Resources

When Being Right Isn't the Right Move

August 10, 2020

Noah Horton

CEO

Most data analysts and data scientists have built their career on precision. In “normal times,” the goal is to identify historical patterns within the business and translate them into predictive models that will be measurably accurate in the future. Spending a significant amount of time and resources to test and analyze these models makes sense in this context.

After all, an analytics function has traditionally built currency on being right –not being fast.

These are anything but normal times, though. A trio of black swans has hit us: a pandemic, social unrest, and the highest unemployment rates since The Great Depression. When so much seismic uncertainty is introduced, predicting the future is an unlikely proposition.

When a black swan devours your models, a slow, methodical approach to analysis won’t suffice for building relevancy. Instead, analysts need to focus on building a process with their business counterparts that delivers preliminary insights quickly. While this may be a departure from the status quo, it’s essential to establishing confidence and reliability with the business units that are starved for insight in a COVID-ravaged environment.

How We Used to Operate

When your primary operating value is “being right,” speed to value rarely emerges as a top priority.

Prior to the pandemic, most enterprise organizations performed data analysis through a laborious and manual process. The details might fluctuate from organization to organization, but the general outline looked like this:

1.    Form a hypothesis about what data might be needed

2.    Join and structure a portion of available data

3.    Generate features from the dataset

4.    Hypothesize about which insights are most relevant

5.    Test the hypothesis using BI or ML

6.    Iterate on the above until you have some good findings

7.    Report back the findings

8.    Interpret into human terms

Even in the best of times, this process is rife with issues. Forming a hypothesis before analyzing the data introduces bias while necessarily limiting the amount of data that will actually be examined. But after COVID, one problem became even more acute: The process requires too much time and too much (now irrelevant) historical data to add meaningful value at a moment where nearly all businesses face an existential crisis.

How COVID-19 Broke That

When COVID-19 first broke, we didn’t know whether the virus would keep us quarantined for weeks or months. Very few people predicted many states would still be in various forms of lockdown in August with the first wave rebounding into a  climb that still doesn’t appear to have peaked.

Businesses reacted in many ways, but the commonality is that every one of them has had huge changes. Many businesses, seeing a sharp drop in sales, moved to cut budgets and preserve cash. Of course, the ripple effects of stalled investments and layoffs moved throughout the economy.

There were obvious losses, but also clear gains. As remote work went from a rising trend to the “new normal,”technologies like Zoom and Slack saw a spike in sales. Home entertainment spending climbed. So did stockpiling. Businesses large and small pivoted to producing personal protective equipment.  

Even these growth areas created havoc for businesses as supply chains shook and buying behaviors morphed. Every business has been changed by COVID.

One might argue that past pandemics, like the1918 flu pandemic, would be useful for predicting food and cleaning supply hoarding. But there have been many drastic, unexpected behavioral changes over the last few months.

Predictive modeling capabilities are broken because our historical data doesn’t generate usable models in this changing world. When we don’t have experience, we can’t come up with the right hypotheses to test. Plus, the “event” isn’t over – we’re in the middle of it with new effects emerging daily. We’re not going to be able to gather enough“new world” data anytime soon to build predictive models.

An example of this challenge is highlighted by the airline industry, which despite being armed with mountains of historical and real-time data, is struggling to find the right pricing models that will lead to sales today and post-COVID.

In this situation, building a predictive model for the next year will likely fail in the next three months. We’re not going to get to “right,” so we have to focus on being fast.

A New System for Today

Our analytics focus today should center on generating insights that are quick and iterative. In a time of profound and ceaseless change, multi-month or even multi-week analysis is more likely to frustrate than inform.

Our conversations with business leaders must be oriented around what’s happening today.

The tradeoff is that if we focus on what’s happening now, we can’t over obsess with being right. Our confidence in the individual insights might not be as high as it has in the past. That’s okay. It’s a discomfort, yes. But it’s a discomfort we need to embrace in this environment.

Instead of gathering requirements and then building models in a silo, analysts must move fast and embrace an iterative approach. Use the data you have to find unusual spikes or dips, bring those patterns to the business, and work together to find where you need to focus next.

Realigning with Executives

Chances are, executives are demanding predictive models from you to help them traverse our new world. Expectations must be realigned.

Analytics organizations can’t spend time on investigating an oddity’s origin. They may not even have the time to identify patterns mirroring one another in different areas of the business. Perhaps some of the findings won’t be actionable, and maybe you’ll hit data quality issues.These will result in discomfort but going to the business every couple of days with new insights will help everyone discover patterns faster.

Business partners will need to take a more active role in their interactions with the analytics organization. Analysts will need business partners to inform them of what is obvious, what needs more attention, and what shouldn’t be a focus due to lack of control. Business leaders will have a better sense of what might be a secondary effect from something else, what isn’t actionable, and where you should be digging deeper.

This change in mentality will result in your time being used much more effectively. You’ll provide incremental and continuous value. Using this new approach, the business knows a little more about what’s going on over a very short time frame rather than staring with a historical trend and wondering why the numbers are different today.

A Real-World Example

Let’s say your company has a question about what’s happening with sales. Historical forecasts are not going to be of much use. Instead, look at what’s happened over the last week or a couple of days. Analyze customer attributes, product characteristics, where the sales are taking place, and how the sales are taking place (i.e. online direct vs. online third party vs. brick and mortar). Ideally, analyze combinations of all of these things.

The insights into the where, what and who of sales are incredibly valuable when delivered in a regular and constant cadence. It allows business leaders to access faster analysis into what’s occurring in their own organization, and to allow their context to inform and refine not only further analysis but taking quick business action.

Discovering the New Normal

Everything I’ve just described here involves behavioral change. And changing behaviors isn’t easy. But it is worth the pain of change.

If you can create a cadence where you’re meeting with the business every couple of days to give them a variety of insights, you’ll have more impact on the business in the short term. This tactic is essential while COVID is changing behavior day-by-day, city-by-city. It’s also going to help you transition to the period when people are attempting to realize a new normal, and it’s going to help your business adapt to the new normal once COVID is under control.

With iterative analysis, you’re going to be able to identify what’s happening fast. Once we get to the new normal, you and your business partners are going to be able to understand what the new normal is faster than your competition—which means capturing more marketshare.

Right now, COVID is driving rapid changes to how our customers and colleagues go about daily life. That change is iterative. Only an iterative approach to data analysis can help identify the emerging patterns.

Don’t focus on being right. Focus on being fast.

Careers

Ready for your work to have real impact? We’re hiring people that want to build the future of AI.
See open positions

Resources

Learn more about how Unsupervised works and how it helps you achieve your business goals.
Learn more

Contact

Interested in getting in touch with us? Send us a short message.
Get in touch