You Need To Know What These Top Companies Have Been Building With Generative AI

As the Generative AI revolution continues, more and more successful companies are harnessing the power of OpenAI’s ChatGPT, Google’s Bard and other systems to support their employees and create better products and services for their customers.

‍

At the same time, many others are still unsure about what it could mean for them and their own business models or operations. So if you are a company leader who’s highly intrigued by the corporate potential of generative AI but still looking for inspiration about how it could concretely help your company, then this post is for you. I gathered some of the most interesting enterprise cases that I encountered over the past weeks and months to help you see the possibilities.

‍

I’ll first start with some basics about generative AI, so if you feel you’ve read enough about the subject - because that is ALL that the entire internet seems to be talking about these days (Hear me out, generative AI is the cat pictures of our time) - then feel free to skip this part. No no, I won’t mind that you’ll be ignoring all the hard work that I put into this. NOT. AT. ALL.

‍

What the hell is generative AI?

It is a specific type of artificial intelligence that is capable of “making new stuff”. In response to your “prompts” - which is just a fancy word for “questions” or “orders - it can generate all kinds of useful content:

• Text
• Images
• Video
• Sound
• Music
• Speech
• Software code
• Product designs

Basically you can ask the AI system a question like “What if the internet was cats?” and then it offers its interpretation of that, which it bases on patterns it identifies in a humongous amount of existing data. I did ask that question and this was Midjourney’s underwhelming answer (I’m a big fan of Midjourney but often you have to try several times to get to a great version):

‍

Though, after some re-iterations, it did end up with something pretty cool:

‍

‍

This was part of OpenAI’s ChatGPT’s answer:

‍

‍

Which was actually pretty similar to Google’s Bard’s answer:

‍

‍

Before you ask: yes, these systems can also answer really useful questions, but why would I do that if I can ask it about cats? Let’s just stay serious here for a minute, right?

‍

Where can you make what?

First of all, it’s always good to know the difference between:

1. The company that built the model
2. The name of the chatbot interface it uses
3. The underlying model: for instance a latent diffusion model or LDM (like used by image-generator Stable Diffusion) or a large language model or LLM (as used by OpenAI’s ChatGPT)

‍

If only because it will multiply your coolness at parties at least by a factor of 1,5 (not more, because, well, it’s also pretty nerdy), if you know when to ask “Which LLM does that tool use?”.

‍

Anyway, for this blog, I’m mostly zooming in on the Large Language Models or LLMs, which lie at the basis of most text-generating chatbots, as this seems to be the main area of attention for enterprises. LLMs are AI models that can generate new human-quality text because they have been trained on a massive dataset consisting of books, articles, press releases, conversation, recipes, company websites etc. in which they find recurring patterns and then predict what the best text to follow your question would be.

‍

Some people mistake LLMs and their bot-interfaces for search engines like Google, but they are in fact built to write (not look-up) text and predict the best words following a prompt. That does not make them the best place to search for existing information because they sometimes “hallucinate”, which is a just fancy word for making sh*it up. They do that because they are so eager to predict the text that would follow a question – pardon, “prompt” – that they invent things if they cannot find the right information.

‍

These are some of the top text generating tools of the moment. I added some Chinese ones for diversity (and because China has been deeply investing in that area) but these solutions are not widely used over here, especially not in enterprise environments:

‍

‍

Some of these are connected to the internet - Bard, HuggingChat and ChatGPT (if you install the WebChatGPT extension) - and some, like Claude and LLaMA aren’t. That’s relevant, because those not connected to the internet will not be able to search recent data, only up to the point where the available data goes.

‍

Other notable generative AI models include artificial intelligence art systems like Stable Diffusion, Midjourney, and DALL-E but, again, they are not the main area of focus of this piece.

‍

There are a lot more generative AI chatbots out there, of course. But most of them use the proprietary LLMs described above. Snapchat’s ‘My AI’ chatbot and Jasper, for instance, are powered by OpenAI’s GPT model as described above. Why? Because it costs a massive amount of energy and a gazillion (true story) of euros to build your own LLM so why would you, if you can use an open source version or pay for the API of one of the closed ones. Amirite?

‍

Yeay, enterprise!

And now on to the good stuff. There are many ways in which a company can use the API of GPT or the Open Source model LLAMA internally to build useful tools for its employees or customers. To be clear: I am not talking about your employees using ChatGPT or Bard, as they are using Google, but about your company using LLMs to mine its own internal data, to support its marketing department and create new features or even services for your customers.

‍

There are quite a few companies, particularly those with sensitive data like Samsung, Bank of America and Wells Fargo, that are building their own “private” and internal models, mostly because they do not want their employees to use the open versions of ChatGPT or Bard because of security and privacy reasons. This fear of employees using the open version of ChatGPT is actually quite common:

‍

‍

But let’s move on to the exciting part: here's how some top companies are using generative AI to their advantage.

‍

Knowledge Management

Knowledge management is the holy grail of large organizations. They have so much valuable information and insight available but it is often not accessible to the right people because it is so dispersed over different departments, in so many formats, and they lack the tools and sometimes even the culture to share what’s useful. In many ways, enterprise generative AI platforms can help transform this wealth of information into usable knowledge and make it accessible to all (though it should be noted that it’s not a magical solution and it will certainly not help you change your company culture, should that be the case).

‍

Here are some of the top examples:

• American medical center Mayo Clinic uses Google Cloud's “Enterprise Search on Generative AI App Builder” to build a chatbot that enables medical professionals to quickly access patient data.
• American multinational investment bank and financial services company Morgan Stanley is testing an OpenAI-powered chatbot for its 16,000 financial advisors which generates responses from a selection of its own 100,000 or so pieces of research - rather than the entire internet - to cut down on errors. To further reduce mishaps, the bank also has humans checking the accuracy of responses. 
• American multinational financial services JPMorgan Chase built its own ChatGPT-based LLM to analyze Federal Reserve statements and speeches in order to expose potential trading signals.
• American multinational investment bank and financial services company Goldman Sachs is looking at LLMs to take document classification and categorization - something it has been using traditional AI to do until now - to the next level.
• The largest supermarket chain in the Netherlands Albert Heijn is rolling out its very own AH GPT platform. For the moment, I believe it’s “just” about its employees using generative AI in a secure internal environment in order to protect the shared data, but don’t forget that they will learn a lot from this type of “small” experiments.

‍

Building New Services for Customers

Beyond internal use, generative AI is also enabling companies to offer new services to their customers:

• ‍JPMorgan's IndexGPT is aimed at simplifying financial investment by analyzing and selecting securities tailored to customer needs.‍
• The Josh Bersin Company is developing its “HR Copilot” which is meant to offer access and analyze more than 50,000 pages of research, 1,000+ blogs, extensive podcasts, as well as the company's vast research library of case studies, vendor information, labor market insights, and economic data.‍
• McKinsey partnered with AI startup Cohere to provide AI solutions to its enterprise clients‍
• Amazon Web Services is investigating how to provide AI models that their clients can use with their cloud services.‍
• Intel is working on Aurora genAI which is a trillion-parameter AI model - almost six times more than in the free and public versions of ChatGPT = that will cater to the science community and accelerate the advancement in System Biology, Cancer Research, Climate Science, Cosmology, Polymer Chemistry, and Materials Science.‍
• Salesforce is creating Einstein GPT, a ChatGPT-like tool for CRM, which can generate personalized emails for salespeople to send to customers or produce responses for customer service professionals to use when handling customer queries. Salesforce customers can connect their own external models to Einstein GPT and use natural-language prompts from their CRM to generate content that is updated with real-time customer data.

‍

Creating New Features for Products & Online Services

Generative AI also enables the creation of new features within existing products and services:

• The Albert Heijn app has a new generative AI “Recipe Scanner” feature with which you can scan a recipe, have the ingredients translated into Albert Heijn products which are then automatically added to the shopping list. The scanner feature is pretty smart, distinguishing between various ingredients like tomatoes and tomato cubes, and effortlessly adjusting quantities based on the number of people eating.‍
• Mercedes-Benz is integrating OpenAI's language model, ChatGPT into its MBUX infotainment system to enhance user experience and allow drivers and passengers to have more natural and complex interactions with the system.‍
• Danish-American software-as-a-service company Zendesk’s “Zendesk AI” recently added capabilities including advanced chatbots, AI-generated content, tools designed to detect intent and analyze sentiment in conversations, response rephrasing and tone shifting.‍
• Oracle is using AI in its HR software to help its customers draft job descriptions and performance reviews.‍
• Snap recently released its “My AI” chatbot to all of Snapchat’s 750 million monthly users for free (after it was first made available to the app’s more than 3 million paid subscribers). It can for instance recommend AR filters to use in Snapchat’s camera or places to visit from the app’s map tab and can be added to group chats by mentioning it with an @ symbol.
• ChatGPT is also being integrated into Slack, Salesforce’s instant messaging platform.
• Language education specialist Duolingo’s Duolingo Max - powered by GPT-4 - enables language students to get in-depth explanations of why their answers were (in)correct, delivered in natural language, just like they would from a human tutor.
• Online course provider Udacity, too, has used GPT-4 to create an intelligent virtual tutor that can provide personalized guidance and feedback to students. It offers detailed explanations that can be customized to the individual learner and can summarize concepts and explain technical jargon as well as translate when a course might not be taught in a learner’s native language.‍
• TikTok is testing an AI chatbot called Tako that can recommend videos based on what people ask it.

‍

Marketing, CX and Customer Service

The marketing and CX landscape, too, will be undergoing a seismic shift due to AI:

• ‍Coca Cola has partnered with Bain & Company to leverage ChatGPT for marketing and creating personalized ad copy, images and messaging.‍
• WPP, the giant ad agency network, has a deal with Nvidia for generative AI to automate asset creation in campaigns.
• UK-based energy supplier Octopus Energy has built ChatGPT into its customer service channels and says that it is now responsible for handling 44 percent of customer inquiries. The app is said to do the work of 250 people and receives higher customer satisfaction ratings than human customer service agents.

‍

Editing and Summarizing

Generative AI is obviously also very helpful for editing and summarizing vast volumes of information:

• ‍Goldman Sachs is experimenting with summarizing earnings calls and research.‍
• Hitachi's new "Generative AI Center" is using AI for value creation and productivity enhancement.
• The city of Yokosuka in Japan officially adopted artificial intelligence chatbot ChatGPT to make bulletins, summarize records of meetings and edit documents.‍
• Amazon is testing the use of AI to generate summaries of reviews left on some products. The feature provides users with an overview of what shoppers like and dislike about a product they purchased.

‍

Developer Support

Another really interesting enterprise case for internal generative AI could be its use as a support for developers. The only ‘official’ example I came across so far, was Goldman Sachs, which has been assisting its developers with AI-powered in-house chatbots to write code. But I’m pretty sure that more companies are or will be looking in that direction. Hook me up if you know any other examples.

‍

The beginning

To conclude, it's true that the enterprises that are currently (or have been open about) experimenting with generative AI to enhance the experience of employees or customers are mostly the bigger technology companies like Amazon and Oracle or multinational financial services providers like Goldman Sachs or JP Morgan. But this is also really just the beginning and we are seeing some early adopter retailers like Albert Heijn or FMCG companies like Coca Cola jumping on the train, too.

‍

I’d probably advise organizations to follow in their footsteps and organize small-scale experiments - where the impact of their failure remains manageable - to learn what the possibilities are, as well as the limitations.

‍

Above all, if your company is experimenting with generative AI – both internally of externally – or if you know which other organizations are, let me know! I’m really interested in learning about what’s going on in this area.

‍

‍