When Everything Changes, Go Back to the Facts

By Brynne Henn - May 14, 2020

A recent article from InformationWeek, Why Everyone’s Data And Analytics Strategy Just Blew Upargues that the COVID-19 pandemic is challenging the very data models companies depend on to make decisions in real time. Everything from customer behavior to market dynamics to demand curves is shifting dramatically. As a result, the months and years companies have spent investing in predictive models for decision-making have been rendered obsolete. 

To keep up, InformationWeek argues, companies need to rebuild their models quickly to adapt to new assumptions and, where possible, use external data sources to be more agile as the world continues to change. 

We disagree. 

While InformationWeek is right that the impact of the COVID-19 pandemic has drastically reduced the value of these models, the correct response is not to create new ones. 

Instead, companies should put these predictive approaches aside and focus on proactively using the data they do have – a wealth of transactions and customer behavior that’s accurate to the minute, to diagnose the actual changes their business they face today. 

Building new predictive models is not the solution 

Trusting predictions when the world is unpredictable isn’t the way forward for companies who need to make rapid, high-risk decisions. Rebuilding data models for a new set of assumptions presents a myriad of problems for responding to business changes quickly. 

There are many reasons why the rapid creation and deployment of predictive models in the enterprise is a problem, and one we’ve discussed before. First, the cost of acquiring sufficient data to tune new models will be high, and the timeline to build, test, and tune a model will be too short. Historical data is an ineffective way to map behaviors in an uncertain future. 

Second, for that cost the accuracy you could expect from a model will be too low to show ROI. A general rule of thumb is that it takes exponentially more training data to deliver incremental improvements in accuracy. And most importantly, the rate of change we expect to see from markets and customer behaviors over the next few months make “prediction” a bit of a fool’s errand. 

A better approach: Tackle today’s challenges with today’s data 

While the value of predictive models rapidly decays, it doesn’t mean analytics teams are without the data necessary to inform decisions across the business. Companies capture the rich transaction, customer, and operations data they need every day, from data sources that are accurate to the minute. 

In the past, it’s been a challenge to use this data. It’s granular, high-velocity, and very complex, which is exactly what necessitated investments in data science, feature engineering, and complex data transformations in the past. 

But with proactive diagnostic tools like Sisu, it’s now possible to use this raw, denormalized data in near-real-time to inform decisions, explain changes in metrics, and help the business make better use of the data they have. 

Stop making hard decisions based on gut assumptions  

To respond to the changes in this new market, companies need to stop looking for perfect data, stop betting on predictions, and stop making bad decisions based on gut feel. 

As businesses adapt to the pressures of COVID-19 and the uncertainty of re-opening the global economy, proactive, diagnostic analytics will overtake predictive modeling and data science as the path for successful companies.

“The most successful leaders will seize on any opportunity to use detailed facts about the current performance of their business to ensure they have a path forward to their future.” — Peter Bailis, Sisu CEO

During drastic change, businesses need to be proactive in their use of data, and they need tools that can proactively recommend the data that’s having an impact. With Sisu, we automate and augment an analysts’ workflow to reduce the manual work and time it takes to test assumptions. Instead, we stack rank results from your data, identify trends you should be focusing on, and test millions of hypotheses at once. Meaning analysts can spend more time making the answers actionable and less time digging through the data.  

When everything changes, don’t reinvent the wheel. Just get back to the facts. Reach out to our team to learn how we can help you get the facts from your data. 

Read more

Three Inconvenient Truths about the State of Enterprise ML

Last week at TieCON 2019, Sisu CEO Peter Bailis challenged conventional wisdom around enterprise ML with three principles for making ML accessible to all.

Read more

Three Data Challenges Shaping the Future of Analytics

During our recent webinar, “The Future of Analytics: Accelerating Operational Decisions,” we discussed the three primary data challenges that companies face and how these challenges are shaping the adoption of new, more modern tools and strategies.

Read more