Hey there 👋

📢 Ch, ch, ch, ch, changes… So there are a couple of improvements to the format of this week’s case study, the weekly report where I put myself in the shoes of a product team and identify how generative AI could add new value to their customers.

(True, this is only the 2nd week of writing these, but I made sure to reach out and get some feedback from readers (thanks all of you!) and that has “prompted” some changes.)

First off, the name. Although I kinda liked the alliteration, “AI Audits” implied I was reviewing existing AI features, rather than suggesting new ones, so we’re making the switch to 🛣️ AI Roadmaps.

Second, my first case study, on Monzo, generated super high engagement rates, around double that for the daily email. So my hypothesis is that it’s worth going all in on these, and you’ll see more product thinking rigour in the report, for example identify customer segments and applying Jobs To Be Done methodology as a core determinant of the opportunities we explore.

Finally, as of next week I’ll be introducing Product Prompts Plus - a paid membership option (as well as the usual free one). This will initially feature extended weekly AI Roadmaps, in which I’ll go super deep with tons more insights and inspiration for your own products, plus free access to future Prompt Books, webinars, and more!

Ok, on with the show!

This week’s AI Roadmap is on LinkedIn.

And given that the company is owned by Microsoft, who’s partnership with Open AI is already translating into new features for core products such as Teams and Bing, it may not be too long before we start seeing some of the features mentioned in this report coming to the world’s largest professional network.

You’ll read about:
- 6 user segments
- 20 Jobs To Be Done for one of those segments
- 20 generative AI product ideas
- 2 feature deep-dives and risk analysis

(Descriptions here of LinkedIn’s product strategy and potential AI features are purely my personal views.)

💪 Vision And Mission

Vision: to create economic opportunity for every member of the global workforce.

Mission: to connect the world’s professionals to make them more productive and successful.

🍰 User Segments

LinkedIn can broadly be divided into the following user segments:

Job seekers: individuals who are actively looking for employment opportunities and use LinkedIn to find relevant job postings and connect with potential employers.

Recruiters: find suitable candidates for job openings in their organisations.

Business professionals: use LinkedIn to network, build their personal brand, and stay up-to-date with industry news and trends.

Employers: post job openings, showcase their company culture and values, and engage with potential employees.

Freelancers and entrepreneurs: find freelance projects or to market their products and services to a wider audience.

Educational institutions: universities and other educational organisations that use LinkedIn to connect with alumni, showcase their programs, and engage with potential students.

Of course in practice there is often overlap, and you could even no doubt dissect each of these segments further, but at least by segmenting like this and understanding the different needs, we can identify opportunities to add generative AI to improve the user experience and help users achieve their goals more efficiently and effectively.

🥅 Jobs To Be Done

Let’s take one of these segments, business professionals, and explore their likely functional, social and emotional JTBD when using LinkedIn.

• Build and maintain a professional network of contacts and colleagues

• Find new business opportunities and potential clients/customers

• Research potential business partners or vendors

• Stay up-to-date with industry news and trends

• Showcase their professional accomplishments and expertise

• Develop thought leadership and establish themselves as industry experts

• Recruit new employees or contractors for their company or projects

• Find potential investors or partners for their business

• Receive recommendations from colleagues or clients to enhance their professional reputation

• Collaborate with colleagues or clients on projects

• Gain exposure and visibility for their brand or company

• Find industry-specific events and conferences to attend

• Research competitors and industry trends to inform business strategy

• Seek mentorship or guidance from experienced professionals

• Share knowledge and best practices with their industry peers

• Explore job opportunities and career development paths within their industry

• Find and engage with relevant industry communities or groups

• Connect with educational alumni and former colleagues to network and potentially work together

• Discover and evaluate potential business acquisitions or partnerships

• Monitor and manage their company's online reputation on the platform.

That’s a huge variety of Jobs! I’ve listed twenty, and one could imagine there being hundreds across all the different jobs and segments.

Without conducting a research survey we can’t know for sure which jobs are most pressing, i.e. with which people have the most pain or difficulty without LinkedIn, and which are most important. (This is extremely useful to know, as it pinpoints areas of opportunity for product teams to consider working on). But I suspect that there are probably 3 or 4 Jobs which make up 80% of the motivation for any given user session. Let’s assume these are:

1. Develop thought leadership and establish themselves as industry experts

2. Build and maintain a professional network of contacts and colleagues

3. Find new business opportunities and potential clients/customers

4. Gain exposure and visibility for their brand or company

Let’s concentrate our focus around these. Remember, we’re interested in generative AI, so Large Language Models like GPT-3, the new ChatGPT API, text-to-image generation and audio transcription.

💡 Generative AI-deation

JTBD: Develop thought leadership and establish yourself as an industry expert

Personalised email newsletters: generate personalised email newsletters for users based on their expertise and interests, providing subscribers with valuable insights and establishing the user as a thought leader in their industry.

Expert opinion pieces: generate expert opinion pieces on current industry trends or news topics, allowing users to establish their thought leadership and share their insights with a wider audience.

Voice-enabled thought leadership: create a voice-enabled feature that allows users to ask industry-specific questions and receive expert-level answers in real-time.

Social media captions: generate compelling post captions that incorporate trending industry topics and hashtags, helping users establish their thought leadership.

Moonshot idea: generate entire keynote speeches or conference talks based on a user's industry expertise and personal style, which can be broadcast live.

JTBD: Build and maintain a professional network of contacts and colleagues

Personalised follow-up messages: generate personalised follow-up messages for users to send after connecting with a new contact, based on an analysis of both profiles, increasing the likelihood of building and maintaining professional relationships.

Intelligent networking: power intelligent networking features that analyse users' profiles, job titles, and industries to suggest relevant and valuable connections for users to make.

People search: use Natural Language Processing to enable users to enter quite detailed descriptions of the kind of person they’re looking to connect with.

Professional writing assistant: power writing assistance features that suggest relevant, professional language for users to use when composing posts, messaging potential connections or creating their profile.

Moonshot idea: generate virtual chatbot-based networking events or "speed dating"-style networking sessions, allowing users to build connections at scale.

JTBD: Find new business opportunities and potential clients/customers

Personalised pitches: generate personalised pitches for users to send to potential clients or customers, incorporating industry-specific language and trends to increase the likelihood of conversion.

Intelligent lead generation: features that analyze users' profiles, job titles, and industries to suggest relevant and valuable potential clients or customers for users to pursue.

Customer personas: generate detailed customer personas for users to use when creating marketing strategies, helping to identify and target potential clients or customers more effectively.

Sales forecasting: intelligent sales forecasting that provides users with predictions on future sales opportunities based on historical data, allowing users to identify and pursue potential clients or customers more effectively.

Moonshot idea: generate entire LinkedIn marketing campaigns, from identifying target audiences to crafting messaging and producing creative assets.

JTBD: Gain exposure and visibility for their brand or company

Social media scheduling: identify the best times and platforms to post content, increasing the reach and engagement of users' social media content.

AI-generated posts: generate high-quality posts, incorporating trending hashtags and topics to increase visibility and engagement.

Influencer outreach: identify potential influencers or brand ambassadors who could promote users' content or products, and provide personalised outreach messages to build relationships.

Product descriptions: generate product descriptions or web copy that are optimised for search engines and increase visibility in organic search results.

Moonshot idea: AI-generated video content: Use GPT-3 or similar technology to generate high-quality video content, such as explainer videos, product demos, or promotional videos, based on a user's industry expertise and brand voice.

🤿 Idea deep-dive

Let’s take two of these ideas and consider how these could be applied, plus understand the risks which could then be tested.

Feature idea #1: AI-generated video content
JTBD: Gain exposure and visibility for their brand or company

Generate high-quality video content, such as explainer videos, product demos, or promotional videos, based on a user's industry expertise and brand voice.

The AI model could analyse existing content from the user's profile and other sources to identify key themes and messaging, and generate a script and storyboard for the video. The user could then provide input and feedback to refine the content, and the final product could be produced using video editing tools or AI-powered animation software.

This would enable business professionals to create engaging video content more quickly and efficiently, and to leverage their industry expertise to produce high-quality video content that showcases their brand or thought leadership position.

Risks

The video quality isn’t high enough, or generates too many “hallucinations” where the model’s image is too random. The likes of Runway are pioneering video creation and editing with AI, so we know the tech is making incredibly rapid advances, but even if a partnership with Runway did happen, LinkedIn may deem it too risky right now, preferring to wait until 2024 to avoid risking brand reputation and loss of trust.

Intellectual property: There is a risk that the AI-generated video content may infringe on existing copyrights, trademarks, or other intellectual property rights. This could result in legal action, damage to the brand's reputation, and potential financial penalties.

Lack of creativity: AI-generated content may lack the pure thought leadership creativity and originality of human-generated content, potentially making it less engaging and effective. This could lead to decreased engagement and user satisfaction, as well as reduced effectiveness in achieving the intended Jobs To Be Done.

Cost: Developing the technology and infrastructure required to support AI-generated video content can be expensive. There is a risk that the cost of developing and maintaining this technology may outweigh the potential benefits, leading to a negative return on investment.

Ethical concerns: There are ethical concerns surrounding the use of AI-generated content, particularly with regards to the potential for deepfakes or other forms of misinformation. There is a risk that the use of AI-generated video content could contribute to the spread of false information, damaging the brand's reputation and leading to legal and regulatory challenges.

Feature idea #2: People Search
JTBD: Build and maintain a professional network of contacts and colleagues

People search: use Natural Language Processing to enable users to enter quite detailed descriptions of the kind of person they’re looking to connect with.

To actually find people to connect with whom you don’t necessarily know, you currently either have to use the search function, which is pretty basic with its filter options, or hope that the “More suggestions for you” recommendation engine throws up some interesting ideas. Personally I can’t imagine the conversion rates of the latter are any good; I rarely see a stranger with whom I feel enough common ground to warrant trying to connect.

So what if there was a way to ask a bot to scour LinkedIn and identify people who might be a good fit for whatever you’re looking for, and even craft a personalised message that would optimise your chances of getting a response?

Your prompt could be something like:

Then you could refine the results with further prompts. How much easier (and surely more accurate) would that be?!

Risks

Privacy concerns: The use of detailed descriptions to search for people on LinkedIn may raise privacy concerns among users. There is a risk that users may feel uncomfortable with the level of detail being requested, potentially leading to reduced engagement or usage of the feature. I see this as low risk and easy to test.

Technical limitations: While Natural Language Processing has shown impressive capabilities in understanding and analyzing text, there are still technical limitations in its ability to accurately interpret and understand complex user queries. There is a risk that the feature may not accurately understand user queries, potentially leading to inaccurate or irrelevant search results. That said, I still think you’d get more relevant results than you do currently, and careful training of the model with the right “embeddings” should optimise the results.

📚 Related reads

LinkedIn co-founder Reid Hoffman leaving ChatGPT maker OpenAI's board (presumably to avoid conflicts of interest) - link

✨ Enjoying the daily emails and weekly AI Roadmap? Spread the word and forward this to others who want to build better products, faster, with generative AI. And if you have any suggestions for future reports, hit reply and let me know!

Thanks, and happy prompting!

Martin