Tableau is a good product. I use it daily. There simply is no better alternative for quick ad-hoc, enterprise-level, intuitive graph and dashboard making.
I've been at large orgs where we evaluated all the cloud services like Mixpanel and the ilk, but they are very different use cases.
Tableau + Redshift (or Vertica but Redshift is way less expensive) is the typical BI tech stack. When I go a client or a new job and ask them what they use for analysis and if they say "Good Data" or "MicroStrategy", I know my job will be a pain in the ass to get done. Its as if you're interviewing for an engineering job and they say "oh we all use NotePad here". Its night and day in terms of productivity.
I don't know what this means in terms of the stock, but all I can say is that I vouch for the product and think it has a lot of room to grow (from a BI perspective).
We used Tableau for about a year and then decided to drop them. The reason was a sudden change in licensing terms where:
1. The minimum step increase in server cores changed.
2. The cost to run ha increased.
3. The cost per core increased.
The only thing that is worse than an expensive product, is one that has unpredictable pricing.
Tableau is not scriptable. 3 places I have worked are moving away from it because of this severe limitation. You have no control over what queries it writes and you can't easily add simple scripts to make things more seamless or performant, like say a quick scripted query to populate a dropdown menu.
It's not very scriptable but that's not the use case of Tableau. Generally data analysts do not know how to program. They meet with PMs to understand concerns, interpret concerns into metrics, write some SQL, and display them in dashboards.
While Tableau performance is a thing that people debate, its a non-issue. Generally, the data analyst produces a report that is consumed by 1-10 people and if that consumption grows >10, then the organization will "productize" the report by putting engineering resources on it and build something custom with HighCharts or D3.
>Generally data analysts do not know how to program
In my experience, the ones that don't often produce some really erroneous or nonsensical results. Or even if they do get some things right, they have no ability to drill deeper.
Most big companies are actually just fine with nonsensical results. It's a bit odd, but I've realized that nobody who uses BI actually cares much about data quality or integrity.
BI in large corporations is usually done by people having no clue about the data, producing bullshit results, leading to "insights" to management (who at that point have have no clue that all the cost columns were added up, regardless of currency), leading to a final "product strategy" that gets implemented over the next years, while in all the companies still producing innovation, this is usually done by a rogue team applying common sense and ignoring that strategy all day long.
I was a SI/BI engineer for a while; let me chime in to defend my people just a little bit :)
Certainly there are lots of bullshit metrics, there is often very little desire to audit data or even do a by hand sanity check once and a while.
That being said, there were _many_ engineers who actually gave a fuck about making sure we had telemetry in actionable, meaningful, and appropriate places, and unfrotunately the gap in analytics would take place far over their heads, when the upper management would produce documents with all the beautiful charts and graphs, communicating.... absolutely nothing.
Graphs with mixed axis scaling (log vs non log) graphs with mixed units, conclusions that are a total stretch from the data that's there and ignoring obvious conclusions that don't line up with what the managers want to say.
For there to be USEFUL BI (and such a thing certainly can exist) there needs to be a "this isn't bullshit" mindset up and down the whole stack, not just in engineer land.
While what I said was actually an amalgamation of true stories I've seen myself, I was certainly exaggerating a bit :)
I think a lot of things come together for this pattern, which definitely happens quite a lot. Best predictor for this scenario is a clear divide between product and BI people, usually exacerbated by the fact that non-technical people get hired for BI.
Group bias by backend engineers who see BI as "just some point and click" which isn't really that challenging (but usually much better paid, as they are catering sales/bizdev/C-level which is always closer to the money) also doesn't help.
Also, as you say, often times the value of BI doesn't trickle back down the chain and that way, tracking is at most a second thought for application engineers when going prod.
Having built two analytics stacks myself (and seen the perspectives from BI, backend, sales and marketing alike) these are exactly the drivers we tackled first at our current company.
Our recipes against this common failure are: Marketing directly working with product engineers for their tracking requirements (with just some coaching from tracking pros) - and dual-using our analytics stack (Snowplow -> Redshift) for both operations (user segmenting, push notifications) and business intelligence.
This is usually considered a big no-no in BI circles which all tend to duplicate data to be on the safe side, but it helps immensely to make sure product and engineering are just as interested in data quality as the BI guys, as they depend on the very same data.
It's certainly not for everyone, but for us, it works really well.
i've only ever worked at small companies, and i guess to avoid things like this is why, but i can't help but wonder how common this is, and if in fact no one has ever used BI effectively. i find that hard to believe but really i have no idea, and its an interesting thought, makes you want to chuckle or shake your head or both.
Of course I wouldn't go so far as to say no one has ever used it effectively, but it seems like in the vast majority of cases, it's bullshit all the way down.
It's a surreal experience when you realize how executives are making decisions based on something that, after a moment of critical thought and common sense, clearly has no more relevance than some integers pulled out of rand(). Chuckle and shake your head is indeed about all you can do.
Just try not to visibly smirk if you're ever involved in acquisition talks. Play along.
In other words, it is for big shitty companies that have hired a bunch of useless people who can't do anything and need reports to get their bosses off their back.
Run a trace on your database and watch the queries Tableau actually runs while building your worksheets and dashboards. It's doing a lot more than you think.
I've played with Tableau, is there no scripting at all? Their previous Director of Analytic Product Management Stephen McDaniel stated that in v8 they were going to add server side Javascript - and he said this in 2012...
Tableau is a great product but with a high price stigma. I really like it, when clients asking for recommendation my first answer is Tableau. They like the product and the features but the reaction is always the same: nice, but it's too expensive for us, we'll go with PowerBI. Competing with Microsoft is a tough game.
It's not PowerBI they are competing with. It's Qlikview. Their biggest issue is that Qlikview is viral: you can use the evaluation version ("personal edition") for as long as you want, but it has restrictions. With Tableau, you get a 15 day trial. Consequently, it gets used in businesses for small data projects, gets seen to be really effective, then the business realises it's effective and buys licenses.
Tableau isn't SAP, it just isn't. If they want to really get into businesses, then they need to be sensible and give people the opportunity to use it and then get it into the businesses they work in.
For instance, I used Tableau to learn it for the 15 days they gave me, and I learned quite a lot as they have great documentation, but then after the 15 days I got no more opportunities to go through their tutorials. Partially I got busy on other things and wanted to revisit it after a few days, but I also setup SQL Server SSAS which took me a bit of time. I got a maximum of about 5 days usage, after that there's no point having it on my workstation as it's far too expensive for me to justify buying a copy.
If I could have had more time with the product, I guess I'd know how good it is so I can recommend it to the business I work at. Unfortunately, I can't without cracking their trial limit code, which I'm just not prepared to do. For now, I guess I'll be recommending Qlikview which is a known quantity and very good also, though nowhere near as intuitive.
This is definitely an issue. It's difficult to really sell a tool like this to your managers unless you can prove over time it saves you. With orgs I've been at the decision process was like
- order 1 license to test it out
- analyst gets order of magnitude work done more than co-workers
- team manager buys 10 licenses and a tableau server
The price is expensive but if I'm charging $100 - $150 / hr and it takes me twice as long to do something in PowerBI that it does in Tableau, Tableau quickly pays for itself for the desktop software. Then you can multiply that across a team of experienced BI professionals and even the product team because they can easily view workbooks and give feedback. That would easily outweigh the costs of Tableau Server.
Looking at their pricing ($500/user/year) seems pretty reasonable for such a high-impact role. If you're at a company, the question is if it saves you ~5 hours in a year.
It absolutely does. I know what life was like before Tableau (MS Excel + webquery, custom highcharts, cobbling together graphs from multiple SaaS services, custom python+R code) and I can say it probably saves 5 hours in a day or two.
That is not a adequate comparison. You aren't done after you bought the software. You need someone to set it up as week and that cost scales with the quality of the software you bought. You need to look at the cost you have once everything is said and done.
It is expensive if you have a lot of people. Paying $2k for a license may be for two or three analysts is a good deal. Compare this to a start up like Looker which costs substantially more to get your foot in the door (last I checked, correct me if I'm wrong.)
IBM has chosen to display "Analytics" as a major piece of their company. Presumably they want investors to think this could generate billions of dollars in revenue yearly. Worst case scenario for me would be IBM to buy Tableau, but clearly for those of us who use it, Tableau is one of the best tools out there for what it does. Will it be able to keep its marketshare 5 years out? Who knows.
For the record, Periscope has an incredibly poor UX. It suggests "run selected query" when you highlight a single column name... As if that's valid sql.
Periscope is in a distinct, lower tier of usability. You write SQL. Or, put differently, the innovation of Tableau was so you don't have to write SQL for most visual encodings, even non-trivial ones.
However that does limit you a bit when wanting to do deeper analysis. Looker/Periscope are tools for analysts who know and are comfortable with SQL. (Disclaimer- no affiliation, just trying to build a desktop tool that sits in the spectrum between Tableau and Jupyter Notebook)
You can still write SQL in Tableau, that's table stakes.
Funny to me: it took years for Heap to introduce that.
Edit: we use and help customers with IPython/zepellin etc, but that doesn't change that these things are 10 years behind as IDE/visual interfaces. Adding charting & drilldowns is much harder than a button to turn a table into a line graph, and analysts are wasting a lot of time due to this.
Yes, completely agreed. There's a continuum of UX between BI tools like Tableau/Qlikview and the interactive prompts used by most data scientists (mainly around the non-interactivity of display)- it's sad that an 80 character static tty display is still the state of the art in that space. While I'm very comfortable in ipython/terminal I often wish I could easily pop up a Tableau interface on top of my pandas analysis. Jupyter is getting there slowly but the widgets people are building are still more geared toward display than interacting with data (both original and derived) directly. Even the new set of SAAS BI tools (Looker, Mode, etc) leave interaction as a secondary concern. I think that's the main difference between tools geared towards reporting/publishing (most of the BI world) and tools geared toward data analysis (R, python, etc). As you point out, the tools for data analysis are years behind the BI tools in terms of UI/UX (or their focus is more on IDEs for developers rather than analysts).
As a side note, I've been following your work for a while now (both Superconductor and Graphistry) - mind if I shoot you a few questions privately?
I am stuck with SAP Business Objects (Crystal Reports) at my company. It is ok, but the features of Tableau look like they blow it out of the water. I've tried to get discovery projects started up to look at other options for due diligence. However no one seems to go for it. Oh well :/
SAP have been gutting Business Objects for some time now. I know someone who is fanatical about Business Objects Universe Designer, but they tried to use the later version and they found that a lot of features were stripped away, so she went back to the older version and refuses to upgrade until she's certain the new versions are on par with the old one.
MS SQL server is very expensive, especially thanks since their recent license changes. Microsoft BI (PowerBI, etc) is to lock you in to their ecosystem (MSSQL, Excel, SharePoint/Office365, Windows Servers/Azure cloud).
The thing is, if you already decided on MS SQL Server, why buy Tableau? The company I work for has both, and Tableau is the most likely to be dropped of the two.
Hello? The MS SQL Server license has been changed to be per CPU core. Consultants want you to create a seperate MS SQL Cluster (serveral bare metal servers with many CPU cores). So MS BI will be a lot of more expensive as you first think. Most see just the Tableau is more expensive than BI and decide for MS BI just to find out MS SQL is rather expensive. Also you need at least MS SQL Server 2014, despite dark pattern speak (lies) about that 2012 may work too. Oh and don't forget they will mention that you will need a seperate SharePoint cluster (web and app servers) to run Excel webservice and the SharePoint based dashboards. Oh and you will need to upgrade all your clients with Office 2013. And they will invite your CTO to an Windows 10 event and give away some WinPhone10 and tablets. And then you will have to figure out how to comply to local state law by upgrading to the expensive Windows Enterprise LTSB license.
At least in the case I'm describing, the company already has an expensive MS SQL Server 2014 license (Core licensing) which allows for an unlimited (ok, something like 50) amount of VMs, and they already have the expensive bare metal servers, so that's not a concern, since it's a sunk cost.
All clients are already on Office 2013, so that's not a concern either.
The Sharepoint cluster probably is a concern, but that's about the only thing, I'm pretty sure it'll be cheaper than Tableau.
I wish they'd invite the CTO to a Windows 10 events and give us some giveaways :) . No local state laws, this is for an insurance company in South America btw :)
Edit: the opinions are strictly my own, I'm not involved in any decision making sadly (no Windows 10 giveaways for me :P ), etc.
Finally someone else who has felt that pain of GoodData.. I was hired to make it work after GD's consultants set it up at my current company. What a nightmare. It took a year+ but I finally convinced everyone that it was terrible and we switched to Tableau, which let us use our own db's instead of GD's terrible "logical data model"
Less moving parts. If I get a request from a PM, I quickly open up Tableau, select a datasource or write some SQL and deploy. Sometimes data asks that they think would take 24 hours or more can be turned around in under an hour with Tableau. You can also train and hire less technical people to use Tableau...its basically like advanced Microsoft Word.
It depends on whether you want your BI to be generated by a battle tested product, or by a melange of a rickety, buggy, half baked stuff that hardly anyone is able to use without paying $10K+/yr for a support contract.
We jettisoned Pentaho about 2 years ago because of the effort required to develop and maintain it.
At the time, between the ETL and reporting layers, it acted more like a set of different open source apps simply branded together, and interop required more effort than what should have been necessary.
Debugging was a nightmare as well. Huge stack traces on simple errors made locating problems difficult, and there seemed to be little information in the community. The number of Java library layers spewing out on a simple JDBC driver error was mind-boggling. Pentaho of course has a interest in revenue from support contracts, and most inquiries into simple issues in the forums led down that path. It may have gotten better since, but there was a long way to go.
We switched to Tableau on the front end, and "old fashioned" ETL scripting in Python (now some Go) on the backend. At the same point today I would consider something like R/Shiny, but for speed of implementation Tableau would also be a contender.
Basically same here, but a few years earlier. I consider a three hundred line stack trace to be a valid reason to replace a system with something less operationally challenging. When engineers are $100+/hr, and some way more than that, it often ends up being cheaper as well.
No, although it required a little ingenuity. We would have considered using Shiny Server Pro for the auth and other features but we wanted to use our single sign on service. What we did was we put Nginx in front and had it call a tiny Rails app that handles authentication via the SSO. If Rails returns the correct status to Nginx, then the client is redirected to the Shiny page they requested.
Its a bit of a one-off but it does work well! One more step towards making R fit for production :)
Just to offer a different opinion: I know a lot of people who have been switching away from Tableau or not choosing it and using a different solution (or multiple) instead
I've been meaning to write a blog post about this actually. Maybe I'll actually do it :)
I guess it depends what you're trying to do. What part(s) of what Tableau offers do you care most about? How big of an organization?
Chart.io is one product I evaluated that met our requirements. We cared most about collaborative querying and the ability to run templatized reports (including regenerating them via the API) so we ended up going with Mode Analytics
'what it means in terms of stock' Exactly- The problems (big price drop) arise only when the stock price is at unreasonably high levels based on unreasonable growth assumptions by Wall Street. So the same people that cause it go 'bubble' then cause it to go 'burst' wreaking all kinds of collateral damage (employee stock options, morale, etc).
Wall Street isn't responsible for the bubble. It's the company leadership misrepresenting the company's road map. A stock won't become overpriced if the CEO blatantly says that sales will probably only go up 5% this year.
Telsa appears to have been overpriced for quite some time. The stock kept rising based on emotional reactions to videos of falcon doors, robot recharging arm, hypertube test tracks, and reusable space craft landing. If Elon Musk can make all that stuff happen surely he'll find a way to drive the stock up to $500? But even Musk was saying the stock was overpriced. And now it is correcting.
From experience, there's a nontrivial intersection between the BI stack and the operative administration stack.
However, in the difference, the admin stack with ELK, Riemann, Graphite, InfluxDB, Airbrake, Icinga and whatever else assumes that 30% of the information right now is worth more than 100% of the information 3 weeks later. You know, my application port is closed, first level support is getting hell, I need any available information right now.
On the other hand, the full data warehouse/tableau stack assumes that 100% of the information is more valuable than anything else. Good DWH-Guys and their analysts can do very awesome data voodoo, predictions and analysis, and the admin stack won't be able to reproduce most of that. It just takes 2 month of data collection and 2 - 3 weeks of analysis to get the result of that grand voodoo. And Tableau can automate that voodoo after it's been done once.
The ELK stack is really fancy next generation "tail -f | grep" in a web browser. Which is awesome and really important, but a very different problem from "tell me how many users have used my app for three or more days in a sliding thirty day window".
Furthermore, their support is amazing. I've emailed their support in the last couple of weeks and they've called within an hour to help resolve the issue.
Their desktop publisher or Tableau Server? Because Tableau Server is the biggest POS I have ever used. Their API is fundamentally broken, their authorization model misses large gaps, and client-integration is seemingly something they never considered.
Never heard of DevExpress but just by going to the website it feels far less intuitive than Tableau. Tableau is so freaking easy that you don't need to be experienced in BI to be user.
I've been at large orgs where we evaluated all the cloud services like Mixpanel and the ilk, but they are very different use cases.
Tableau + Redshift (or Vertica but Redshift is way less expensive) is the typical BI tech stack. When I go a client or a new job and ask them what they use for analysis and if they say "Good Data" or "MicroStrategy", I know my job will be a pain in the ass to get done. Its as if you're interviewing for an engineering job and they say "oh we all use NotePad here". Its night and day in terms of productivity.
I don't know what this means in terms of the stock, but all I can say is that I vouch for the product and think it has a lot of room to grow (from a BI perspective).