How to Select Your Analytics Software
Selecting the Best Analytics Software at the Right Price – What to Consider In our last Useful Business Analytics Summit Q&A post I asked our client side data scientists about unstructured data and text analytics.
Today I ask them their thoughts on how they select analytics software whether for structured or unstructured data, and how they think about costs.
I’ve written before about the idea of “free software”, and I know that in the area of social monitoring at least it seems to be a topic up for debate… But in my experience, if you need highly specific software, that does what you need it to do efficiently without lots of set up and work, then there really is no other option than professional/paid software. I realize not everyone including some of our speakers may agree though.
As someone whose firm develops text analytics software, and who has previously evaluated and purchased all manner of analytics software for both structured and unstructured data in the past I can’t help but provide my own 3 tips on how to select software today. If at all possible always:
- Ask for a demo of the software (under mutual NDA) with your own data. Don’t let them give you a canned data with some fake data.
- Ask difficult questions about the value of the software. If your sales person can’t give very specific examples about how the software is successfully used by the company or their clients, then chances are you are evaluating software developed by developers only, and not by analysts. Chances are high that if they can’t tell you how valuable it is, there isn’t much value in it. You must learn that it can be used successfully!
- Don't forget to think about ease of use. While there is no such thing as magical analytics software in which you upload data and dashboards automatically tell you everything you need to know (if someone tells you this – run!), the software should however be intuitive and user friendly enough for a junior or mid-level analyst to easily learn how to use it. Anything that requires lots of customization and learning will not be used very much.
I think our speakers bring up some good points below, and I’m very curious which of their points you agree or disagree with?
Q. IN YOUR OPINION, WHAT IS THE BEST WAY TO SELECT A BUSINESS ANALYTICS (STRUCTURED OR UNSTRUCTURED DATA) SOFTWARE SOLUTION?
I’d suggest starting by talking to others in your industry – what are they using? Are they happy with the solution? What are the shortfalls? What are the reasons they selected this package? From there, you should be able to identify a shortlist of relevant candidates. Pressure test them all – take a set of data and ask to be able to “play with” the solution.
It depends on your maturity. If your maturity is low (1 or 2), using one or more light-weight tools for 6-12 months is a good bet: You'll get the capability that is not overweight to your maturity at low cost and you'll learn a lot, which will make your eventual move to a sophisticated tool less risky. If your maturity is low and you buy, you'll probably let the tool drive your analytics rather than vice-versa.
I have never done this, but Sara Shikhman of Rent-a-Gent had a brilliant idea for solutions research generally, and that is to 1) get a bunch of interns from her business school, then 2) get 30- to 60-day free trial periods from the solution vendors who are eager to speak with her on the exhibition floor. She puts one intern in charge of learning each solution as it applies to her business, then like the Joker in the Dark Knight movie, she basically breaks the pool cue and lets the best solution win.
One should have an analytics strategy by understanding the difference between descriptive, predictive and perspective analytics and whether you are looking for an analytics and modeling tool development and/or governance and management.
If you work up to it by building the lower layers of your data capabilities stack first ...like identifying the important information, validating your ability to accurately get the important data, making the important data easily accessible and able to be rolled up to any level of aggregation needed etc., then it makes it much easier to identify the POC's needed to test candidate software solutions.
1. How relevant it is to solve the problem you are faced with - a generic solution usually doesn't work for advanced stuff.
2. Interoperability and ease of use (user friendliness) is key.
the costs of improper decision making you will make
2. For low impact improper decisions, choose a tool that favors visualizations and visual exploratory analysis
3. For high impact improper decisions, invest in human expertise first and involve them in the decision regarding software.
Q. THESE DAYS THERE SEEMS TO BE SO MANY ANALYTIC SOFTWARE SOLUTIONS, FROM EXTREMELY EXPENSIVE SIX OR SEVEN FIGURE "ENTERPRISE LEVEL SOLUTIONS", TO COMPLETELY FREE AND 'OPEN SOURCE' SOLUTIONS, AS WELL AS EVERYTHING IN-BETWEEN. IN YOUR OPINION, WHEN IF EVER IS FREE GOOD ENOUGH, AND HOW SHOULD ONE THINK ABOUT THE VARIED COSTS WHEN SELECTING A SOLUTION?
This is a great question – often an enterprise level solution may seem like the only option, especially if you work in a large corporation, especially if you don’t do your homework. You have to honestly ask yourself what you will be using the package for and then, in my opinion, start from the cheaper and more basic solutions and work your way up. Try to make the cheapest solution pass your evaluation, but if it doesn’t, eliminate and move one up the food-chain. It’s hard to do and requires discipline – I mean honestly, most of us don’t shop like this, we go for the flashiest car or shoes!
You have to think about analytic software solutions in terms of the value of the decisions it will help you make. If reducing shopping cart abandonment by 10% nets you $1K/week, Google Analytics is going to be your site monitoring tool. With regard to Big Data slicing, there seems to be a changing of the guard as the established stats PhDs have a strong SAS preference while the kids coming out of school mostly know R. Given the license costs and the fact that the Revolution Analytics version of R now enables multi-million-row processing (which had been a big SAS advantage), I would stick with R until you have to go to SAS. That rule is probably generalizable: stick with the cheap solution until you find yourself straining against its limitations.
Free is always good, analytics is no different, when you are starting. Once analytics production becomes integrated into the company operational and/or decision making process and/or formal management / governance processes then the investment is inevitable but should be scaled to enterprise (company) that you are in (i.e. size of the balance sheet, size of portfolio, equity, size and complexity of risk, etc.)
In the absence of a fatal flaw associated with the free product (e.g., lack of support), if the free product performs well in the POC phase (see question to answer above) AND the free product does not close future options, then consider it. Keep in mind that the software license cost is only part of the total cost associated with adopting an analytic software solution.
Small and middle size companies can sometimes get away with free solutions. As the business grows and company needs more specific and detailed Information it will have to pay for more robust tools. Cost of new tools should be justified by incremental business that can be potentially brought with additional business intelligence. (ROI)
Free is almost always good enough if you have a highly qualified / talented team that knows how to get the best out of it. If you need a lot of hand holding and support then definitely use a commercial solution.
When we look at the landscape of powerful, flexible, well proven and even scalable analytics software solutions, there are still very few serious players out there. Some that cost hundreds of thousands of dollars, some that are free and open source, and some in-between. Long gone are the times when free is synonym of cheap. I started using R in 1999 when a professor gave it to me, hand written on a CD, who in turned received from colleagues in New Zealand. I could not afford S-PLUS to do my homework at home. He told me it was a clone. My immediate thought was that he was providing me with pirated software!
Fast forward to today when R is integrated in almost all analytics and database solutions, from SAS to Statistica, from SPSS to SAP, from IBM to Oracle. Yet, R has a steep learning curve. It's definitely not for everyone. We are starting to see solutions nowadays that attempt to fully incorporate R but in a way that is more user-friendly, at least for the simple things. How successful these efforts will be is too early to tell, in part because of R's licensing model. Other solutions leverage a different programming language, Python.
The question is not so much one of free vs. paid; the crucial question is how serious a company is about analytics. A small team of data scientists and analytics professionals usually has no problem leveraging free software to their full potentials. Support here is seldom needed. But if a company is testing the waters or has very simple analytics needs or wants to increase the analytical mindset of its people they may want to look for solutions that are easy to learn, favor visual explorations but that at the same time promote good analytics practices.
Thanks again to our speakers. Check back again for our next two posts which will give tips for consultants and software vendors wanting to do business with our client speakers as well as what recommendations they give in how to best communicate findings to the C-Suite.
[Full Disclosure: Tom H. C. Anderson is Managing Partner of Anderson Analytics, developers of patented Next Generation Text Analytics™ software platform OdinText. For more information and to inquire about software licensing visit ODINTEXT INFO REQUEST]
[Above also posted on the Next Gen Market Research blog]