Hacker News new | comments | show | ask | jobs | submit login
Evernote CEO Shares The Numbers that Makes Freemium Work (toddsattersten.com)
108 points by toddsattersten on Feb 6, 2010 | hide | past | web | favorite | 24 comments

Some metrics:

From the article:

60 servers

35% active users (defined as being active a given month)

2,335,676 total users

41,598 premium users

38972 users per server

4 sysadmins

$68,641 - Total variable expenses (hardware + software + hosting + network + operations staff + support staff)

$145,000 - Total revenue from active premium users

From this we can deduce some interesting numbers:

free/paying ratio (all users) - 1,7%

free/paying ratio (active users) - 2.2%

revenue per user (all users) - $0,06

revenue per active premium user - $3,5

Variable expense per user - $0,29

38927 users per server

583919 users per sysadmin

I particularly find the $0,06 revenue per user interesting, since it gives a rough number to work with if you're doing a freemium website. Eg. if you need $10.000 in monthly revenue to break even you will need 166.666 users.

I'm starting up my own tiny business and have been spending a great deal of time understanding how to optimize every part of the product infrastructure to minimize cost (to a reasonable level).

$68K per month blows my mind if the number represents monthly expenses instead of yearly given the number of active users. I've an iPhone game with over 3MM active unique users and by my math I can host a multiplayer solution for less than $800/month. Granted, my service will not be sending giant PDFs, notes, voice memos, etc. but still... What am I missing? I am not being snarky. I need to learn.

Also, is the 4 ops guys per 60 servers a non-shocking number? When I worked at Yahoo, the ratio of ops staff count to production server count was IMHO ridiculously high and I never understood why it was required. Sure, some redundancy and so on. Clearly I'm missing something though - what?

I'm eager to correct my misunderstandings if I'm the one off base here. Can you recommend where I might find similar (perhaps more detailed) descriptions of production environments by other companies?

"Also, is the 4 ops guys per 60 servers a non-shocking number?"

4 people is the bare minimum for realistic 24/7 support. Even if you just do "on call" support outside office hours, making sure someone (capable) is available every weekend and at o-dark-thirty every day is difficult with less than 4 people.

(Says a guy who spend too many years as part of a 2 or 3 person sysadmin team...)

I suppose it depends on the thing(s) being monitored though. I certainly do not want to have 4 people for my always-on development project. I am designing the system to be as fault tolerant and self-healing as possible so I do not have to have people (read: me) awakened at 3AM with alerts.

Maybe I'm insane or overly optimistic or both. Designing things to bend and not break is really difficult and time consuming. "Thanks Captain Obvious! You are so insightful!" No, I know, but I'm going through the cutting-of-teeth ritual now so these kinds of things are on my mind.

It doesn't necessarily depend on the things you are monitoring. If you are running one server, don't sweat it - if you are running anything more than 20+ server (my preference) and the capacity is constantly growing and you still don't have a 24/7 sysadmin. I think you are doing it wrong.

The key point is, if your business model depends on more user usage ie, more uploads more money (on a 60 server scale), you would like to have someone standby all the time.

I've got the same question. Anyone with an idea of why the monthly expenses would be so high, and why so many ops guys for just 60 servers? I don't want to be so cynical as to assume it's because of bad management or something; I must be missing something.

Just speculating.

Assuming they have a sysadmin available 24/7 (which they should) 4 employees is hardly too much. Its just enough.

When your business is on the cloud with 60 servers, there should be at least one person available at all time, whether the server ever goes down or not.

I'm aware that some Europeans use commas to do their decimal separator, but I've never heard of anyone using the same symbol for both decimal separator and thousand-multiple separator. All the same, thanks for breaking out the numbers.

I'm ashamed to say that it boils down to two factors: Laziness and copy/paste


I'm surprised at the low amount per active premium user, my numbers are much lower overall, but the amount per active user is almost 7 times as high.

The free to paying ratio is (coincidentially?) almost exactly the same.

Well, in engineer's/scientist's terms, 7 times approaches the bar of significance, 10 times :)

I suspect their retention rates could be a lot better..

I paid for Evernote a year ago. The other day I got a mail saying my account had been reverted to a free one until I renewed. But I haven't bothered. Evernote is great but I rediscovered that the free account works just fine.. oops! If they'd notified me a week or two in advance, I'd probably have paid for year two for fear of "losing" my privileges (whatever they are).. but as it is, it still "just works."

Just pointing this out in case other startups consider dealing with renewals in the same way. If people are reminded of how awesome the free account is, there's less motivation to renew! I suspect I'll end up renewing just as a token of support though..

That seems like an excellent place to put an engineer to work improving the marketing. It is really shockingly easy to put a pipeline between the product and marketing rather than treating them as two ill-fitting bits grafted onto the Frankenstein monster that is the company. For example, your company has nigh-omniscience regarding the value they're getting from your product. Your company wants to ask them for a sale (the renewal). Combine both of those into one email: you should renew this because you saved 47 photos with us, saving you several hours and loss of precious memories like Cousin's Wedding and Study in Violet #8. Did we mention it is only $PRICE? (Obviously, handle with care, particularly for regulated industries.)

I spent $250 on MailChimp credits this morning because email marketing is printing money at the moment. Previously customers received a hints & tricks followup email 1 day after signup (if they asked for it). I tweaked that a bit so the computer watches their activity in the first day and customizes the subject line to fit their interest. That took thirty minutes of work to double open rates, which helps retention considerably and ultimately leads to more sales.

Back to Evernote for a second: they could also probably make serious money by A/B testing obvious candidates for improvement on their front page and their purchasing page. Revenue per user is not a fixed number if you start doing conversion improvements.

I find Evernote's free model oddly generous and I suspect their conversion rates could be much higher with some tweaking.

I use it every day and I don't come close to reaching my monthly limits. It's actually become one of my most used apps in recent months and I'd happily pay a monthly subscription for it - but the free plan is so generous I've never needed to.

Google's freemium services, such as GMail and Google Apps, are plenty generous, and they don't really get conversion until you buy an entire institutional version (and for universities, they even give you an institutional version for free).

Now, Google does also make money on ads, but the point remains that it really doesn't take a high conversion rate to get a freemium service to make money; as long as the cost per user is low enough, it can be beneficial to be very generous to your free users in order to attract a lot of people and get just a few more premium users.

In Google's case, though, it probably doesn't pay to waste time setting up systems to properly deal with small fry customers. And since the free users are being advertised to anyway, they don't miss out on much.

You're right though. Someone ran the numbers on Spotify (the music streaming service). They charge $15 a month for ad-free listening or you can just use it free. Based on average listener profiles, it worked out at something as small as 1-2% of the userbase needing to pay to cover all the streaming costs involved (from what I recall).

makes me think of http://sivers.org/1pct - there's a number smaller than 1%...

Heh, except Spotify is good enough that most people should be paying for it.. g Barely touch iTunes anymore and don't buy any music as it's nearly all on Spotify :-)

the problem with spotify is that storage and bandwidth costs are nothing compared what they will have to pay labels for not suing them.

It's really interesting to me that he even considers taking Moore's law into account when doing cost planning, but of course it makes sense. I also like this line:

For a really quickly growing startup that's always doing something new, you have to multiply Moore's Law by Murphy's Law; "The number of things that will go wrong will double every year." Sure servers get cheaper, but you probably bought the wrong ones anyway...

I cannot help it.

The number of things that go wrong with their text cum HTML editor does double every year (EN for OS X)!

It's really quite poorly implemented. Plain text suddenly transmogrifies into bold text with a different font, deleting a CRLF causes the current paragraph to randomly have double-spaced lines, etc.

>> 1. Make a product that a billion people will fall in love with and use for the rest of their lives.

>> 2. Make it easy for a single-digit percentage of them to pay you a few bucks a month once in a while.

>> 3. Make sure your variable costs are low enough that you can make a mountain of profit if you get #1 and #2 right.

This is great advice.

The vast majority of products (and web products) do not meet number 1. Appealing to a billion is not a prerequisite to being valuable or/and successful.

I think 1. was exaggerated. The model still works if you only appeal to a few thousand people who use your product for a while, if you can get some of them to pay you money.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact