Cohort — A New Professional Social Graph
“Professional? Social? Isn’t that what LinkedIn is for!” Nope.
This week marks 12 months since Alex, Brendan, Eoin and I started working on Cohort. We want to make it easy for you to find the people you need through the people you already know and trust.
So, I wanted to take a moment to reflect on why we’re doing this, what we’ve achieved so far, and why Cohort is in private beta, and still pre-launch. (Hint: this shit is really hard!)
Couldn’t we think of something easier to work on?
When I started thinking about this space a few years ago, I wanted to make it really simple to make introductions. At first, I thought the best way to do that was to automate the double opt-in intro. I soon realised, however, that before you can introduce two or more people, you really need to understand the world they live in, all the moving parts around them, the hows, whys and whens.
After all, why make an introduction, if you’re not reasonably confident of a successful outcome? That got me thinking about social capital — how it’s acquired, maintained, lost — it’s a lot like real capital in many ways.
I started to think about the kind of social graphs that exist today, and how they’re rarely reflective of relationship strength, or representative of the social capital contained in them. People try to use their real world networks everyday, and intuitively know some of the people they can count on for help… and it’s usually because of the existence of social capital.
Maybe it’s because social capital is intangible, that it can sometimes feel like a fuzzy concept. Indeed, a number of different definitions and perspectives have been written on the subject over the years. When we started to think deeper about a social graph built around real relationships and social capital, we needed an easy way to a) explain it to people, and b) make it a dynamic component of the Cohort product.
In the context of what we’re building with Cohort, we think of social capital as the ability to ask someone for a favour.
To be able to do so implies that you have spent time with each other, you have shared experiences and shared conversations. There is trust, open communication and the ability to collaborate — the foundations of a healthy relationship.
This is important, because a network cannot be leveraged without social capital — favours. If you have a list of people who you are “connected” to on LinkedIn, and you can’t ask a favour of any of them, then that is not your network — it’s just a list of people exchanging invites to connect.
There’s another important implication here. It is much easier to ask someone for a favour if you have already done one for them. LinkedIn’s push to get people to “connect” has created a culture that thinks simply hitting that blue connect button is how you “Network”. Really, all that is required to network is to be willing to offer to help first. This is how you start to acquire social capital.
Over the years, I’ve noticed that people I’ve met who are most successful (by whatever definition makes sense to them) are those who ask me how they can help. If, when meeting someone for the first time, they offer to help, I immediately hold them in higher regard, and it’s usually the start of a real, meaningful human relationship.
There are three main problems with human networks, as I see it, today.
- Fragmentation. Our relationships, and the social capital contained within them, are fragmented across multiple domains — social networks like Twitter, Facebook and LinkedIn; and communication tools like email, phone contacts, text messaging and IM. Some relationships only exist in real life, and not anywhere online.
- Reach. Even if we have a reasonably good idea of who we have social capital with, it’s hard to discover that about their friends. The very best human networks are built on strong Friend-of-a-friend relationships, yet it is surprisingly difficult and time consuming to understand not only who knows who, but how well they know them, what those people are knowledgable about or have interests in, and how can that be leveraged to a positive outcome.
- Application. Because of fragmentation and lack of insight to reach, it is hard to know where and how to apply the social capital we have worked hard (whether through conscious effort or not) on acquiring through out our lives and careers.
Have you ever been asked by a friend or colleague if you know someone who can help them with something? Maybe it’s a skill they need in work. Or maybe it’s someone at a particular company they need an intro to. Or maybe they’re just looking for a good recommendation or piece of advice.
Were you able to answer them immediately? “Yes!” you say, “I know just the person!” Perhaps. But in reality, this is rarely the case. And it’s because of the three problems outlined above: Fragmentation, Reach and Application.
This is the problem that we’re solving with Cohort. People don’t know how to effectively, easily, realise the untapped potential in their networks. And guess what? Neither do organisations.
Every day, people turn up for work at your organisation, and they bring with them all the things they need to do their jobs — qualifications, skills, expertise, experience — but they leave their networks at the door.
“Must have a great network and deep understanding of how to leverage it” — said no job description ever.
No one has ever been asked in an interview about their awesome network of peers. And even if they were, where are the workflows? How do people, teams or organisations effectively realise the untapped potential in their networks?
100 million people & 700 million relationships…
…and growing
When we started a year ago, we asked ourselves a question: How can we make it easy for people to understand their friend-of-a-friend networks and what are the things they can do with that.
Turns out, that’s a complicated question to answer, never mind packaging it up in an easy to consume interface. The first thing we needed to do was take a stab at understanding a big chunk of people and how they were related, and what the kind of things they were interested in.
We believe that if we can get a reasonably good idea of this, then we can probably make a decent effort at recommending when and how people should be introduced. And in doing so, we solve the the problem of Fragmentation, Reach and Application.
As of today, Cohort is a social graph of over 100M people, and over 700M relationships. We are understanding 10,000 people and their networks a day.
We take public and private sources of data, run it through models we’ve spent the last year developing, to create a social graph that is concerned with proximity of relationships and interests. It looks something like this:
It’s not an exhaustive list, but here’s some of the things we are doing. Watch out for a follow-up post that will dig deeper into this.
- Conceptual Search and Recommendation — generating relevant information to bridge the gaps in specific queries, i.e. “AWS” is part of Infrastructure.
- Natural Language Processing — mapping natural language needs to queries, to surface relevant helpers.
- Deep Learning — used in both conceptual search and natural language processing, deep learning is an advanced machine learning technique we use to form contextual connections between terms.
- Classical Machine Learning — a number of models we use for relationship and knowledge prediction draw on well-known machine learning techniques.
- Social Network Analysis — we analyse people in the context of their networks, detecting communities they belong to and their importance to those communities.
Cohort learns over time. We make recommendations, and through usage, we gain insight into real human relationships, interests and skills. We take that feedback and work it back into the models.
So, when you use Cohort, you are not only helping to improve it for yourself, but also for everyone in your friend-of-friend network — your cohort.
Making the complex, simple and immediately useful.
As you can see from this tweet I posted only a few weeks ago, we are starting to accelerate the rate at which we consume, analyse and graph relationship data with context and meaning. In only a few weeks, Cohort’s graph has grown from 360M relationships to over 700M.
So, how do we make sense of so much data, and provide relationship insights to people in a useful way?
We’ve focused our product decisions on first principles/how things work in the real world. Which has helped clarify some fuzzy areas.
From the outset, we decided that real world social dynamics should drive product decisions. If we couldn’t imagine a scenario happening in real life, then no amount of clever UX or application of data could make it work.
Our priority is helping people discover who in their friend-of-a-friend network could help them solve a problem.
Keeping this in focus, helped us work through key parts of the overall experience, for example:
- Using favours as a driver of value creation made it easy to frame relationship closeness, both to the user, but also as part of our ability to predict the same.
- Framing network search around problem solving. i.e “I’m looking for someone who can help me with sales” or “I’m hiring a product manager with mobile experience in San Francisco”, — searches can be interpreted by the machine to make recommendations, but also read by people in your network, paving the way for a more social search.
- Using a Broker-Helper paradigm as a means to present options for solving the problem. This makes it easier for the user to sort through people who can potentially help out.
- Team collaboration, where a number of friends can combine their networks to achieve their goals, effectively piggy-backing on favour exchanges made by others in their team.
- Building in reciprocity, where we balance the idea of self-interest with helping those in your community and extended community. This is such a fundamental part of good networking practice, and Cohort is able to show users how they can help those close to them.
A 52-week sprint 😱
We knew this was a big problem to solve, and we expected it would take us time to get to a point where we could demonstrate power, utility and ease of use. If we were to start on this problem today, knowing what we know now, it probably would take less time, but not by much.
A product like Cohort is technically complex to build, and takes time. This isn’t a problem that can be solved simply by building on top of LinkedIn, or syncing up phone contacts.
We set out to understand the world you live in, all the moving parts around you, the hows, whys and whens — so when the time comes, and you need to ask your network for help, Cohort can point you in the right direction. And when your network needs you, Cohort lets you know how you can help.
Before I go…
As an early stage startup, we mainly exist because a handful of people were able to see a better world with Cohort in it. Shout out to Jonathan, Martin, Curt, Karl, Dave, Dylan, Harper, Joe, and Sean — the true believers ✌️💰
Thanks also to Matt, Gil, Eoghan & Laura for listening to me ramble on about this stuff over the past two years, and all those who have tried out and given us feedback on the beta.❤️
If you’d like to sign up to Cohort’s beta, you can do so here.
