experiences

For department stores interested in beauty and fashion, he mentioned that Gen AI was specially adapted to 5 main use cases:

• Marketing campaign booster,
• Website content optimization,
• Community management,
• In-app beauty coaching,
• In-store beauty advisor personal assistant for salespersons

When it comes to customer experience exemplifications, Vincey showed a few demos of actual use cases:

• The US-based insurance company USAA uses a combination of GPT and DALL-E to create dozens of personalized ad copies according to a specific customer profile and a problematic, and A/B test them at scale,
• Adidas uses GPT in social media, where Gen AI makes proposals of answers to be sent to customer comments, in order to generate interaction,
• USAA uses GPT in up-sale and cross-sale with the analysis of credit card history, it can make mass-personalized emails (that can be also tweaked with external information) to be sent to customers.

*How can retailers move forward?

As of now, there are three ways for companies to address the challenges and opportunities created by Gen AI:

• Some decide to go into a full 360° transformation journey, based on a leap of faith, and the purpose is to go fast in the transformation in order to reap market share and reduce the cost of operations. This is the case with Coca-Cola or Carrefour.
• Some see Gen AI as a use case acceleration and adopt a test & learn approach with 1 to 2 use cases, which are developed and implemented quickly, to prove out the opportunity and identify the requirements for scale. This bottom-up strategy allows them to act and then strategize.
• Finally, some companies are voluntarily slower in their approach and decide to identify the value at stake by looking at their most critical use cases opportunities compared to their business strategy, and then implement those projects with a top-down approach.*

Whatever the option is chosen, no retailer has the choice to ignore what is going on now with AI, as the risk would be to be left on the side of the road. The change is equivalent to the 1990s when PCs and Internet invaded office spaces, with the difference that this change took place in 10 years, whereas now the timeframe for adaptation is much more reduced. For instance, at Bain, the technology component of the organization involves 1,500 technologists, including 400 data scientists, of which 150 specialized in text mining and analytics. In the course of only one year, these 150 people got rid of their in-house toolbox of natural language processing techniques, and now exclusively use GPT. The disruption in the consulting field is huge, and retail should expect a similar shift.

Vincey also reminded the audience that Gen AI is a great way to catch up with big tech as the leverage effect can be huge, not only in terms of business approach or organizational processes but also in terms of company mindset. It is all about granting the right to the teams to innovate, test and learn, sometimes at the risk of making mistakes.

For that reason, change can only happen if CEOs show the way and encourage such behaviour, if these tests do not take place in the mission-critical part of the business.

Credits: IADS (Selvane Mohandas du Ménil)