Faster Delivery = Happy Users
Automated Process = Fewer Errors
Standards = Cost Reduction
Order Visibility = Confidence
Linking Systems = Efficiency
The tech world is filled with buzzwords and hype. While these phrases can be meaningful or helpful, they aren’t always.
This is true of “analytics.” It’s also true of its sibling terms “predictive analytics” and “business intelligence.”
Tools marketed as “analytics” are sometimes excellent. But not always. To tell the difference, it’s important to know what analytics is, in plain language. Otherwise, you might expect analytics to be the silver bullet that solves everything when the reality is a little more nuanced.
Just ask the companies surveyed in Duke University’s 2018 CMO Survey. It revealed that, on average, companies don’t see incredible benefits from analytics. On a seven-point scale of analytics performance, in which 1 meant ‘not at all effective,’ and 7 meant ‘highly effective,’ the average performance rating was 4.1. Despite this, the average company planned to direct 198% more of their marketing budget to analytics over the next three years.
You can probably do better than that average and for less money. But doing better requires a clear understanding of what analytics actually is – along with its powers and limitations.
“Analytics” isn’t a precisely defined term. Mostly, it’s just a fancy word for “descriptive statistics.” It doesn’t necessarily have anything to do with “analysis.” The only functional difference between analytics and old-fashioned statistics is availability. Analytics tools collect statistics way faster than the tools of the past. It’s a difference in presentation, not really a difference in kind.
Any user of analytics tools should be mindful of the classic truths about statistics. Like the fact that data varies in quality. Or that it’s easy to include irrelevant data in statistical reasoning, or miss vital factors. Analytics is no substitute for human discernment.
According to the Harvard Business Review, analytics function better with goal-oriented data collection. You shouldn’t just gather mountains of data indiscriminately and plug it into a model. Otherwise, ‘big data’ becomes ‘too big data.’ In the real world, this often occurs because data collection is driven by IT expansion, rather than marketing aspirations. People gather data because they can, from whatever fancy new analytics sources are available. As a result, valuable sources of insight are thrown in with irrelevant strings of unstructured information.
This is not to say that the speed and thoroughness offered by today’s analytics cannot improve your decision-making process. They can. But only if they’re used with sound statistical principles.
Predictive analytics take a step beyond statistics into the world of forecasting. They use tailor-made math to make predictions from presently available data. Sometimes these forecasts are created with new techniques, like machine learning. Sometimes they’re created with old-fashioned techniques like linear regression.
Here’s an example: Predictive analytics software might estimate the growing cost of producing certain widgets by looking at trends in the commodities market.
This kind of forecast can be useful. But no predictive analytics suite is a crystal ball. If someone had invented a way of accurately forecasting stock prices or other ripples in the marketplace, you’d have heard about it.
Moreover, you’ve got to make sure your models are measuring the right things. In this interview, data scientist Claudia Perlich gives a common example relating to advertising buys. According to Perlich, advertising strategies based on click-through rate often don’t account for the fact that people click on ads accidentally. As a result, if you adopt that kind of model, you might end up favoring people with eyesight issues. Or people who hand their devices over to young children who click all over the place. “You’re going to end up with something that is technically correct but doesn’t actually do what you want it to do,” she says.
Forecasts from predictive analytics, even if they’re conservative, should be treated like any other forecast. Which is to say, like guidelines, rather than truths.
As is true of “analytics”, business intelligence isn’t an entirely new thing, nor is it a precisely defined term. It tends to refer to applications that present big clumps of analytics in a single pane of glass. BI tools, like IBM Cognos Analytics, bring availability to automated data collection and automated forecasting. With BI software, you won’t have to make as many reports manually. You can just look at consistent slices of data collected from everyday transactional information.
These capabilities are impressive. If you find that your data collection procedures are cumbersome, a BI suite could be right for you. But, like analytics tools in general, BI isn’t the end-all and be-all that it’s sometimes made out to be. In fact, a flashy interface and a high level of automation could be counterproductive by leading to complacency. Just because your statistics are presented nicely, it doesn’t mean they’re the right ones.
This is backed up by research from the Aberdeen Group. In 2017, they surveyed 342 companies to figure out what separated companies that were getting the most out of BI solutions. Among many factors, one of the largest was “formal efforts to develop analytical skills in-house.” In other words, companies that did best used BI to enhance the cultivation of analytical skills, rather than as a replacement for those skills.
Again, none of this means that these aren’t excellent tools. It just means that they’re precisely that: tools, not magic wands. They shouldn’t be adopted for hype reasons.
So, how can you do better? First of all, remember the classic saying: ‘garbage in, garbage out.’ Don’t use whatever data you have lying around. Be intentional with your data collection—start with a specific goal or question in mind, and do everything you can to ensure that you go after it with specific, high-quality data. Equally, make sure that you’ve designed a model that’s measuring what you want it to, rather than just relying on a potentially misleading proxy variable like a click-through rate.
Remember, too, that analytics tools are more effective if used as part of a knowledge-gathering toolbox. There are other great tools that shouldn’t be overlooked, like direct customer engagement. By using multiple sources of information, you can paint a more complete picture than the one provided by a single pane of glass in an analytics suite.
Do this, and analytics can really sing. Ironically, the only way to fully take advantage of their immense power is to be aware of their limitations. Take the time to separate the data from the noise.
Interested in learning how AI can help you develop smarter business processes? Explore the resources in our IBM AI hub to get the full story on how practical AI is for your organization.