About a month ago we launched Recotype a recommendation engine built on Twitter.
Our goal with Recotype was to make it dead simple for people to get recommendations from their followers on Twitter — Just add #reco to any tweet and we do the rest.
We could see 2 things that inspired us to build Recotype: people using Twitter for recommendations and shortening ‘recommendation’ to ‘reco.’
So we took a leap of faith to see if we could do create something interesting with those 2 observations.
We built a quick application and started to get real feedback on the key assumptions behind the Recotype model:
- Would people know how to use has tags at all?
- Would people use #reco?
- Would enough people use #reco?
- Would we be able to pull together all the pieces of the conversation and really add value for folks?
- Where in the mix would be the real value we could provide?
And since launching we’ve had some incredible responses, feedback and usage of #reco. It’s been awesome.
We’re really thankful for all the cool people who have picked up the #reco and run with it.
Our first 30 days provided a ton of the lessons we wanted them to provide.
And since so much of the success of Recotype relies on people using it, we want to share those lessons.
So here goes — our lessons from the first 30 days of Recotype.
Locality Rules in #Recos
When it comes to recommendations, the location of the recommendation is almost always important.
We don’t have hard data to share on the percentage of recommendation specific to a location, but I’d guess about 75% of #recos are specific to a place.
That place is most often where the person seeking the #reco lives. But it’s also often associated with travel or a new place they don’t know well.
Some examples:
We also noticed people started to add airport codes as additional has tags to their #recos to mark them as specific to a place.
#Recos Have Patterns
Reviewing the #recos of the past 30 days, we see some key patterns emerge.
These are the big 4:
- Declared #recos — someone says ‘I #reco this.’ just because they want to.
- Requested #recos — someone says ‘Anyone have any #recos for …’ because they want information and help from their followers.
- Provided #recos — someone responds to someone and says ‘I #reco this for your context…’ because they want to share and help.
- Passed #recos — someone retweets a requested #reco to pull from a wider network of people.
We also see another kind of #reco pattern that deserves its own section.
It’s so explosive, we try to describe it below in the section Networks Explode #Recos.
Fast Is Good; Not Always Best
We originally thought that speed was going to be the key factor in whether a #reco thread was a success.
And we were right, sometimes. Some #recos are time sensitive and only answers within a few minutes or hours are valuable.
But many more are not too time sensitive. They’re middle sensitive. They need input in a few days but not in a few minutes.
We found that many #reco threads stay active for a few days at a time. New people chime in, new replies are added and conversations continue.
And a few #reco threads go dormant then revive after a period of time.
So speed isn’t necessarily the key factor.
#Recos Are Alive
As an addition to the assumption above that speed was a key factor, we originally thought #reco threads would flare up and burn out quickly.
Old threads would be good archival materials but static. We were wrong.
What we learned is that #reco threads can stay alive in a few different scenarios:
- When the same #reco is requested and someone can point back to an existing thread as the answer.
- When new information or context or people connect into the #reco thread, they want to add to it.
- When the person who requested a #reco returns to the thread to provide feedback on the #recos received or their experience — to make the thread complete.
So far the application we’ve built is really limited in how you can search past #reco threads, how you can discover #reco threads, how you can add to #reco threads and how you can add to #reco threads with the new information, context or people.
We have to change this. We’re working on ways to do so.
#Recos Blend
We’ve found that recommendations blend along the lines of how people using Recotype have their networks mingling.
For instance, many of the folks on Recotype combine personal and professional interaction in their Twitter account.
One moment they’re seeking a #reco for a course of action for when a client domain expires before being noticed and gets scooped by a porn site?
The next they’re seeking a #reco for a manufacturer of quality sportswear not covered in crappy logos? I hate logos (ironic, I know…).
So we’re just letting that all blend together.
In addition, many of the folks on Recotype have their Twitter account hooked up to their Facebook account.
So when they ask for a #reco both their Twitter followers and Facebook friends see the request. They get some responses on each network but we’ve only hooked up Twitter so far, so they only see those responses on Recotype.
Lastly, many of the folks on Recotype get both public and private responses to requests for #recos.
So we’re challenged.
On the one hand, we want to pull together the disparate fragments of information into a comprehensive #reco thread — to make it simpler and easier to get help and the right answer.
On the other hand, we want to preserve the mode and settings of the communications channels — ensure private communications remain private, keep Facebook feedback distinguishable from Twitter.
And if we add on more professional-oriented networks like LinkedIn the challenge doesn’t get smaller.
We’ll use the principals above to guide our decisions but we’ll still have to make good, interesting, challenging decisions.
Networks Explode #Recos
The most interesting and unexpected lesson we’ve learned since launching Recotype is that networks can explode the value of recommendations.
What do we mean?
Megan posts Need best ever earbuds for running, no over-the-head kind – Suggestions pls. #reco.
I see her request but don’t know the answer. What I do know is that my friend Corey has some earbud-style headphones he raves about. So I loop Corey into the #reco thread.
In 2 simple steps, Megan has a personal recommendation from someone with expertise and experience in the area of earbud-style headphones.
But maybe your question is more technical, like is it wise to serve JS code off of github or should I serve it off my own server?
That might as well be Greek to many folks, right? But not to a javascript expert like Brian Leroux who also uses GitHub.
Brian gets connected into the thread and has the right answer, (your own CDN).
And there’s something really interesting in that dynamic — matching the right person to give a recommendation to the request for a recommendation.
That simple action of bridging the #reco request and singling out a specific expert creates a ton of value.
So how do we encourage people to make those connections and bridge the recommendation to the right person?
Sounds like another good challenge — one of the top ones we’re working on right now.
Your Advice Needed
Last up, we want to know what you think of the first 30 days of recommendations.
Have your tried out the #reco tag?
Do you have questions about how it all works?
Are we missing anything that we need to know about?
Tell us. Comments below for public action.
Email at recotype@adhack.com. Hit the AdHack contact page. Tweet us @recotype
And if you like Recotype, please make it one of your #recos!
Tags:
AdHack Labs, feedback, lean, learning, lessons, projects, recommendations, Recotype
More AdHack Goodness? Subscribe!
AdHack Blog by RSS | 
AdHack Blog by Email
AdHack: Talent, Tools & Tactics
AdHack works with agencies and brands to develop nimble, responsive, compelling digital campaigns and movements.
Here's the link to learn more.
