Big Data: Bad Data Doesn’t Deliver Good Answers

Bad data hides everywhere and making decisions based on this data can lead to serious consequences. In the era of big data, the effects of bad data are amplified significantly. Forbes argues that these well-meaning efforts to “Moneyball” every aspect of business, from hiring to operations, place an increased emphasis on data quality. At the same time, maintaining data quality at a time when massive amounts of data come into your business at the speed of light is challenging. But where does this type of data come from? It sneaks in through a variety of sources, including internal data from transactions, email, and log data like Google Analytics and IoT (internet of things), as well as external sources from social media, partners, and customers. But all is not lost. There are ways to handle bad data and to keep it from entering your system in the first place. So, read on to learn how to handle the data monster.

What is bad data?

Bad data can creep into your databases from any number of sources, which we’ll discuss next. Gartner estimates the cost of bad data to average $15 million per year (per company, not per region), and 60% of those surveyed have no idea how much bad data exists in their systems. Even worse, a 2016 study by IBM estimated that bad data costs the US economy $3.1 trillion per year. But here are some examples of data we classify as “bad”.

• Missing data points, which might throw off calculations if you’re missing a significant number or lead to transposing numbers
• Duplicate records that misrepresent actual values
• Inconsistent formats resulting in flawed analysis, i.e., birthdates displayed with both two and four-year formats
•  Incorrect data types, such as numbers stored as text
• Uncorrected data, such as outliers that have an oversized effect on the analysis
• Mismatched data, such as misspelled words

Some data is just dirty and needs cleaning. Maybe some missing values shifted all your columns off — just like when you were in school and messed up the responses on your scantron. Sometimes, weird values got inserted. Maybe your key was off when you merged two databases. The bigger the data, the more likely it is that you have some dirt in there.

Whatever the reason, you need to clean your data periodically, and there are good tools available to help with this.

A bigger, more subtle problem might exist in your data — it’s not representative. Yet, you make broad assumptions based on this skewed data. For instance, Facebook comments over-represent young, affluent, outgoing folks, which don’t represent the total population and might not represent your target audience. The discrepancy between predictions and results from the US presidential election is a good example of how your predictions are wrong when you assume a biased sample represents the whole.

There are plenty of errors even with credible data sources

That’s right. You heard me.

In addition to being biased, your data might be wrong — full of errors, duplicate counting, and just plain wrong!

In a recent PPC campaign for a client, I discovered just how inaccurate data is—even data from Google, which makes its living by providing accurate data. As a data-driven agency, we monitor analytics daily, sometimes hourly. I noticed something funny (or not so funny) in my data: the PPC campaign charged my client for a higher number of clicks than were recorded by Google Analytics (combined with the Google Search Consol).

Which number was accurate?

I never got a satisfactory response from the Google Ads Team.

I noticed a similar problem with Sprout Social. I routinely unfollow Twitter accounts (now X) that remain silent—why follow someone who never Tweets? I saw a friend of mine’s account appeared among my silent accounts, but I thought that didn’t seem right. So, I switched over to Twitter and found Tweets as recently as one hour ago. Now, I’m afraid to delete supposedly silent accounts.

Which leaves me pondering:

• What other data is inaccurate?
• How big is the difference between my digital analytics and reality?
• Are my decisions based on accurate data or just some fiction?
• Is there some way to fix data problems or even understand the extent of these problems?

Where does bad data come from?

Bad data comes from a variety of sources, which we’ll discuss below.

Human errors

People make a variety of errors when entering data, sometimes by accident, sometimes through inattention, and sometimes through outright fraud.

• They might miskey data because they read it wrong, a finger slipped onto another computer key, or they transposed numbers. Inconsistent data input forms increase the chances of these miskeys.
• Duplicate data is difficult to detect manually unless the two pieces of data are in close proximity on the input form.
• Inconsistent standards or misunderstanding directions between employees in different departments or shifts.

System issues

• Legacy systems might use outdated software to store and process data, which can introduce errors or inconsistencies.
• Migrating data from one system to another might introduce errors due to missing data or corrupted records.
• Integrating systems might generate problems with conflicting values or other inconsistencies.
• A lack of standardization in the way data is collected, stored, and managed can lead to bad data.

External factors

Often, systems must interface across organizational units, geographical regions, and companies. This makes your system vulnerable to mistakes by external entities. User-generated data is especially problematic as it might be inaccurate or manipulated. Web scraping from external sources also results in errors when the website data is flawed or the API used to extract the data isn’t good.

Cybersecurity and fraud

You might face issues of intentional manipulation, especially when including data from devices such as IoT, where hackers can introduce data for malicious purposes. Data breaches and fraudulent data, such as from fake accounts, can also introduce data problems that might impact your decisions.

Data analysis problems

Before you can make data-driven decisions, most data must be analyzed, which introduces another opportunity for mistakes. Below, we discuss the top problems that can crop up in your data analysis.

Problem #1. Response bias

Sure, we use that term more when we discuss survey responses, but I think we can use the same term to represent the problem in digital analytics.

Response bias (or bias of any type) refers to the problem you encounter when your data is systematically different from your population of interest. Using data from your website or email list results in response bias because data comes from a select group that doesn’t encompass your entire target audience, only those who visited your website or subscribed to your email list.

In a client project, I built an algorithm to score email list subscribers. I discovered that a very significant number of subscribers weren’t even IN the company’s target audience but were competitors, students, and others who weren’t valuable to the organization. Without this algorithm, the company wasted resources trying to convert folks who weren’t even prospects and likely made content and other types of decisions based on feedback from readers who didn’t matter to the company’s bottom line.

There’s also a fundamental bias when it comes to posts on social media, where most of the conversation comes from a small number of users who don’t represent the consumer population at large. That problem happens with your website, where visits from different devices by the same individual further obscure the customer journey by fragmenting it across various devices.

Problem #2: Correlation isn’t causation

By extension, it bears repeating that correlation isn’t causation — no matter how significant the coefficient of correlation is. But, big data offers the tantalizing option of seeing correlations and using them to inform decisions.

As an example, let’s say you pick up a series of clicks about my search for a new car. You pick up similar signals about me using cookies or other data sources that indicate my lifestyle and predict income. You then start sharing information, encouraging me to buy your high-end car.

But, whoops. You guessed wrong.

I don’t have the kind of income necessary and I’m not a prospect for your expensive car. Instead, I lent my computer to a friend visiting from out of town. She was the one doing the searches that threw off the analysis. We aren’t discrete, nor is our behavior reflected in each of us as an individual—more on this in the next section.

Now, you’ve wasted resources and may even damage a potential relationship with me in the future. This happened to a friend of mine just last week.

Problem #3: Behavior isn’t understanding

Watching what people do, even in digital and mobile space, isn’t the same as knowing them or explaining why the customer journey evolved the way it did.

• I may let someone use my mobile device to find information.
• I may research a product for someone else or as a gift — something I have no interest in.
• Maybe I’m researching a term paper, job prospect, or blog post and have no interest in the results personally.

Using behavior to infer WHY I chose a particular path along my customer journey is dangerous, but when combined with big data about other behaviors, the danger is compounded.

Even when we ask consumers why they do something, we might get inaccurate data, but certainly, inferring attitudes based on behaviors is wildly incorrect.

For instance, my daughter is getting married this summer. I’ve been bombarded with emails for wedding albums, honeymoon trips, and other related wedding paraphernalia because I attended a wedding expo with her.

But, I’m not getting married- I’ve been there, done that, and have the bruises.

I’m not making the decisions. In fact, I have little time or expertise in such matters, so she’s getting other folks to help her put the wedding together. All a company does by sending me this deluge of unsolicited emails is incur my wrath — especially when I ask multiple times to unsubscribe from their list.

Bad data, bad analysis, bad decisions

If you want to make data-driven decisions (and you really should), bad data and bad analysis lead to bad decisions. A good example is the situation Coke found itself in a few years ago. Based on a biased survey, the company changed the formula for Coke to make it sweeter. Consumers universally rejected the new formula, causing a PR nightmare that caused the company to revert to the original formula.

When you use inaccurate insights to make predictions regarding essential business operations, such as how much raw material you need to produce forecasted sales, you waste money to build a surplus or run out of material when demand is higher than predicted. This wastes resources and efforts, as well as potentially leading to poor customer relationships.

Another big problem you can encounter when you face inaccurate insights is meeting compliance or legal requirements. For instance, a firm using AI to predict the best new hires might face a problem when the algorithm is based on prior biased hiring practices. Because you used historical data that was biased toward white men, for instance, you perpetuate these inequalities and might face legal action.

How to prevent poor insights that drive poor decisions

Automating data entry goes a long way toward building better insights that drive better decisions by reducing the human factor. That’s why systems such as EDI, which automates ordering and payment between supply chain partners, results in efficiencies. In cases where you must use employees to input data, educate them on the importance of accuracy and ensure they have sufficient breaks to reduce mistakes.

Another data management best practice is to only gather data from reliable sources and then cross-check critical information to ensure it was appropriately collected. Monitor and clean your data continuously to ensure its accuracy.

Finally, policies for data collection, storage, and use should be implemented to reduce errors and keep data safe from unauthorized use.

Conclusion

Data should drive every decision you make for your business. However, bad data results in poor decisions that can cripple your growth or even sink your business.

Need marketing help to support business growth?

We welcome the opportunity to show you how we can make your marketing SIZZLE with our data-driven, results-oriented marketing strategies.  Sign up for our FREE newsletter, get our FREE guide to creating an awesome website, or contact us for more information on hiring us.

Hausman and Associates, the publisher of MKT Maven, is a full-service marketing agency operating at the intersection of marketing and digital media. Check out our full range of services.