Amazon, Google and Facebook are acknowledged leaders in using data to drive their digital companies. But it is interesting to note that they really never underwent a digital transformation. They started out as digital companies and have merely accelerated from that point forward.

What of companies that didn’t grow up digital from the beginning? Companies that have been producing real world products for many decades and now need to deliver on a digital transformation?

It is much more difficult for an established company to make the transition to digital than it is for a company that was recently born in the cradle of the internet sometime in the last decade.

This is the challenge for Monsanto, an agricultural and chemical company founded in 1901. It is the challenge for the vast majority of companies in the world today.

The Promise of Data Science is Immense

As part of my current research on the barriers to and the best practices for operationalizing data science, I recently spoke with Jim Swanson and Naveen Singla about their efforts to operationalize data science. Jim is the CIO of Monsanto and Naveen leads the data science center of excellence.

For a company like Monsanto, the promise of data science is seductive. It provides that rare promise of increased revenue for little or no increase in cost. It makes use of data that already exists and refines it to make decisions that improve the profitability of all operations.

The Promise of Data Science is Unfulfilled

But despite the hype and some amazing isolated projects, companies have struggled to make data science fully operationalized (Forbes reports survey results that the majority of enterprises saw revenue increases of 3% or less from their data science initiatives).

By ‘operationalized’ I mean that data science becomes part of the operations of the organization and becomes automated, standardized and routine. It becomes boring. But in a good way.

Consider other things that we have operationalized in the past. Vast amounts of technology and process are required to deliver the electricity flowing from the electric outlet in your home or office. Yet the end product is simple, consistent and dependable. In a perfect world, in the future, the application of data science should also be delivered with this level of simplicity and stability.

Today this is not the case. The vast power of data science must be delivered by superstar data scientists and most often is delivered as heroic one-off projects that are difficult to repeat. More worrisome is the fact that without tight controls on quality and process the decisions delivered by data science can go wrong. Sometimes in amusing ways but often in ways that have significant impact on a corporation’s top and bottom lines.

Four Challenges

Today Monsanto has one of the most successfully operationalized data science ecosystems in the corporate world. Built on a highly scalable platform their data science initiatives have recently saved some $15 million in costs while growing revenue by over $17 million.  They are realizing the full promise of data science.

But it has not always been this way. Not unlike most businesses, Monsanto had experienced many challenges in the past in recognizing the full value of their data science initiatives. While already a leader in utilizing data to improve their businesses, these challenges had been keeping them from becoming a fully digital / data-driven company. Some of these common challenges included:

1. Coveted data. Important data was tightly connected to the applications that generated the data and just as tightly guarded by the owner. For example a sales organization might have no incentive to share the data generated by its CRM system and might perceive some risk if other groups in the organization had access to their data.
2. Too long to deploy. There were long cycles to create, test and deploy predictive models. In some cases they had crossed an important tipping point where it took so long to deploy a model that they had to deploy the next model before learning from the deployment of the previous model. Sometimes the models would arrive and the business had already changed and the models were no longer needed.
3. Best practices and standard operating procedures. There was a lack of operational discipline and process across the enterprise. Individual data science projects would occur but there was little collaboration or reuse based on standard protocols.
4. Unused models. The worst thing you can do to a data scientist is to not use a model that they developed. Yet roughly half of all models that are completed, tested and ready for deployment are never made actionable due to a lack of understanding or trust by the line of business.

Disruptive, Not Incremental Change was Required 

While most organizations would try to make incremental changes to overcome these challenges Monsanto was different.  Instead of incremental change they made bold, disruptive moves that challenged the status quo and brought individual data fiefdoms inline to benefit the overall enterprise. There were four overall best practices that guided them in their digital transformation.

Best Practice #1: Plan for Scale

One reason that Monsanto has been so successful is that they planned for a fully scalable solution and focused on operations and modernization. They recognized that small incremental steps were not sufficient when much more was needed and they built a strong centralized organization to carry out that plan.

Best Practice #2: Focus on Building a Platform

One important trait of those who have successfully operationalized their data science is their focus on building a ‘thing’ not just writing down best practices and policies. Their thinking is that of a software company that is building a data science factory that regularly churns out high quality and valuable product. In this case these ‘products’ are predictive models and prescriptive insights.

Best Practice #3: Decouple the data from the application  

Culturally recognize that the data created by line of business applications are an enterprise-wide asset that needs to be shared. No longer will the data be viewed as being owned by the application that generates it. In some ways, the functioning of an application may be viewed as subordinate to the data that it produces. For example, you could imagine in the extreme viewpoint that the purpose of your CRM system was not to support your salespeople and customers but to generate sales and customer data that could benefit the enterprise. This viewpoint will help to create a marketplace for data that is open and collaborative. Tools and best practices that encourage collaboration in building and maintaining data dictionaries and master data management will be critical.

Best Practice #4: Make data science a priority

To make disruptive changes you will need to change the thinking and the culture not just at the senior management level but at intermediate levels below the C-suite. Data science is inherently more complex than most other technologies. It also has the paradoxical property of recommending decisions that are not intuitive (if they were intuitive you would have been doing them already). To build organizational buy in consider bringing in outside practitioners from other industries and letting them tell their stories directly to your management. Or hold mini-conferences within your organization where management can get hands-on with the tools and results. It will certainly be a process that requires significant patience. Once you have built critical mass however you will be able to move forward without 100% buy-in as the laggards will soon recognize that they need to get on board or be left behind.

Four Years from Now

There are continuing challenges for Monsanto even though they are among the very few who have succeeded in operationalizing data science. Many of the challenges, such as data fiefdoms, are continuous and need constant vigilance to combat. Other battles won, such as collaboration on data dictionaries and master data management can be held in place by best practices that are adhered to and enforced by a strong central organization.

As other companies make the transition to operationalized data science it may not be that obvious from the outside looking in. But you will recognize their success by their ... success.

In the future, we expect to see certain companies mysteriously overtake their peers in market share. They won’t necessarily exhibit any external signs of change from technological breakthrough, new innovative marketing initiatives or new product launches.

These will be the companies that have operationalized data science and have improved all of their business processes. They will be the companies that are leveraging their own data to predict and plan. They can see into the future just a little bit better than their competitors. And that little bit will make all the difference in their success.

Expert Insights

Many thanks to Jim Swanson and Naveen Singla for speaking to me and sharing their expert insights on this topic. Jim is the CIO and Naveen is the Data Science Center of Excellence Lead at Monsanto.

Further Reading:

• Operationalizing Data Science: 13 Challenges 
• Data Science is Plutonium: Powerful, Dangerous and Handle With Care 
• The Demise of the Data Warehouse
• Ok, I Was Wrong, MDM is Broken Too: Insular, Dictatorial MDM Doesn’t Work
• IT Activation Energy: Cloud Considerations in Retail and Insurance 
• Cloud Data Warehousing: Producing the Infrastructureless Culture
• 17 Things a CDO Should Know: A Report from MIT’s CDO-IQ Symposium  
• Deep Learning, AI and Privacy 
• Enterprise-Grade Data Science
• For Best Predictive Analytics Results: Don’t Move the Data