Shutting Twendly down - for now

Farewell HiveMind, Introducing Tribalytic

It's been far too long since we've updated you on what we are up to.  Time to rectify that!  Those that follow Tim on his blog will probably be more up to date, but it's time to let BinaryPlex followers know what's been happening and where we are at.

Like all startups we have our ups and downs and the last few weeks have been no exception.  For the price of entry it's proving to be an interesting ride.

When we formed, our initial goal was to build software to search and locate experts inside large enterprises.   As we started discussing this problem with people who had signed up for our beta we started to see some consistent challenges:

1. There was incredible diversity in the response to the problem - people had clear views on what it would take to address the challenge, what content should be mined and how structured and unstructured content should combine.  We struggled to find enough consistency to identify which part of the product we would build that could appeal to more than one or two customers at a time, particularly in the beginning.
2. There was insufficient pull - lots of people were interested, a few were engaged, but generally speaking, the problem was looking like a nice-to-have, not a must-have.  With the economy where it's at in the US and the UK, we decided it wasn't (if it ever is) a great time to launch a "nice to have" business.
3. There was generally speaking an expectation that it would be a "point and click" type experience.  The reality of course is that the technology required is a fairly complex installation and required some substantive resources both in hardware and time for any meaningful implementation.  We could also see that there were barriers of entry around enterprise technology choices (must run on <<insert technology choice of preference here>>) that were going to be difficult to overcome - our resources were directed at solving the problem in the most efficient way possible (which for us meant Linux and lots of open source), not catering to Enterprise requirements on technology platforms.

Rather than simply drive for the market, we really began to question what sort of organisation did we want to be and what did we want to do?  Some things were clear - with limited resources we need a single point of focus, we wanted something clients could experience quickly and easily and we wanted to develop in the technologies that were best suited for the task.  Finally we couldn't let go of the work we'd been doing with the HiveMind engine - the challenge of finding and understanding people was just too interesting.

With these thoughts in mind we launched Twendly.com to help address point two - let people feel and touch what we were doing.  People really responded and engaged with Twendly far beyond potential HiveMind customers and we have seen sufficient traction that we believe there is a market interested and ready for a solution in this space.  We also realised that a SaaS model would help us address number three listed above.

We took this early prototype, our market research and our response to Twendly to Sydney and started pitching to Angels and VCs.  These are smart people who've been around and throwing ourselves into the deep end helped us learn and iterate very quickly across a week to really focus our idea onto a market with a big enough problem that we could address and help address point one.

Which brings us to today.  We've made the tough decision to farewell HiveMind and instead introduce Tribalytic, a social media analytics tool that shares a lot of common heritage with the work done for HiveMind.

Using the technology developed for HiveMind, Tribalytic goes beyond a simple monitoring of mentions and instead deliver fast analysis and segmentation to help products and brands Explore, Discover and Engage with the tribes holding conversations about them.

While we regret that we won't be delivering HiveMind, we are very excited about Tribalytic and this new focus has helped us drive more rapidly towards a product which we anticipate should be in beta by early April.  If you want to go beyond simply knowing that 300 people tweeted about your brand yesterday and actually start to understand more about who these people are, what drives them and interests them, we invite you to visit us at BinaryPlex and sign up for the Tribalytic mailing list.

* Rollercoaster image by Scott Ableman,licensed under Creative Commons.

Reflections on start-up life: Week 10

Last week was very interrupted for a number of reasons - family events and then a HDD failure on my main PC which stopped me dead in my tracks.  This turned out to be a blessing in disguise as I'd been putting off moving to Ubuntu, so this was my opportunity.  Fortunately I lost only time as all our data and code is well backed up - DropBox is a blessing for documents as well.

I won't rave about the Ubuntu install other than to say it took me around 1 hour to get a functioning Ubuntu machine up and running - almost 8 hours for the Vista re-build (patch, reboot, patch, reboot all day).

So the last weekend wasn't as relaxing as I would of liked, mostly consisting of building new machines (2 boot images on my main desktop and I also purchased a new MSI X340 laptop I installed with Ubuntu as well).

The reflection this week for me is that decision leads to opportunity.  Being decisive about things opens doors and provides more information - sometimes it might be that the decision was wrong, which is still good information, but it does lead to action.  Indecision kills start-ups. 

Having decided to focus on the consumer play we found that we could start to do a number of things:

1. Write a pitch deck and circulate for feedback.
2. Act on the feedback to refine and define what we are doing.
3. Have more focus in our meetings.
4. A clearer idea on what we "want" when we meet with someone, and exactly what it is that we have to offer them.
5. Actually forecast a plan for two weeks in advance.

Technically not much happened from my perspective (although Alex has been doing a lot of planning on his side), but from a business perspective we made some major steps forward.

We will be in Sydney all week next week - if you're following this blog and interested in catching up, let me know - we are there to pitch to potential investors, but we would love to meet with anyone who is interested in what we are up to.

Twendly continues to do very well and is gaining some real evangelists for the service.  Again, doing leads to learning - without a real live demo like Twendly we could never have learnt as much as a we have about our HiveMind engine and how it works in the real world.

• Wasted time rebuilding PCs.
• Still not enough time on the business plan as is needed.
• Hating the touchpad on the MSI - need to get the proper driver for it so I can turn off the stupid touch to click.  Why anyone would think this is a good default behaviour is beyond me.  I constantly brush it with my thumbs on it when typing and end up typing somewhere else in my document.
  

• Refine and rework the business pitch so we are good to go for Sydney next week.
  

What next for Twendly

What a week it's been!

Twendly has grown and grown after TheNextWebAU article was published, sparking a wave of interest across the globe which has been fantastic.  We've also seen something in the order of around 500 retweets about Twendly as people share the news.

Our initial suspicion that there was something useful about this way of searching for people, has now been have been confirmed by lots of enthusiastic people.

So where to from here?  There are a few things that we are focussing on now in the short term:

1. We've decided to put the enterprise version of HiveMind aside.  Twendly needs our full attention to realise its potential and we have lots of great ideas we want to build into it.
2. We've expanded our server capacity - as the search load increased, we've run into trouble both keeping the search up AND also loading and indexing the data to expand the search without dramatically impacting the search experience.  A second server will help.  This should come on line this weekend and will enable us to index new users much faster while still keeping the search experience for existing users running well.
3. Catching up! We experienced a massive surge in growth that we were not properly prepared for - we are working hard to get all these new sign ups into the index and deliver the experience we have promised.
4. Squashing Spam.  We have had an anti-spam component to the engine from day one, but we have refined it a lot over the last week as we get more and more accounts indexed.  Roughly 10% of accounts the we index, we consider spam and we remove from our search results. Our ultimate goal is that when you search you will find a real person, not a bot or a person who tweets like a bot.

You should start experiencing the benefits of these changes by next week.

In the medium term, we want to re-write the search engine from scratch and include more sophisticated algorithms for people matching.  Twendly works quite well at the moment based on the feedback, but we know that it can be a lot better and for us to continue to scale we will need a different underlying architecture for the engine.

Thanks to everybody who has tried Twendly for your enthusiasm, your tweets and your blog posts - we are excited about the tool we are building and are looking forward to seeing what we can turn this into with our 100% focus and your support.

Cheers,

Tim Bull (@timbull) and Alex Dong (@alexdong)
Co-Founders of BinaryPlex (@binaryplex

How does Twendly differ from a Google or Twitter Search?

We are really excited about the traffic we are getting on Twendly - it's great to see people getting value from it.

Of course, in a world that is largely used to searching for documents, it can take a slight "twist" of thinking to easily understand the difference between a document Vs. a people search.  We'd like to illustrate this with some examples.

In Twendly, put simply, you're searching for the things people are talking about.  If you search for your name, you're searching for the person who talks about you the most, not necessarily yourself (unless of course you talk about yourself a lot!).

These three screen shots help to summarise this difference and give a good illustration on how it works by comparing a Google Search, a Twitter Search and a Twendly Search using exactly the same search terms "Lotus Connections".

In the following Google Search you can see that a search for Lotus Connections returns documents and sites-  it's shown you the most relevant locations to go and research further yourself.

In Twitter search, you see the people who are mentioning Lotus Connections "right now".  This also has its place - you can join in the current conversation.  The problem is that although these are people, it doesn't give a sense of who the relevant ones are - perhaps the leading people are silent just at the moment.

Finally there is a search in Twendly.  It returns people, with an easy gauge on both how much and how frequently they talk about Lotus Connections.  Having had some experience, I can vouch for 7 of these being excellent people to talk to about Lotus Connections and the rest as good information sources.

To keep the test fair, we used the same search terms for each tool, however there are lots of ways in each search engine to refine the search results.  If you want to refine the result in Twendly to exclude people (or Twitter bots in this case) who tweet lots of links, a handy tip is to include -http at the end of your search e.g. http://twendly.com/?q=lotus+connections+-http This simply says don't include any thing mentioning HTTP.  Depending on the search topic and the type of people you are looking for, you'll find the relevancy often improves further if you do this.

 

 

Launch of Twendly

Yesterday we launched http://twendly.com which is a Twitter people search implementing a version of the HiveMind engine we have been working on.

It's still very much a preview of the concept, but it does provide a hands on implementation that you can quickly experience to see the concept a lot more clearly.  There are lots of features we are working to add in, but for now this is the very core of the idea - you search for information and we return people.

Even if you don't have a Twitter account you can still search and play with it - you'll get the basics of the concept very quickly, but if you do have a Twitter account, we'd love you to sign up and let us index all your tweets to add more users into the database and make the demonstration more meaningful.  We'd love to get a couple of hundred people signed up over the Christmas period, so feel free to invite and encourage your friends to join in too.

What we aren't demonstrating here is the plug-out concept yet - we've implemented a Twitter plug-in, but we aren't returning the data yet.  In an Enterprise implementation (for example a Wiki) you'll be able to plug-in various different sources and then build and save Topic searches.  The results of the Topic searches (for example Lotus Connections Experts) will then be able to be assigned through plug-outs back to source data, for example labelling the top 20 individuals who match the Topic with the topic name (i.e. tagging them as "Lotus Connections Experts") in the source system.

We'd love to hear your feedback, so please let us know what you think using the feedback on the site.