How LinkedIn Could Create a World Connected Through Mentorship
Designing a LinkedIn product that helps professionals grow faster through expert mentorship, made possible with AI
The right mentors can change your life.
But finding the right mentors, and actually getting to speak with them, is a massive challenge.
That’s why Tehseen Dahya and I spent the last 2 weeks prototyping LinkedIn Mentor — an AI tool that recommends experienced LinkedIn users to reach out to based on your most burning questions.
This article breaks down the massive problem for professionals LinkedIn Mentor solves, explains our product design process for the prototype, displays our market research, features the revenue opportunity for LinkedIn through boosting LinkedIn Premium subscribers, and highlights how this is technically feasible.
To kickoff, below is a breakdown of the problem we are solving, the solution we have designed, and the impact this could have on the world.
Problem: Professionals face roadblocks that slow them down from hitting their goals. Mentors that have encountered the same roadblocks can help, but finding mentors today is hard because of limited search abilities, a lack of clarity on what to look for, and effectively connecting with a desired mentor. According to MentorLoop, 76% of people think mentors are important, but only 37% of people have one, clearly demonstrating the gap that LinkedIn Mentor solves.
Solution: Ask a question, get an expert. LinkedIn Mentor helps professionals discover and connect with relevant mentors on LinkedIn by leveraging AI and the platform’s user base. By typing in a burning question, like “How do I raise money as a 16-year-old?” LinkedIn Mentor recommends the best people on LinkedIn to seek mentorship from, generates better questions from your original question based on the recommended expert’s experience, and lets you send the expert an InMail message to start learning and growing.
Impact: LinkedIn Mentor could launch a world connected through mentorship. The platform has undeniable potential for accelerating economic opportunity and the ability for people to make a positive change in society. From students looking to understand different career paths to first-time founders looking to learn from experienced CEOs, LinkedIn mentor empowers people to transform society.
Discovering and connecting with mentors is broken. LinkedIn can fix it.
LinkedIn is the biggest social platform for professionals.
The platform’s users join to advance their careers — whether by getting hired, growing their businesses, or pursuing other professional opportunities.
LinkedIn is also where many discover mentors. For me, it’s how I’ve met incredible mentors like Jay Singh (who used to work at LinkedIn) and Joshua Weiss from Stanford University. Mentors help you identify and achieve career goals.
What if LinkedIn made the process of finding the right mentors easier?
Finding mentors today is hard because of limited search abilities and a lack of clarity for what to look for
Finding the right mentor can skyrocket your development and learning. While people run into roadblocks everyday, the reality is that many personal roadblocks have been encountered before by others who can help you move faster. Having access to a consistent line of communication to someone who has walked before you is like getting to see the future.
Home Page of LinkedIn Mentor: Clear, Simple, and Focused
Question examples: At the bottom are question examples, letting users familiarize themselves with something they have never seen before. These question examples could be especially magical if they were based on data LinkedIn already collects on users
Go back to LinkedIn: Clicking the LinkedIn logo in the top left will send you back to LinkedIn, letting users go back and forth between their LinkedIn feed and LinkedIn Mentor at ease.
Get a Recommended LinkedIn User That Can Answer Your Question
Krishiv is a LinkedIn Mentor user. He searches “How do I raise money for my startup if I’m 16?” which leads to the following page:
Personalized mentor descriptions for a faster UX
This is not Eric’s LinkedIn bio. It is what the LLM behind LinkedIn Mentor has generated for the user.
Showing the user his full bio would be distracting. At this point, the user wants to know if the recommended mentor is the right person to answer their question or not.
So instead of showing Eric’s full bio, LinkedIn Mentor creates an AI-generated bio about why he is the right mentor for the user’s question, based on what’s found on Eric’s LinkedIn profile.
This feature could also incentivize LinkedIn users to create more robust, up-to-date profiles, for discovery on LinkedIn Mentor.
The personalized mentor description highlights what you actually need to know about Eric — AKA what he’s done that is useful to answer your question.
Only see the user’s relevant experiences for a faster UX
Being able to add your work experience to your LinkedIn profile is an integral LinkedIn feature. Without it, earning credibility and the trust of others on the platform would be tougher.
We want to show the recommended expert’s experiences to our users because it helps them make an informed decision, and see clear credibility. But something key for any search engine is that it should be fast, and having to read through an entire experience section to make an informed decision creates friction for LinkedIn Mentor users.
For context, this is what Eric’s full experience section on LinkedIn looks like:
Based on the user query ‘How do I raise money for my startup if I’m 16?’ this is what LinkedIn Mentor would show in a ‘Relevant Experiences’ section when recommending Eric:
Much faster.
Just how there is no need to see Eric’s full bio, there’s also no need to view all of his experiences.
So LinkedIn Mentor evaluates each of Eric’s experiences, and then only shows the most relevant ones. Relevance is decided based on how effectively each experience proves his ability to answer the user’s question.
This feature is made possible through the LLM behind LinkedIn Mentor that decides which experiences are more relevant to the user’s query.
Get personalized question ideas for a more frictionless experience
LinkedIn Mentor takes both the user’s query and the recommended expert into account to generate 3 personalized question ideas.
This turns the user’s original question into more personalized and relevant questions for the expert. This helps because:
- User will understand the value of this recommendation quicker
- User can spend less time personalizing the question
- Better questions get better answers
More value, in less time.
How this feature would be technically implemented: The LLM behind LinkedIn Mentor takes the user’s original question, and the text from the recommended user’s LinkedIn profile. When making the call to the LLM, prompt tuning takes place to turn the original question and LinkedIn profile into three more personalized questions.
Message the user, straight through LinkedIn Mentor
For the best LinkedIn Mentor experience, you need to be a LinkedIn Premium user to reach out to recommended mentors with more ease. This is a big win for LinkedIn, because its users will be further incentivized to subscribe to a LinkedIn Premium plan. This is why:
Users log-in to LinkedIn Mentor with their LinkedIn account.
This allows LinkedIn Mentor users to reach out to recommended LinkedIn users instantly.
LinkedIn Mentor does not just revolutionize identifying the right mentors — it revolutionizes getting in contact with them too, leveraging the LinkedIn platform.
Given the cost of running LinkedIn Mentor, it would likely be most effectively long-term implemented as a tool for LinkedIn Premium users.
LinkedIn Premium users receive InMail credits, allowing them to reach out to LinkedIn users outside their network. They would be using those InMail credits here.
On-Demand Discovery of Mentors, Based on What You Have Been Using LinkedIn Mentor For
LinkedIn Mentor is like that one well-connected friend in your life.
As your friend, it understands your needs, and can help fulfill them at your request.
Clicking the “Discover mentors” tab will generate a list of three LinkedIn users that could provide value, based on the user’s recent and frequent questions.
These mentors also come with personalized descriptions, instantly showing the user the personal value they can provide.
In this case, the user has been asking a lot of questions related to raising money as a 16-year-old. That’s why LinkedIn Mentor serves the user with three more people that can help with that.
Ambitious professionals want LinkedIn Mentor (Market Research)
To validate the need for LinkedIn Mentor, we asked students at TKS how often they would use LinkedIn Mentor. (For context, TKS is a 10-month global innovation program for ambitious high school students ages 13–17. TKS students have gone on to be the youngest employees/interns at Microsoft, Google, SpaceX, OpenAI, Neuralink, and more.)
More specifically, our exact question was: How often would you use a tool that let you ask any question about your work to get the ideal LinkedIn profile to reach out to for mentorship?
Out of 22 survey respondents, 31.8% said they would use it every day, 54.5% said they would use it multiple times per week, and 13.6% said they would use it once a month.
How LinkedIn Mentor Could Skyrocket LinkedIn’s Revenue Through LinkedIn Premium Subscriptions
LinkedIn’s biggest driver of revenue is its premium accounts segment, so that’s the revenue avenue we focussed on boosting with LinkedIn Mentor. Therefore, incentivizing users to desire cold messaging people on LinkedIn more will encourage more users to purchase LinkedIn premium, thus increasing LinkedIn’s bottom line.
Systems diagram of LinkedIn’s Mentor’s sales funnel
Breaking Down How the Technical Implementation of LinkedIn Mentor Is Feasible
On the technical front, implementing LinkedIn Mentor will mean leveraging LLMs (like OpenAI’s models) and LinkedIn Profile data.
There are 4 main steps:
- The user inputs their query.
- LLM selects keywords on which to query for LinkedIn profiles.
- Leverage retrieval-augmented generation (RAG) to feed relevant LinkedIn profile data into the LLM so it can match profiles to keywords from the user’s query.
- LLM compares profiles to check if the current profile is a better fit to the prompt than the previous best one. If so, replace, if not, keep previous one.
In the future, LinkedIn Mentor could also collaborate with other companies or leverage APIs to increase the user-recommended mentor fit. For example, leveraging Google Custom Search API to search Google Scholars.
Any powerful enough LLM can be used in place of GPT, as long as it can effectively interpret user prompts and select the best profiles to match the user’s query.
A way to streamline this profile search would be to use LinkedIn’s existing recommendation engine. This may not be as effective as the user will likely already come across these profiles without the use of LinkedIn Mentor, so it would only be used in edge cases to limit the number of pages required to scrape through.
Also, an internal LinkedIn API is not the most ideal for profile search due to it being time intensive to search through every individual profile when a query is inputted. When testing technical implementation, we used a third party API to scrape pages, which was not effective at all. Alternatively, it would be preferred if LinkedIn has other methods of storing all its profile data in a database that is easy for an LLM to read from (JSON, CSV, etc).
We Recommend the RAG Approach Because LinkedIn Profiles Are Dynamic
Retrieval-augmented generation (RAG) is an approach to improve the efficacy of LLMs by using custom, dynamic knowledge bases. LinkedIn Mentor requires specific data, and RAG allows the model to retrieve the relevant data from the LinkedIn database and use it as augmented context for the LLM. A RAG workflow pulls relevant information and connects static LLMs with dynamic data retrieval.
The keyword here is “dynamic.” LinkedIn profiles are ever-changing, and having to re-train or re-fine-tune a model every time a user makes a change to their profile is computationally near-impossible, or at least absurdly expensive. Feeding data to the LLM using the RAG approach ensures users are always receiving the most up-to-date user information, and LLM hallucinations are minimized.
With RAG, a vector database must be created to store the data. However, there are arguments that state there are more efficient ways than using a vector database — this really depends on the format of the profile data.
LinkedIn Is Sitting on an Opportunity To Accelerate the Next Generation of Professionals
There has never been a better time for LinkedIn Mentor. Recent advancements in large language models (LLMs), have made it possible to effectively interpret user prompts, analyze vast amounts of data (linkedin profile data), and provide highly relevant recommendations. These new capabilities, combined with LinkedIn’s extensive user base and wealth of professional data, creates an ideal platform to democratize access to mentorship.
Moreover, LinkedIn Mentor capitalizes on the current trends and needs of the professional world. Advancements in natural language processing (NLP) and information retrieval techniques, such as retrieval-augmented generation (RAG), enable more accurate and efficient matching of mentors to mentees. AWith LinkedIn’s focus on premium subscriptions and generating revenue from value-added services, a product like LinkedIn Mentor represents a compelling opportunity for the company to enhance its offerings and drive subscription growth.
We designed LinkedIn Mentor because as ambitious young people, it solves a problem we face every day. We want to build what we wish we had sooner, and would love to help make this a reality as soon as possible!
