- High signup-to-subscription conversion rates
- Another 13M net photos added (i.e. taking into account deleted photos from accounts leaving trial)
- Good feature work from Highlights view on website to AirPlay support in iOS app
- Lots of unexpected maintenance on our infrastructure
Like in December we had about 1,600 signups this month. However out of a total of “only” 9,000 trials since 1.0 launch, we have now close to 1,200 subscribers (keep in mind people wait the end of the trial to subscribe so it’s closer to 1,200 subscribers out of 7,400 trials).
Now that we have a few months of data, the very interesting data point is how signup-to-subscription conversion rate increases significantly with product experience:
- 14% among all new users
- 20% among users importing at least 1 photo
- 34% among users importing at least 1 photo and also trying Mac or Windows Uploader
- 42% among users importing at least 1 photo and also trying Mac or Windows Uploader and iOS app
We assume with these stats causation and not correlation since we have a 30 day trial period and the subscription price at $40 is unlikely to be an impulsive purchase (i.e. few users if any would barely try the product, wait the remaining 30 days, finally subscribe, and then only decide to try the uploaders or iOS app to get the full experience).
We’re told these conversion rates are quite high which means we’re onto something. We are fully aware these high rates could go down when we start reaching a broader range of people (instead of mostly early adopters), at the same we have a lot of room to improve our conversion funnels. For instance, we are still working the email flow to keep trial users engaged during the 30 days (today, we still send a pair of generic emails at the very beginning and very end of the trial).
What this also means is that today we don’t have a product problem, far from it, but essentially an exposure problem.
We signed a contract with Allison Partners (http://www.allisonpr.com/) starting mid-January to represent us in the next 3 months. This firm was highly recommended to us by people connected to Index Ventures and certainly gave us a good impression among all the firms we interviewed.
Allison Partners is a national PR firm which typically doesn’t work with startups pre-Series A, even less on short term assignments, but they really liked Everpix and we were able to work something out. Their goal is to get us widespread coverage, both regional and national i.e. getting out of the Silicon Valley bubble. We work with their San Diego and New York offices to ensure good coverage on both coasts.
January was a difficult month for our infrastructure as you can see if you follow updates on http://status.everpix.com/. It’s all due to an unfortunate chain of events (aka “perfect storm”):
- On January 9th, we got covered by TUAW (http://www.tuaw.com/2013/01/09/everpix-transforms-overwhelming-photo-archives-into-enjoyable-me/) which brought a surprisingly high number of visitors (10,000+ over the week i.e. about the same as for a good TechCrunch article).
- Our invite system had its rate-limiting feature disabled, so people signing up immediately received their invites and started importing photos.
- The infrastructure was rapidly importing more than 2,500+ photos per minute and scaling up.
- Soon afterwards, Amazon Web Services ran out of the fancy EC2 machines we use for image analysis. This prevented our system to continue scaling up and a photo backlog started to build up.
- Because AWS was incorrectly reporting the error, we didn’t notice it until hours later. When we switched data center, the backlog was huge (more than a million photos).
- Our image analysis is heavily parallelized so it was able to work through the backlog fine. However, the next step of adding the photos to the user accounts is mostly serialized (i.e. 1 photo at a time per 1 user account), which put a lot of pressure (i.e. I/O contention) on our database servers.
- At the same time our databases became quite full and painfully slow compounding the problem (even after increasing storage and upgrading to the highest performance AWS databases).
- By this time the entire system throughput was abnormally low and it wasn’t getting any better.
We addressed the root problem by bringing online new database servers and having new users signing up from that point on created on them. Then we had to figure out a way to restore performance on the database servers containing the old users and deal with backlog. That second part took about 3 weeks of on and off maintenance work.
In conclusion, in the 48 hours after the TUAW coverage, more than 5 millions photos were imported to Everpix, a number we should normally be able to handle fine. Clearly that wasn’t the case here and the opportunity cost was high: we had to spend countless hours fixing the infrastructure instead of focusing on building the product.
Going forward we now have 10 database servers to distribute the load as best as possible while preserving the overall system performance. We also use multiple data centers for our image processing machines. Finally, we have started looking at external consultant firms we could use for Sys Ops.
Sometimes you may wonder why we spend so much engineering time building our infrastructure and a concrete reason is what happened this January. We are dealing with very large amounts of data combined with very computationally expensive image processing that early stage startups typically don’t have to. To make our ambitious photo platform accessible to more than a few thousands users, we must design and build a very scalable system; there’s just no choice.
January expenses were particularly high ($140,000+) due to several factors:
- An exceptional $16,000+ AWS infrastructure invoice (as explained above)
- A $15,500 invoice for the first month of work from our new PR agency Allison Partners
- A $8,000 late invoice from our Marketing consultant for November which was paid in January instead of December
- A one-time $5,000 Mechanical Turk bill to crowd source the tagging of thousands of photos to train the updated image analysis system
We now have a new system to track all our metrics based on Chartio (http://chartio.com/). We built it to work with everything from internal infrastructure metrics to AWS spending, Google Analytics, App Store reports, etc… It’s also fully automated and much easier to maintain.
As a consequence, the KPI format is changing in this report compared to the previous ones: it’s now just a couple of PDFs exported from Chartio. Unfortunately not all previously tracked KPIs could be implemented in ChartIO for technical reasons but we’ll improve this going forward.
Meetings with VC to raise our Series A have started very late January. The target is USD $5M and the goal is to wrap it up within 3 months.