Hacker News Reader

As a second-time founder, I found Mixpanel to be the most expensive piece of SaaS we were using in our stack: at barely 30k users, it was already far more expensive than our whole hosting bill combined!

Then I switched solutions for a while, looking for a good combination that'd give me Facebook attribution data plus user behavior. So far, I haven't found my ideas combination and in a way, Amazon doesn't change that.

I find that solutions like segment.io are great in the way that you can install their SDK just once, and have that proxy the events to all other SDKs. For now that's what I chose to do, to avoid week-long delays for approval publishing our apps in the app store.

We effectively went through the same thing - moved off of mixpanel when it got too expensive.

We instead started throwing all of our events into Google big query. You can steam events directly into a table or proxy then through your server to add more metadata. You can even use mixpanels great open source client libraries and point them at your servers.

Initially I was worried that we would miss mixpanels built in analysis and tools but we honestly preferred querying a database with SQL better. The tricky queries are the funnels but once you wrote the query once you can just save the template. The only thing we had to build was a chrome extension that could graph the results of an arbitrary query.

Thanks for sharing! I'm not a developer, but an analyst. I know SQL fairly well and some javascript, but no coding experience apart from writing code for google analytics. How easy/cheap would it be to build a custom analytics solution like that? It's probably beyond my personal skills, but do you think it's possible for someone with my skillset to find a developer and supervise them? Can you point me to some resources or stories of it being done?

Sorry I forgot to qualify my comment in that I think its a great solution for startups who are comfortable with SQL and a little bit of coding to glue a system together.

For non technical people, Mixpanel def seems like the best product.

We recently launched mobile SDKs for iOS and Android at Snowplow (https://github.com/snowplow/snowplow), so that's probably the easier route to go than a full-blown custom build :-)

Yeah seems good and I will check it in depth. It seems not provided out of the box, but I assume I can use MS SQL/HDInsight as a backend, instead of Redshift? Maybe you should consider including Azure as an infrastructure partner. I'm a consultant in the non-valley space (rest of the world). Lots of folks like me are looking for what you've developed. Most are pretty big MS fanboys. Just saying :)

Great! Yep we're not based in the Valley either. Our first focus is on AWS and Redshift, but yes Azure is interesting too :-)

Segment.io is awesome, would highly recommend. Makes it easy not to be locked into a single platform by lowering the switching cost to almost nothing.

Attribution is a tough problem in itself, most companies either do attribution really well or user behavior really well, but not both. The best analytics setups I've seen are ones where you have a separate attribution platform that feeds data to a user behavior platform.

You’re definitely ahead of the game if you’re thinking about using a solution to prevent vendor lock in. The problem though is that running a mobile app is much more than choosing your analytics provider and streaming your data to it. That same stream needs to be used to power marketing activity such as new user acquisition and retargeting to lower acquisition costs and improve engagement.

With mParticle, data portability is a feature within an end to end data platform. So is segmentation with the ability to create any audience you can think of and send it to partners like Facebook, Twitter, Google DFP, AppNexus, ActionX, and many more. Think about seamlessly running an App Install campaign on Facebook to all users that crashed and did not return and then measure it with Kochava or MAT, all with just one client side integration.

Nearly every comment here reflects concerns around cost but cost also includes labor costs, switching and opportunity costs, as well as wasted user acquisition costs. The reality is that you don’t have to send all your data to every third party. You should be able to send a configurable sample of your data to a third party, but also very specifically include or exclude data to reduce costs and protect users privacy.

Disclosure: Co-Founder of mParticle

A word of warning on selling features, perhaps customers don't value those features as much as the seller does?

My history is that I built a real-time analytic engine for reporting project/program progress within large multi-nationals.

We had this idea of a maturity scale to guide a client along a path and to show how our features differentiated us from competitors:

Level 1) A client is recording progress against key milestones

Level 2) A client is trending data against those key milestones and comparing against past data

Level 3) A client is using drill-downs to understand how to affect the trends and deliver milestones sooner against specific metrics from L2

Level 4) A client is defining and reporting exceptions (+/- %/n from expected)

Level 5) A client is defining and using event-driven reports from those exceptions

All clients were at L2 within days of being set up on the system. Few clients ever reached L3, and none ever reached L4 or L5.

The problem was, our price tag was derived from the cost of storing, reporting on, and executing workflow around L4 and L5... we were expensive. This was fine from the perspective of the sales process because everyone wants to reach L5 and so buys into it. But in reality it turned out that once clients had mastered L2, the incremental value received from the weight of extra process required to make L3+ valueable to them just never added up.

We sold ourselves as the only company offering L5 reporting, and whilst clients were aspirational they only ever had the need for L2 or L3 reporting. We were giving them L2 reporting for the price of an L5 system.

Reading the comments here, and I personally have not used Mixpanel for a while now, I suspect that you guys are on the same path we were: Selling the complexity and the capability of a Formula 1 car, when most people just want a Volvo.

It's not even the first time I've seen this cycle repeat itself. Even when I think of companies like Vignette (15 years ago) I recall they sold caching + personalisation, the price tag being determined by the personalisation but it turned out almost every client only needed the caching. I can't think of many clients who stayed with them after that epiphany had been had. Good enough is frequently good enough.

Which reminds me of this: http://archive.wired.com/gadgets/miscellaneous/magazine/17-0...

"When Cheap and Simple Is Just Fine" - and if the orgs do not have data scientists to make sense of the advanced reports you're offering, then those orgs may likely choose to go for something cheaper and simpler that is giving them high returns at the level they're operating at.

I have wanted to use Mixpanel for a while. But your product is very very expensive for a Startup. When we're trying to be frugal, there's no way we can justify spending $150 per month on collecting data on just 5,000 users (where 98% of them will be free-users). Your product becomes interesting to a Startup only if they won't have to worry about paying for analytics till they're really successful. That means many millions of data-points. Not 25,000 or even 500,000.

The only Startups that can afford your service at this point are the ones that have only paying users.

What pricing structure do you want for your startup?

Same problem here. I recommend having a far higher limit on free users. Maybe you can also charge less if the growth rate of active users is less than x% per month. They won't have money to pay anyway without traction.

We circumvent Mixpanel's limited events early on and stayed under their free tier for as long as we could. This is what our setup looks like:

We collect millions of data points a month and save them on our servers. We do not use Mixpanel's SDK to collect these data points, but have written our own code for that. We save the raw data on our server, aggregate a bunch of similar events to 1 event and then push it to Mixpanel using their API. We heavily use event properties to send as few events to Mixpanel as possible.

This way we get most of the data into Mixpanel and can build funnels, etc. without any issue. There are some edge cases where we aggregated data is a limitation. But for those cases, we end up using the raw data that we store.

Not sure about your use case, but hope this helps.

That's precisely why we price on a per-user basis at Indicative (https://www.indicative.com) and not on a per event basis.

What you described is a lot of work to build and maintain and very brittle. We've had quite a few customers switch to our platform from Mixpanel that were doing the same thing - aggregating data or sampling.

The problem with either is that you basically blow up your user-level data (e.g. John Smith doesn't have the right activity history), you will miss potentially important trends that are statistically significant that only appear over time, and you limit the type of analysis you do by "property stuffing", particularly around cohorts and retention.

Theres not much Mixpanel can compete with anymore. You're too expensive for more than the most simple, low traffic setups and your support is terrible. I've emailed you several times after dealing with your support team and never heard back. I really don't see the point anymore.

I understand you may be frustrated but would you mind emailing me at suhail@mixpanel.com? I'd like to look into why we missed picking up your email so we can fix a bigger issue we might have. We are really serious about our customer support so this really surprises me and likely surprises my support team.

I am really sorry that we messed up here.

I'm not sure why you experienced that but on my end I've received great support from the Mixpanel team. In fact I remember one case in particular where we were aliasing our users wrong and a team member there got on a call at 1am to resolve it.

Mixpanel is one of the few pieces of software I've been excited to pay for. Hate to see them get a bad rap.

It seems like there may be certain start-ups being targeted that aren't a perfect fit for the pricing structure. If you are B2B SaaS the price is well worth it.

I never tried MixPanel due to the pricing issues due to our high traffic destination.

My startup builds products for news publications and we are quite happy with GA and using GA APIs to show the same on our dashboards. There are a couple of Bootstrap themes that do a really good job of displaying the same GA data in a dumbed-down, visually appealing layout. There are also some pretty good charting libraries which makes it easy to talk plot GA data on dashboards.

I may be wrong but it seems MixPanel is targeted toward startups who value conversions very strongly and are already spending boatloads on advertising, affiliates, etc.

I found GA is very good at the beginning also. Latter, I found that I need ip addresses info and GA is not giving that to me. I end up adding api to my backend app to log interesting backend events into the my backend json db.

I found that Per IP address analystic give me much deeper understanding of events flow. Geoip lookup also allows for me to gain inside to correlate engaged user level to the type of contents to organization name/type.

> Given that, over time it is our plan to lower prices relative to the rate that hardware and infrastructure becomes less expensive.

Could your COGS really be more than 10 - 20%? Usually margins on SaaS is 85%+ and money is spent mostly downstream on payroll for sales, marketing, support, and development.

So I'm a bit surprised to hear 'once infra costs come down we'll lower prices'. When, outside of 'have 1TB of S3 storage for free for your photos', is infra ever the driving cost?

You're right, the primary cost for us (and like most B2B companies) are the employees that serve our customers. That's about 25% for our business. However, the % of cost substantially increases to become infra for the highest volume customers.

So, in a way, you make an interesting point: it's quite difficult to reduce costs for your smallest customers since the cost of supporting them isn't generally infra related (unless you're a consumer startup?).

There are number of advantages to have lots of smaller customers, no matter how seemingly "unprofitable" on paper they are: marketing, product feedback, etc. Your mileage may vary though for each business and its timing in the market.

We are in the mobile space, but my company is not a potential customer for Mixpanel or Amazon Analytics. I haven't looked at the offerings of either company. (we developed our own simple analytics)

However, I read all three responses to your comment. One comment points to your high(er) price, one more says your price is unaffordable, the other one says the giants will "will douse the flames of your party." Imo this is a problem - not just for your company, but for every startup/small-company out there.

Amazon and Google can both afford to lose money on their analytics engine. In the old days, people may have described it as predatory pricing. However, that doesn't matter because the legal definition of "predatory pricing" is somewhat limited and consumers/customers mostly care about short-term matters like the price of the product today. Of course, MixPanel itself is a venture-funded company that can absorb some losses (so you're in a better position that what a bootstrapped company would be in). However, this just points to the challenges that small companies face in the market. Of course, I root for the little guy (and I myself started a company that has never sought funding and hasn't gone through the consulting/services path either). I also want giant companies to have a smaller market-share, so I wish you the best.

every startup/mid-size company I worked/contracted for went through the same thing: Flurry sucks -> MixPanel too expensive -> Back to Flurry or GA.

We spent a lot of time thinking about this exact problem when we started Amplitude two years ago- how can we provide premium analytics while at the same time having an affordable price point?

It's a hard technical problem but we feel like we've cracked it. Would love to hear about what you think look at what we're up to. Same sort of power as Mixpanel (more in some cases) as well as pricing that's more than 10x better: https://amplitude.com/pricing

Hi,

What I'd like improved about Mixpanel:

* lower pricing (love the free tier, and until now I had just one client went beyond it and afforded to upgrade)

* be able to send notifications on events. I got sick of the excuse that "you've actually come across a really important characteristic of Mixpanel's data structure. People & Events both apply to users' distinct IDs, and sometimes appear to be working in tandem (e.g., people profile activity feeds), but the two sections of analytics actually live in two completely distinct data stores"

* do simple enhancements / bugfixes. I've sent suggestions to support, they said Yes, great, and nothing happened. Gave up. Examples: prefill notify me test email with current user's email address; fix properties in live view which are filled with custom links -- mixpanel will open those relative to report, instead of new window.

For our startup the initial development scenario was: "add an event at everything the user does, lets think about what cares and what doesn't later". And for this mixpanel's "people" plans are perfect (until you have 25K users). But letting me 100M events is a no-brainer too (it's 4000 events/profile-month).

Relevant: Heap automatically takes this "capture every event" approach for you. Would be curious to get your thoughts: https://heapanalytics.com.

(I'm a founder at Heap.)

Looked at it.

I like this approach. Is closer to what web developers are used to: We included analytics on every page, we don't need to create reports until we find a problem or something to optimize... and we still can do forensics.

2 side notes:

– Asked who did this in our app, he told me it was really easy to add mixpanel on "every event". But anyway, he would prefer that it was automatic... and not having to do it again on every view we include.

– As said on my previous comment, for this you need it to be cheap. If not, im creating a problem instead of a solution. At this scenario, mixpanel' people looks less exprensive than heap' sessions.

A question:

Does a "pay for the metrics you use" model makes sense?

Collect everything, what you pay is some sort of multiplier of (metrics used in reports or segments per user). It would be great, at least for this scenario.

I read in the first comment (from mixpanel founder) that the hard part is not collecting data but making them work as formulas.

That makes you able to not storing the data that has not been included on any report or segment for more than a month or something like that if you want to reduce costs.

"This is far more of a competitor to Google Analytics instead of us."

Keep telling yourself that. Microsoft and Amazon will douse the flames of your party before long; they are taking their first steps now.

While your opinion is a fair and valid one, you're extremely mean person on Hacker News: https://news.ycombinator.com/threads?id=nbevans

Maybe so. The truth can't always make everyone happy.