It's not the size of the data, it's what you do with it.


With all of the discussion about big data these days, it is easy to think that every problem is a big data problem. Yes, there is a lot of data out there these days, and of course we all love a nice big data set, but you don’t always need tons of complex data to derive important insights about your business. Continuing with the theme of avoiding big data paralysis from my last blog post, when it comes to data and analytics, it isn’t the size, it’s what you are able to do with it that really matters – as long as you are doing it right, of course.

Don’t assume every analytical problem is a big data problem.

As I alluded to the in the introduction to this blog, just because you are faced with an analytical problem, doesn’t mean that you automatically have a big data problem to go with it. Remember, before there even was big data, there still was math and statistics. For predictive analytics, the more observations you have, the more confidence you can have in the results. However, remember the rule of thumb from back in your statistics classes – 30 observations is enough to have confidence in the results such that you can derive insight. Detecting patterns, or identifying drivers, even with small sets of data, can give you directional guidance, even if you are not coming up with an exact answer with 100% surety (which you never have in statistics anyway, but that’s a theory for another day).

For example, if you were trying to improve guest satisfaction scores, you could run a basic correlation analysis using a small sample of guest surveys, maybe 50 surveys drawn at random from all of those submitted in the last month, to see which detailed questions correlate most strongly with overall satisfaction. If you found out, for example, that the time to check in was highly correlated with satisfaction, and both scores are lower than you like, there is probably a need to invest in more staff, more training or better processes at the front desk. This insight, even from a small sample, will give you directional guidance as to where to invest your efforts. Very important to remember, however, that correlation does not imply causation – it does not identify the direction of the relationship. You can’t say that long check in lines are CAUSING overall satisfaction decreases. You can say that, because they are related, as check in scores improve, overall satisfaction scores should also improve.

Even plotting a series of data in a trend line instead of looking at them on a static report can result in additional insight. For example, plotting cover counts at Tuesday lunch every week would identify overall trends (increasing or decreasing counts), seasonal patterns (we tend to be slower in cold weather), or even outlier events (the large conference from last month).   This information could help to build programs to increase business and track the success of those programs after implementation.

Many analytical problems can be solved without big data storage or big data analytics.

Most hotels will have to invest in their IT infrastructure to handle big data and big analytics initiatives. Unless you work with a SaaS or cloud provider, hotel companies will need to invest in new database infrastructures and new processing methods for solving huge problems calculated against complex data sets. (I cover these kinds of investments at a high level in my blog “Big data is a big opportunity for hotels” Part 1 and Part 2.)

However, there are many analytical problems that can be solved without these “big” technologies. Before the technology innovations that facilitate the storage, access and processing of big data went mainstream, many organizations relied on sampling (working with a manageable subset of the data – like the guest survey above) simplifying assumptions, or data sets that were smaller, and therefore, were able to be processed in a reasonable amount of time. You do not have to wait for the modernization initiatives to be complete before you are able to gain any insights from your data.

For many organizations that are just beginning a journey down the analytics path, starting with big data storage and big data analytics (even if delivered in the cloud) can be a bit like killing a mosquito with a sledgehammer. Too much technology, too fast. The organization needs to grow with the data and technology, or it simply won’t get used. Excel has its limitations, of course, but if it gets people comfortable working with data and performing some basic analytics, then it’s serving a good purpose, but only if, clearly, the organization is able to take action on those insights. Even better, powerful new visualization tools are facilitating broader access to data, and some even have some light analytics like correlation or trend analysis that can help analysts derive predictive insights in a wizard driven environment. The best part is that any value derived from these smaller analyses provides justification for a larger investment down the road.

Understand the problem first – and then select the right data and analytic technologies to solve it.

This cannot be repeated enough. Analyzing data for the sake of analysis is not productive – in fact will just lead to distraction. To move the business forward, you should start by defining the problem you need to solve, not by analyzing a data set. This will keep you and the team focused and productive.

It may be obvious from the title of this section, but here is a good way to think about it:

  1. Define the goal or problem you are trying to solve (increase revenue, decrease labor cost, better guest segmentation, improve engagement from loyalty program members).
  2. Figure out what data you need to solve the problem – this is blue sky thinking
  3. Match that list to the data you actually have – this is where reality sets in. Determine whether what you have is sufficient to provide insight. Make a plan to collect the data you don’t have.
  4. Pick the analytic technique – understand whether you are looking for something descriptive (How many? What’s the average?), which you could derive using any standard reporting tool, or something more predictive (Why is this happening? What are the factors that are causing this to happen?), which might require a statistical package or application.
  5. Decide how the results will be communicated – are you building a report, displaying results or providing recommendations? This also involves understanding who the results should be communicated to, how often and when.

No one type of analytics is better than the other. Each does different things, solves different problems and requires different software and architectures.

Much like a hammer won’t fix everything that goes wrong with your house (much to my disappointment, as that is pretty much the only tool I know how to use effectively), one analytics methodology or technology architecture won’t solve all business problems.

Descriptive analytics like reporting, determining averages, or setting up alerts are based on historical snapshots. They are very useful for keeping your finger on the pulse of the business. Predictive analytics are forward looking. They will help you anticipate trends and identify opportunities. Statistical analysis helps you figure out why something is happening. Optimization tells you the best that can happen given your operating constraints.

A related series of analytics, like revenue management, where each result feeds the next step (demand modeling, then forecasting then optimization), requires a completely different technology architecture than quarterly performance reporting. The heavy-duty analytics in revenue management require an architecture that is designed for fast analytical processing, especially considering that prices need to be updated at the speed of business. The data-intensive process behind a performance report requires a data architecture that loads data fast, calculates report fields efficiently and enables the flexibility to drill-down, sub-set or explore.

You can’t get an optimal price from a historical report, and your revenue management system isn’t designed to be a full-service, business intelligence tool.

What you do with it matters. Find the “so what”

Whether you are dealing with big data or small data, as the title says, what you do with it matters. All of the best analytics in the world don’t matter at all if the consumer of the information doesn’t know what they are supposed to do with it – or worse yet, there is nothing that can be done. The single most important thing about data and analytics is the action you take based on the results you are presented with.

Even with small data, you need to carefully consider how to present the result to the end consumer so it is very clear what action needs to be taken. Is it a performance report that will be used to inform shareholders and stakeholders? Are the numbers they care most about highlighted, with enough backing to ensure they understand where they came from? Maybe you just need a single answer. Instead of providing reams of charts and graphs, just give the recommendation up front. We need to work on our check-in process. Tuesday lunches are getting busier, so we need to schedule more staff. Your analysis is the backup for this recommendation, but only if the information consumer needs backup.

In these last two posts from me, I hope I have convinced you that some data is better than no data and some analytics are better than no analytics. Even if all you have to work with is small data, you can still move the business forward. We have a tendency in this industry to get trapped by inertia, and by “the way things have always been”. You will never convince your organization to do something if you keep doing nothing waiting for something to happen.

Big data isn’t just for big business - learn more with SAS Insights.


About Author

Kelly McGuire

Comments are closed.

Back to Top