The Tale of Credport
This is a post that recaps the product and business decisions we made over the course of Credport. Prepare for a long post with a mix of high-level overview and deep comments.
Credport started way back in the Fall of 2011 as a product of one of many philosophical discussions with Connor that generally range somewhere between meaning of life and merits of political systems.
Rachel Botsman’s TED talk solidified our confidence in a more decentralized economy in which traditional middle men, facilitating transactions via arbitrage, would be replaced by the technical innovations that would allow us to trust and transact with one another more efficiently.
Immediately, we saw that the then (and current) affairs of trust were suboptimal and we could think of two ways to improve it:
P2P Marketplaces were the way to go to disrupt individual verticals, yet there was no sufficient way of sharing reputation data across marketplaces in an efficient way. You reputation would be siloed in marketplaces, greatly reducing the potential reach of Collaborative Consumption.
There was a wealth of data available in the form of social network data that would allow us to more reliably increase both the objective nature of trust in overall risk reduction, as well as the subjective nature of trust in comfort. In particular, connectivity data leverages the truly powerful friend-of-a-friend concept. Not enough marketplaces were taking true advantage, which we attributed to technical limitations.
Besides these insufficiencies, we were also dreaming about the very obvious huge potentials of an implementation of the principles above:
With low confidence in contemporary protocol design and distribution, a system like this ultimately is only feasible when designed as a central system. In consequence, if successful, a central identity system for the Internet had many other applications. We were dreaming of a world with no lock-in effects and perfect data portability. As product geeks, the possibilities seemed endless.
A system like ours would become a very central and integral part of effectively a big chunk of economic activity. Even with minuscule margins, we would be rich as fuck.
It was hard not to get excited. It seemed like no one yet has came across that angle (I know right, how naive?). Thankfully, I have been able to work with Professor Byers to make this my Independent Project in the Spring of 2012, freeing up time from school work (which I thought was a colossal waste of time at the time).
The first Webcred Presentation
The first Webcred Executive Summary:
Basic Design Principles
The basic outline of such a system was very simple:
- Have a central website that ultimately holds the identity and associated data.
- Interface with relevant parties to facilitate the exchange of identity and associated data in an efficient way (read: great API).
- It should be consumer facing. We specifically wanted to avoid the situation of contemporary, in-transparent credit bureaus or online advertising trackers that effectively already consolidate your data across various systems, but generally do so against your favor. Our slogan would be: “Take advantage of what is already done and use your data in your favor”.
How do you get initial distribution for such a system?
We believed no one before has given enough priority to the above question. We knew that for distribution, we ultimately needed to satisfy two important criteria:
- The inevitable complexity addition by such a system to the User Experience would need to be held to an absolute minimum.
- The value-add to both the user and the marketplace, especially in the beginning without network effects, have to be clear and convincing.
- To the user, it needs to significantly improve their (often non-existing) trustworthiness and effectively chances of succeeding in a transaction.
- To the marketplace, the system needs to significantly increase conversion rates and trustworthiness of the marketplace itself.
For example, he hypothesized that the biggest player then, Trustcloud, had a suboptimal UX with an inconsistent signup flow. The trust increase because of the score and badges were irrelevant. Data exchange and hence trustworthiness increase were absent. Because of these issues, marketplaces were not adopting their system. In hindsight, we truly believed we improved upon these criteria, but ultimately, not aggressively enough and that didn’t end up mattering anyways.
The objective behind Webcred I was to create a functional prototype that would help us articulate the somewhat unusual position of Webcred between a consumer facing profile and tightly integrated API. In the original design, we effectively build an Open Graph clone that would allow third parties to exchange data with one another, unlike the original with Facebook being the exclusive consumer of pushed data. We always thought Open Graph was incredibly clever designed and remains in my opinion one of Facebook’s most under-appreciated efforts, but only the reuse of third party data would allow a true exchange of trust and reputation data.
The prototype consisted of 3 parts:
The Core System
The Facebook. Holder of all data which the marketplaces would be able to fetch through OAuth.
They would rely on us as their user account and trust system by plugging “Sign in with Webcred” just like “Sign in with Facebook”.
Third Party Validators
Our implementation of a decentralized system that allows infinite credentialing of arbitrary attributes.
- We were way too tied to the utopian idea of one-and-only-one-profile-everywhere concept. We very clearly overestimated the marketplace willingness to use us as as a source of profiles, it wasn’t until Credport where we realized we had to play a much more ancillary function on the marketplace. Something we could have easily figured out by just asking first.
- Verified Information was a typical “if we had scale” and academic feature. Not practical at all and quickly scrapped thereafter.
- We rather mistakenly focused on the exchange of data part way too much. We were thinking in Open Graph terms with Facebook scale in mind, which is now obviously very stupid.
- OAuth was the best compromise in security/privacy and ease of use across platforms. I cannot recommend the doorkeeper gem enough, creating an OAuth Provider would be way harder without it.
- The social connections feature was by far the most popular feature, which ultimately led us to make it our long-term emphasis by doubling down on Neo4j, the graph database.
Getting (not really) real
This is a really weird chapter to write in hindsight, but we didn’t do anything productive. We were not building a company, we were trying to spread an idea. Being super smug about people writing business plans, we didn’t really do much better either. Instead of finding and talking to customers, we spent the majority of our time polishing up the applications to YC and Summer@Highland and talking to people we knew for their feedback. We were sure that once we would have built the very obviously superior product, everything else would fall into place. SOOOO obviously never the case. Yet perhaps most weirdly, we managed to get into the highly prestigious Summer@Highland program. I think on some level, our exaggerated confidence in our product idea resonated with many people, especially in personal face-to-face communications. In hindsight, I blame a lot of our lack of focus on us successfully overselling what we had.
Summer@Highland: But Seriously
With the financial freedom and lack of pressure to find something practical for the summer (Connor dropped his GE and I my Facebook internship), we reached an important milestone and added a lot of legitimacy through Summer@Highland. We had big plans and thought we were the prototypical startup story: Connor would stay with his girlfriend, I would stay in a triple in a MIT frat. We would be the first one to enter and last one to leave the office, or stay over night all together. We were eating way too many cup noodles and Easy Mac. I remember to wake up one morning to the sight of HCP's limited partners and Apple’s former VP of Marketing Allison Johnson on an office tour.
On one occasion, SigEp, the fraternity I was staying at had a party. Let's just say I got fairly intoxicated and thought it would be a good idea to climb up the fire-escape. (it's not). A brother ended up yelling I should get down immediately, but I just took the first open window and jumped in. At the time, we were looking for some engineering help and lo and behold, the two MIT students in the room ended up helping us out for the summer.
Our game plan for the summer would look a little bit like this:
- Build the production ready system that we could actually sell with.
- Sell. (Note how it’s not the other way round like it should be)
So while I spent any wake moment cranking out a product that we were not qualified to make, Connor compiled a list of all potential customers and reached out.
Luckily, we were able to get Samir on board. He was essential while he worked in his free time next to his Microsoft internship in Seattle.
Our distribution strategy was:
1. Find small marketplaces, use them as “guinea pigs” since they don’t have a trust system anyways. We thought they were more likely to cooperate. 2. Gain initial set of marketplace to launch with and add marketplaces as we go by seeding the network effect.
From a product perspective, we spent way too much time on building with a set of features that we really couldn’t know were wanted. Although we enjoyed immersing in technologies we have not used on such a scale (Neo4j, Backbone).
Especially the “granting visibility” feature was completely and utterly irrelevant and I’m not sure why we did that. Privacy was a defensive argument we needed use to justify our centrality. I fondly remember sitting in a Venetian Las Vegas hotel room, showcasing that feature for our TechStars Seattle interview and looking into not-so impressed faces.
Following the previous tenor, we also focused way too much on some theoretical scale and interactions with third parties, when we really had no one. We normalized everything into Neo4j, thinking that calculating shared information was crucial, when really, only connections benefit. The single worst technical decision we made is to completely rely on Neo4j for our persistence. Not that Neo4j was bad as a database, but for traditional relational data use cases, you can’t beat a mature RBMDSs like Postgres. Instead, we ended up architecting this incredibly complex hypergraph with nearly 1-to-1 parity with Facebook’s OpenGraph objects up to a point where people’s college majors would be their own node in the graph. Needless to say, managing that complexity wasted a lot of our time, writing our own ORM/persistence layer instead of just being able to use ActiveRecord.
Under the pretense of Lean Startup, we also tried to get more potential user feedback despite lacking a functional product. We admittedly scraped users on Zimride and RelayRides messaged them with a survey about trust and their experience, we did some simple blind tests on room mate searches on craigslist, as well as survey people via Amazon Mechanical Turk. In all these cases, we had very clear feedback that our product was not needed. I’m not sure why we never acted on that feedback to be honest.
Though to be fair, we were also looking for alternative options, the all hailed pivots. We briefly considered going into roommate/renting or financial services where applications of increased trust seemed beneficial, but quickly decided that this is not where our interests lied.
By the end of the summer, we were still weeks away from anything usable. A lot of mockups and prototypes. Not only that, but we weren’t that successful with marketplaces either. We managed to get some interest, but as we would later learn, interest is worth nothing without commitment.
Nevertheless, we were in a similar situation like before the summer: There was no one telling us to get real, stop building a product no one wants and start selling (Though we did sign one pilot agreement during the summer). Despite objective slow progress, anyone we talked to expressed enthusiasm. That enthusiasm led us to seriously think about possibly not returning to college in the fall. However, my missing student visa would have been in the way, but coincidentally, Startupbootcamp Berlin popped up, and after a hopeful Skype interview with Alex Farcet, we realized how much sense it made: Further financing (we weren’t really ready for an actual round) and mentoring (from which we benefited greatly at Highland and SBC had plenty of marketplace leaders), ability to live with my parents and Berlin being generally cheap. Within just days, we booked one-way tickets to Berlin.
Berlin: The Next Chapter
While we were ready to make Berlin our next chapter, we weren’t quite willing to risk everything. We told ourselves that if we were not to get into Startupbootcamp, Connor would start his supposed study abroad semester in Dublin and I would return to Boston for my RA position. For me, that complicated our trip to Berlin as I only had a short window of applying to Visas and getting a flight back. In fact, I missed the first half day of the selection days due to my US visa interview as a backup (sorry Alex!).
Maybe unsurprisingly, the selection days kept the same level of enthusiasm. People loved the idea and industry. We also had very encouraging talks with the marketplace people, especially with Gunnar who was the Director of Airbnb Germany at the time. After a deserved short break, it was time to get back to work.
Again in hindsight, we pretty much kept repeating the same mistakes: Kept ourselves busy with work that was ultimately irrelevant and avoided doing the tough sell. We kept saying that our product was just not ready to show to anyone, hence kept our conversations with marketplaces at an informal level. We tried to gain a voice within the Collaborative Consumption Movement, and become a thought leader when it comes to trust and reputation. Connor did most of that outreach.
We realized that the US market was already dominated by few big players like Airbnb, TaskRabbit, RelayRides and figured that the rest of the world, including Europe, would be a lot more fragmented and hence easier for us to get a foot in. We learned that despite advances in technology, physical proximity is huge when it comes to partnerships, which is why our eventual pilot partners would be all German marketplaces.
From a product perspective, our goal was to drive adoption by marketplaces by making it easier to integrate and less obtrusive on their UX. While our previous iterations assumed a lot about marketplaces integrating with our incredibly expressive and flexible API, we eventually boiled it down to one line of code.
In the process, we ended up moving away from our overly complicated hypergraph system to a more sane relational model with parts of it persisting in Neo4j to support our fast Common Connection/Common Interests queries, while preserving our rich third party data support, where Marketplaces could push data like reviews to our system to be displayed on the users’ profiles.
From here on out, the product wasn’t really changing much, we built on top of our three pillars: Verification (of profiles, phone numbers, email adresses), Aggregation (of reviews, work/school history) and Connections (across arbitrary networks). The goal: humanize the marketplace user profile.
During this time, Rachel Botsman’s second TED talk about trust and reputation came out, and we felt validated by our aggregation/context > scoring approach etc.
While we benefited greatly from SBC’s connections in the local tech scene, we did feel like there was an oversupply of people we had to meet, and we ended up wasting a lot of time talking to people who were irrelevant to us and were just there to throw in their often unqualified opinions. We were at a stage we we just needed to get shit done really. The big emphasis on Demo Day didn’t help either. Connor spend the better part of of November preparing for the 5 minutes on stage, a distraction in hindsight.
Ultimately, we did end up completing a usable product (it had to happen eventually) and “soft-launched” at HyBerlin:
In the following weeks, we finalized our integration with Tamyca (for which we are eternally grateful), just in time for Demo Day. Shortly thereafter, Kinderfee joined us as partners. With that light at the end of the tunnel, we gained some more confidence. During the program, I personally exhibited some burn out symptoms, so it was great to hear some good news. That wouldn’t last very long though.
Christmas break hit like a thunder. While we enjoyed the time off a lot (Connor went home and I traveled with friends), the emotional disconnect for two weeks was deadly. Talks with investors just prior to the break died, talks with marketplaces had to be restarted, things were just not going very well. We assumed that once we actually had marketplaces on board, getting the next one would be significantly easier due to the network effect. That was simply not the case. Turns out it is very hard to get people from the “I’m interested, this sounds very cool”-stage to the “I’m in, sign me up”-stage.
Even with the marketplaces we already signed, we did not see the growth we needed. We weren’t really prominently featured in their sign-up process (which we totally understood), not enough people were signing up and worst of all, we couldn’t really produce any metrics to prove that we actually helped the marketplaces which was a critical presumption to convincing additional marketplaces. So the next weeks were all focused on improving our critical metrics. We had many little improvements, but nothing we could come up would feasibly have a big enough impact to change the game.
This is when realized the ultimate problem (rather late, but better than never): Trust just isn’t that big of an issue for marketplaces.
It seemed so antithetical to everything we’ve been working on up to this point, and that’s why it probably took us so long to get to this point. Everyone in the industry complained about how trust was a game changer, how it was so important, whole TED talks and conferences were being given for that sole topic. Everyone has read about scam stories on eBay and Craigslist. Even marketplaces said it’s important to them. However, that was only the PR game, internally, marketplaces have a lot of other things to worry about. Growth, market liquidity, user responsiveness, payments, etc. We learned about it firsthand from Gidsy, with whom we became great friends. And we needed that, because no one else was going to tell us, no one else had any incentive to be that honest with us.
Weirdly, as far as the public was concerned, everything was great. Collaborative Consumption received more treatment in the media and so did we. Here are some examples:
Gründerszene, TAZ, Netzwertig, Silicon Allee
We even had a class at CMU talking about us, as well as a Master’s thesis use us as a case study. As confusing as it was, the question became: Are we gonna keep going?
Obviously, we didn’t and here is why (as taken from our email we send out to our mentors and supporters shortly thereafter):
Trust is not an immediate problem
We talked to many marketplaces and heard "yeah, trust is super important for us." We heard many complaints about how sketchy craigslist is. However, looking at people's behavior, trust isn't the problem for peer to peer marketplaces - it's the incentives they offer. Trust is a problem that people often talk about, but don't really act upon. We assumed that increasing trust in P2P Marketplaces would increase their conversion rates tremendously. Not only did we fail to prove that, many P2P marketplaces need to focus on growth and getting critical mass before focusing conversion rates.
Not really a market (yet)
Our approach was a bet that many marketplace similar to Airbnb, but in different verticals, would spring up and grow quickly. However, it turns out that many of the collaborative consumption companies that were started didn't fare very well, and the ones that did have build their own sophisticated trust solutions. Our potential market, both in number of companies and revenue they were taking in is much smaller than we originally thought. Credport was really designed towards "many to many" marketplaces in the eBay style. Many of the companies doing well right now operate in a "one to many" style where one side has a much smaller population - Uber, Lyft, Handybook, etc.
Product vs. Competition
Though not necessarily why we failed, differentiating our product did cause a major timeline setback. Other companies had already attempted to solve the same problem in the same niche (trustcloud.com) - we assumed that it was because they had the wrong product, so we spent a long time developing a significantly different product utilizing the social graph as a focus. However, potential partners were more concerned with "why do I need a solution in the first place?" than "why should I pick you?"
What's up with Collaborative Consumption?
Collaborative Consumption is a serious movement, and will have some big long term implications - however we don't think that you'll be renting everything from your neighbors or strangers. Services like DriveNow (https://www.drive-now.com), which provide you access to an asset without ownership, but where the asset is managed by a company, are where we believe where most of the potential lies. That asset may be a good, or a service (handybook.com), but it will be simple and on demand. If we had to pick a phrase for this movement, it would not be the sharing economy, but the "technology-enabled access" economy.
We knew we were young and have, as cheesy as it sounds, a bright future ahead of us, so there was no point in trying to chase an infeasible dream. We still had our scholarships at BU and could graduate along with our friends. We enjoyed our time a lot doing Credport, but at the same time felt that we maybe tried to grow up a little bit too early.
What came after
After the admission of failure and defeat there was a mix of two feelings: Relief that it is over and being able to close this chapter on the one hand, natural feelings of anxiety and self-doubt on the other. Luckily, we had the unconditional support of people we have met along the way, easily the best thing that happened to us with Credport. That and learning a ton and being glad that we did it. However, saying it wasn’t hard would be a lie. I won’t go into further detail here, if you are interested in my personal struggles, read here.
Connor ended up returning to Boston for the Summer, I traveled to South East Asia until I ultimately joined Connor in Boston for our senior year at Boston University.
Read a bit more technical account here. As a bonus, here is our last ever presentation: