Experts from Microsoft, DeepBrain AI, Intercom and more shared their hands-on experiences in building generative AI-powered solutions.
Generative AI is transformational and revolutionary, but it has some flaws and limitations. Should you start implementing it in your organization?
The consensus from all the experts that gathered for our Disruption Forum Generative AI event was yes, you definitely should - but you need to have a solid plan for it and be prepared to face surprising obstacles along the way.
If you’re planning to use GenAI in your business, here is all you need to know to get started in the right direction, from seasoned experts who work with this technology daily.
Expert opinions based on hands-on experience with generative AI
💬 To get the most out of AI in your organization, Microsoft's Marc Teipel explained that you need a lot of structured data.
- The right way to implement GenAI is to find a specific problem and build out your solution from there.
- You need to be crystal clear about what you want to achieve with AI if you want the implementation to go smoothly.
- Leaders need to make sure that GenAI is used in a constructive way.
- Some of the most exciting benefits of GenAI is for companies that have fragmented data that’s hard to extract insights from. For example, a call center can use GenAI to analyze thousands of conversations daily, which humans could never do. One company was able to generate customer review summaries with GenAI in 3 weeks, and calculated that it would have taken 11 years to do it manually.
💬 DeepBrain AI's Michael Jung said that ChatGPT showed a lot of people the potential of GenAI and many C-level executives are pushing for GenAI initiatives, but they need to be careful.
- In some situations, applying GenAI can cause backlash, including job loss, legal repercussions, or leakage of internal expertise.
- One of the most exciting things will be how the customer experience changes thanks to GenAI. For example, LLM-powered chatbots for customer support will be much smarter and they’ll be everywhere.
💬 On the topic of how GenAI improves customer support, Christian Brucculeri from Quiq mentioned that customers don’t care much if they talk to a bot or human, they mostly want a question answered quickly and GenAI accelerates this ability extremely.
- It provides accuracy and consistent availability.
- It reduces response time for agents, but also it levels the playing field because it helps lower-performing agents to catch up with the top performers, improving the consistency of customer experience.
- The mass adoption of ChatGPT is already causing consumers to expect the same kind of experience from chatbots. Basically, every structured, tree-based chatbot is now outdated and obsolete.
- However, there will always be some experiences where humans will be irreplaceable, for example high consideration products, or situations where there’s a dispute and customers want to talk to a human.
💬 Intercom's Fergal Reid explains that GenAI brings customer service into a completely new dimension.
- The capability of generative models is new and transformative. It’s much better than previous technologies at navigating conversations, understanding what you’re saying, even building them is a much better experience for developers. For simple informational queries, GenAI is magical, it’s a massive shift.
- A lot of Customer Service is immediately automatable because of GenAI, but there is still work to do before GenAI can handle all CX activities. For example we don’t want it to handle return cases and issue refunds yet, but with time it should be possible.
💬 With GenAI you can use bots not just to help customers but also generate revenue, said Maja Schaefer from Zowie.
- Not only can GenAI provide answers that fit the context and the brand, but also be a personal AI shopper that helps customers pick products they like. Because of GenAI, human customer service will become kind of a luxury service.
💬 For a more technical use case, Morgante Pell from Grit explained how GenAI has been a huge unlock for complex software maintenance.
- For example, modernizing legacy applications used to be a grueling process, something that tired developers out and rarely got 100% done.
- With GenAI, you can put it to work modernizing the code, it won’t get tired and will go on until it’s all finished, while developers can focus on business logic and other high-level considerations.
- One of the things about GenAI is that its power is obvious, but to be effective in an enterprise environment, you need to put a lot of guardrails on these systems to make sure they work as intended.
- One of the most exciting areas now is simulation environments where you insert AI agents, and they learn how to behave from first principles. We could see environments in which you can train an AI agent to be a senior software developer.
- When building with GenAI, the biggest thing is to look at where you want to go, make your goal as specific as possible. You also need a strong business case, if there is no potential ROI then it won’t make sense to deploy the budget and resources to build it.
💬 Jarek Swiecicki from Morizon shared his insights from adding a GenAI experience on top of a search engine for his real estate platform.
- Thanks to GenAI-powered natural language search, users can write out their train of thought and the system converts it into a structured query for the search engine.
- Because of the ChatGPT revolution, users are already starting to expect these experiences, which is a good reason to start building them now.
- Another reason to start now is that there’s a learning curve that has to happen before you can extract value from GenAI.
- You will encounter surprising challenges, for example GenAI models might provide the correct output for a specific input most of the time, but then suddenly it just fails and it’s unclear why.
- To start testing with users, it’s good to play it safe. Jarek mentioned that they do A/B testing of the standard search against the GenAI experience, and analyze which is working better based on their main KPI - the ratio of leads to visits. Only few users get the GenAI experience for now, ensuring that if it goes wrong, it won’t bring down the business.
💬 Sebastian Graf from True Carbon mentioned that anyone who uses ChatGPT today is providing fine-tuning data to improve the underlying models.
- Often, the data isn’t anonymized enough and the model might end up using it to answer somebody else's question.
- Sebastian expects that we’ll see a lot of content online generated on an individual basis for each user, especially ads.
- At the same time as OpenAI has exploded in popularity, there’s also a legion of open-source developers putting together GenAI models you can run on your own machines.
- The GenAI renaissance is coming, and Sebastian predicts that it might happen in Europe because of the EU’s dedication to regulating AI. Also, thanks to the multicultural character of the EU, there’s a possibility to develop localized AIs that operate within existing laws and licensing infrastructure.
- Right now we’re at a point where GenAI can create conversations with libraries and with individual documents. For example, you can tell it to give you specific parts from a book, and it will provide them for you along with source information.
- Sebastian calls the current stage of AI the “immunization phase”. People have learned to tune out ads and sponsored content that tries to influence them in some way, but now they have to learn to identify and respond to GenAI content. People need to use these tools and discuss them to become immune to being negatively influenced by AI.
💬 ChatGPT democratized GenAI, and made everyone think “how can I use that in my business?” said Microsoft's ShiSh Shridhar.
- GenAI in retail is transforming the way people shop. You no longer have to plan your shopping out in detail, you can just ask the AI to do it for you - for example, “I have such a diet, these allergies, please prepare a meal plan and shopping list for the week”, or “I’m going to a wedding in Hawaii, pick out some appropriate clothes”.
- Another benefit is that AI automates a lot of grunt work in different operations. For example, in recruiting it can take over the process of analyzing CVs and making sure they fit the job description, providing a human recruiter with a shortlist of candidates.
- Also, personalized communication at scale is a game-changer. You’ll no longer send generic ads, you’ll have the system generate individual ads that fit a customer’s profile perfectly.
- There’s also potential for mixing GenAI with RPA (Robotics Process Automation).
- When it comes to complex things, like making million-dollar decisions about the direction of your business, you can’t fully rely on GenAI yet. But you can use it to analyze and summarize opinions from your team and stakeholders, to identify the biggest concerns that need to be addressed and create action items to drive your decision forward.
- Exciting new experiences are coming all across retail, like models that create new fashion designs, or models that can analyze your skin and suggest the right cosmetics.
- Today, if you’re a Customer Support agent, you’re constantly working within the CRM, inputting data, checking for information, planning follow-ups and so on. With the right GenAI system, agents won’t have to do any of that anymore, the model will do it for them and they’ll focus on communicating with the client.
- What AI still can’t do is make decisions that require a balancing act between data and human factors. AI is great at data, but it can’t understand nuanced human behaviors and motivations, so you can’t fully trust GenAI to make your decisions for you.
How will your organization use generative AI?
The mass adoption of tools like ChatGPT has raised customer expectations regarding digital experiences. Generative AI is revolutionizing business operations, customer service, and even software maintenance. The rewards of GenAI are undeniable, but it needs to be implemented with clear objectives and an understanding of the risks.
Ultimately, the key to effective GenAI use is finding the perfect balance of technology and human expertise, combining the best of both worlds for optimal results. This will take time, testing, and ongoing learning, but the potential rewards make it well worth the effort.