Supercharge Lead Generation with AI Powered Persona Development

**(00:43)** This is an interactive group, so please feel free to share comments and ask questions. Francesca, would you like to start by asking if anyone here has experience with conversational AI? For instance, has anyone worked with tools like Dialogflow?

**(01:28)** Karen: Yes, I’ve done an integration with our knowledge base using an older version of Dialogflow. It involved a lot of top-down guidance, where we mapped out each interaction. It was effective at recognizing questions, but it was tedious since we had to map out every possible conversation flow. I think combining this kind of setup with ChatGPT’s more dynamic capabilities could be very powerful.

**(02:16)** Francesca: Absolutely. Imagine if a brand like Nike created a conversational AI that could recommend products and recognize customers’ needs without overstepping, like discussing unrelated topics or promoting competitor brands. Moving from rules-based chatbots to more human-like conversational AI using generative technologies allows for a much more fluid experience. However, it’s crucial to keep the interactions focused on brand-related topics.

**(03:26)** Has anyone else used conversational AI, maybe for marketing, customer service, HR onboarding, or user research?

**(04:04)** Attendee: I’ve participated in studies using open-ended conversational AI in surveys. The tech still has some flaws—it often doesn’t seem to recognize pauses in conversation, and sometimes it repeats questions I’ve already answered. The listening and contextual abilities definitely have room for improvement.

**(04:34)** Francesca: Exactly. AI needs to develop a better understanding of tone, cadence, and phrasing to mirror human behavior effectively. This can only happen if it has access to prior conversations to personalize each interaction. The goal is for AI to make each conversation unique and tailored.

**(05:18)** I’ll dive into my presentation now. Today, I’ll be discussing unified conversational AI and its potential to transform customer experience. Let’s imagine a future scenario: Suppose Taylor Swift partners with a brand like Good American to launch a product line. She could have a conversational AI avatar, created using photogrammetry to render her in 3D. This avatar could converse with customers like Sophie, a 13-year-old superfan, who would have the chance to interact with “Taylor Swift” virtually and learn about the brand’s new product line.

**(06:42)** Sophie would get a personalized experience—she might even get an in-store experience where she can take a photo with the digital avatar. This combination of conversational AI and avatars could personalize marketing at a level that really engages customers, making it interactive and memorable.

**(08:27)** Conversational AI, combined with avatars, is an area that has huge potential, especially considering the hype around the metaverse during COVID-19. There’s so much more that can be done with these technologies for creating brand experiences.

**(09:06)** A bit about my background: I’ve been in tech for about 15 years, mostly in startups. I went to the same school as Richard Branson, which sparked my entrepreneurial journey. One of the first products I built was a social network called Uni, a UK version of Facebook. Later, I shifted to a mobile messaging app that evolved into a dating app called Fling.

**(10:48)** Last year, I held the first generative AI conference in London in partnership with Informa and the AI Summit. It covered the impact of generative AI across industries like music, fashion, and marketing. Day two focused on ethics, addressing topics like intellectual property, celebrity likeness, deep fakes, and bias.

**(11:26)** We even created a generative AI music video, which used AI for music composition, lyrics, and melody, with motion capture for the visuals. It was a fun project despite being experimental.

**(12:04)** Attendee: Where did you do the motion capture for the music video? Did you have access to a studio?

**(12:10)** Francesca: Yes, it was a university in southern England with a high-end virtual production setup. I also have experience with motion capture through Move.ai, so I had the right connections.

**(13:10)** Now, back to conversational AI. It’s already being used in customer service, where it can handle common questions, leaving complex queries for human agents. This 24/7 support in multiple languages is crucial as companies scale. AI can also qualify leads, assist with HR by screening candidates, and conduct product research to get customer feedback and help with market fit.

**(15:04)** For instance, in HR, instead of just screening CVs, conversational AI could handle preliminary interactions with candidates, filtering the best ones for interviews. For marketing, AI can nurture leads and personalize recommendations and content. However, if each department has its own conversational AI, it can create a fragmented customer experience. Ideally, a unified conversational AI should be developed that integrates all departments for consistency and data sharing.

**(17:35)** Let’s talk about building a conversational AI system. You’d start by defining the use case—hopefully, one that unifies functions across an organization. Then, you’d select a tech stack based on your industry, regulatory needs, and infrastructure, choosing platforms like Google Dialogflow, Microsoft Bot Framework, or Amazon Lex.

**(18:44)** You’ll design conversational flows, which can be streamlined by analyzing existing conversations, sales calls, and social media data. Generative AI, like ChatGPT, can help map out these flows, including multiple languages.

**(20:05)** When choosing a platform, compatibility with your infrastructure and data compliance are crucial. Rasa, for example, is open source, which is great if you want custom integrations. Vendor options, like Google and Microsoft, also provide strong support if you’re already using their ecosystems.

**(21:23)** Training the AI involves gathering and cleaning data, breaking down sentences into tokens, tagging parts of speech, and recognizing named entities like dates or product names. Sentiment analysis adds another layer by detecting the emotional tone of interactions, helping the AI understand and respond appropriately.

**(23:15)** Once trained, the AI will need to handle real-time integration with data sources to prevent hallucinations and ensure accurate responses, like inventory or pricing details. Platforms like Skyscanner require real-time updates, so data integrity is critical.

**(24:36)** Question: In the Taylor Swift example, would you be tagging all relevant marketing material for her brand? How would that be tokenized and tagged?

**(25:16)** Francesca: That example is a bit futuristic, but yes, you’d need to tag details about her brand and partnerships. For now, I recommend building unified AI for customer support, sales, and other areas, using high-quality, resolved interactions for training.

**(26:28)** Question: Do you use problematic customer service calls as guardrails?

**(27:02)** Francesca: Absolutely. You need to train AI to handle both ideal and challenging scenarios, including potential hacks. For instance, a supermarket AI bot was misused to create harmful recipes, illustrating the importance of managing risks.

**(28:15)** Legal compliance is essential, especially as the EU’s AI Act requires accuracy and bias testing for high-stakes applications like finance and healthcare.

**(29:35)** Attendee: Do you think celebrities are more open to chatbot versions of themselves for marketing?

**(30:14)** Francesca: They can be, but it depends on their brand alignment and long-term commitment. B- and C-list influencers may be more open, as AI avatars could help extend their reach. However, high-profile celebrities need to think carefully, as avatars might become ongoing companions for fans, which is a huge responsibility.

**(31:29)** Question: How would you manage this with time-limited campaigns?

**(31:35)** Francesca: Limiting campaigns to set time frames could make it easier to manage. Restricting chatbots to specific questions and topics also helps maintain brand alignment.

**(34:36)** Data privacy and compliance are crucial, especially as you integrate conversational AI with multiple systems. Federated learning is a promising approach, with training done on devices, potentially even on future IoT devices like smart fridges or cars.

**(37:47)** By building two-way conversational AI, brands can gather insights about customers for personalized engagement. But companies need to respect privacy, secure consent, and be transparent.

**(39:05)** After extracting insights from conversation data—such as sentiment, intent, and segmentation—you can minimize data storage and focus on insights rather than raw data.

**(40:21)** A unified customer data platform (CDP) is ideal if you have multiple AIs for different departments. Adobe and Salesforce offer strong integration for enterprise users, while Segment and Treasure Data are more flexible and cost-effective for mid-sized businesses.

**(42:37)** [Video: Demonstrating H&M’s interactive mirror using conversational AI for personalized shopping experiences.]

**(43:33)** Host: Thanks, Francesca. This has been incredible. Given that some here are more tech-oriented and others are from marketing, how tech-literate do you think marketers should be in this AI age?

**(44:14)** Francesca: Everyone should become tech-literate to some extent. Marketers don’t need to build the tech, but they should know what good interactions look like and participate in testing to ensure AI aligns with their brand’s

 tone.

**(45:00)** Question: What are the best practices for collaboration between marketing and tech teams?

**(45:10)** Francesca: Start with customer support and sales, then build out from there. Marketers can help by defining customer segments, buyer personas, and the type of personalized experience they want to deliver. It’s also beneficial to have agencies that combine marketing and tech expertise to facilitate communication.

**(49:29)** Question: How do you measure the success of conversational AI?

**(49:35)** Francesca: Good question. Metrics like throughput, customer retention, and lifetime value can help gauge success. For me, conversational AI’s strength lies in customer retention—deepening relationships and personalizing interactions to create memorable, human-like experiences.

**(51:32)** Quantifying ROI is complex, but tools exist to estimate potential savings and returns. You’ll need to consider whether the investment will pay off immediately or over the long term.