Automating Unpredictable Processes

June 29, 2017
The key to the pharmaceutical industry’s R&D efficiency

While the $2.6 billion cost of bringing a new drug to market as found in a recent Tufts Center for the Study of Drug Development (CSDD) research report is jaw-dropping, of greater concern is the fact that this figure has increased 145% since the last time the study was conducted in 2003.

Despite a significant industry focus in recent years on improving drug development efficiencies and lowering development costs, R&D efficiency in the industry remains low, and costs continue to mushroom. Why are drug R&D costs rising so rapidly? The reasons are multifaceted, but they include the increased complexity of administering clinical trials, a greater focus on developing drugs for complicated diseases, increased regulatory burdens, and many other factors.

However, there is also strong evidence demonstrating that pharmaceutical management shares much of the blame. Questionable drug development projects are too often pushed forward when perhaps they should be discontinued early in the process, too many processes across all aspects of the life sciences industry are not automated, and data is often not being transformed to allow it to serve its proper role of enabling decision-making. 


Drug development is an inherently unpredictable enterprise. Partially due to this unpredictability and partially due to compliance issues, a large portion of the pharmaceutical and biopharmaceutical industries still use fairly arcane process management strategies to manage a wide range of activities. The use of traditional paper-based approaches and the use of technology tools not fit for the purpose are still far more common than is ideal.

Planning for Inevitable Unpredictability & Needed Course Changes

In addition to unpredictability, another challenge in pharmaceutical R&D is automating uniqueness — seemingly a contradiction in terms. Drug development often consists of a series of established processes, but mid-process decision-making is always needed as the uniqueness of each project reveals itself. A great example is the design and execution of clinical trials. Every clinical trial undergoes the same series of basic steps for setup and execution. However, every trial quickly delivers a set of unique conditions that could not be definitively anticipated and characteristics that are different from other trials being conducted.

How Can Automated Decision-Making in Life Sciences Work? Let History Be Our Guide

Technology that supports automated decision-making is needed for unpredictable, complex, and rapidly moving scenarios. An example that we can all relate to is the evolution of transportation and navigation.

In the days of the horse and buggy, travel was slow so people tended not to travel very far from home. They were typically very familiar with the routes they needed to travel and when they did venture to unfamiliar areas, there were not all that many roads to choose from — remaining on course was fairly easy.

As automotive travel became more prevalent and people traveled more frequently to unfamiliar areas, more sophisticated navigational decision-making tools were needed, and the ubiquitous road map played a critical role. Road maps, despite the near-impossibility of refolding them, were extremely helpful decision-making tools. However, they could not provide directions based on changing factors: enter Mapquest.

Mapquest, and other digital tools of this nature, were revolutionary. After entering starting point and destination information, possible routes were calculated and offered for user selection. The driver selected the route that seemed to be the best, printed out turn-by-turn directions and carried these directions with him/her to help with accurately and efficiently reaching the destination. The primary drawback was that these directions could not respond to changing mid-journey conditions. While these directions were extremely helpful, when confronted with new information, changing conditions, or navigational error, their usefulness quickly dissipated.

Finally, we arrive at modern GPS navigation technology that calculates predictive routes based on recent historic information and in some cases, real-time information. These navigation technologies are robust decision-making support tools when course changes are made; users are notified immediately of navigational errors and provided with suggested course corrections. Finally, arrival time estimates and distances are continually updated in response to executed decisions.

How Does Travel Navigation Apply to Pharmaceutical R&D?

Drug development processes share common steps and stages, and the end goal for every drug is regulatory approval. Like the navigation analogy outlined above, there is a clear start and end point to pharmaceutical development, and many decisions need to be made along the drug development journey as new learning occurs, with inevitable course changes.

So how can drug development processes, with unanticipated and/or changing mid-process steps, be automated? The answers lie in first understanding the differences between flow-centric processes and decision-centric processes.

Flow-centric process designs assume that the order of actions is not variable. A workflow engine determines what happens next based on unchanging rules. Returning to our navigation analogy, a journey that never changes midway in response to traffic, road closures, error, or other factors, represents a flow-centric process. The primary advantage of flow-centric processes is that they are easy to understand and construct. However, they break down within workflows that do not progress in a serial or completely predictable fashion.

In contrast, decision-centric workflows are driven by the user and constrained by business rules that determine which activities are selectable and/or required. Based on conditions (e.g., the data captured, decisions made, the status of the workflow), the process workflow may present different options. Conditions may change in such a way that some of the planned activities are no longer appropriate, and other activities might become appropriate that were not available before.

Trying to manage a changing process using a flow-centric/procedural workflow involves building complex internal routing, leading to a spider web of possible paths in a workflow diagram. This is where workflow and process breaks down, because users cannot easily interact with the system to add intelligence, rules, and guidance.

This is a significant reason why many pharmaceutical organizations use Excel worksheets and other ad hoc tracking and process design tools. Pharmaceutical development, and many other processes within the life sciences industry, are most definitely decision-centric. But, because most process design software tools cannot handle decision-centric processes without extensive IT development costs, high levels of complexity, and highly trained users, life sciences organizations often piece together process design and tracking tools. However, this ad hoc approach does not facilitate the automation of life sciences processes or cross-team collaboration, greatly escalates the risk of errors, adds time and costs, and results in many other undesirable outcomes.


Over the course of my work within the life sciences industry, I have found that most organizations are using many unconnected applications including huge Excel files overloaded with complex macros, and other ad hoc tools to design and track the many processes inherent to life sciences R&D.

Moving to automated decision-centric process management approaches presents challenges, largely due to the lack of connection among many company systems and the lack of ability to transform data into actionable decision support.

Platform and Tools for Successful Life Sciences Process Automation

Although there are multiple approaches that could be explored, I have had success utilizing  enterprise as a platform to connect disparate systems and as a dynamic business process management tool; for example Work-Relay, which is capable of automated decision-centric workflow creation and execution. is increasingly being utilized by life sciences organizations as a platform to assemble data sources and systems in order to fuel applications that can transform this data into actionable information and processes. To support decision-centric processes, Work-Relay uses user-created flow and form engines, and rules evaluate the conditions pertaining to the entity to determine which actions are appropriate. This stands in contrast to trying to conceptualize the problem based on an overall flow, where users work to define the portions or entities within the process structure as they are working, leveraging pre-established decision-making and process rules.

Returning one final time to our navigation analogy, the combination of Salesforce and Work-Relay offers automated mid-journey decision making support as course changes are needed and decisions based on unexpected data and learnings present themselves.

Successful Life Sciences Outcomes

The life sciences functions and situations that can benefit from process automation are nearly limitless, but I’ll share what I can about two specific scenarios. Clinical trial managers often struggle with bottlenecks in trials, unexpected trial outcomes that require course changes, changing regulations, and other factors that cause delays. One life sciences organization identified the need to automate the entire trial creation and execution process, including rolling up process information to create project performance views that allow actionable visualizations of the status of a trial, real-time reporting of potential trial delays, and potential resource shortages so that management can make appropriate decisions in order to keep trials on track and finish on time.

Another recent project automated decision-centric process workflows within a biopharmaceutical company’s pharmacokinetics processes. Rules and decision-making processes were established that mined available data and provided the framework for a much more automated process. The processes were streamlined  and stronger decision-making was enabled, which is contributing to a more effective drug development process.


Statistician and risk analyst Nassim Nicholas Taleb has stated, “While in theory randomness is an intrinsic property, in practice, randomness is incomplete information.”

This concept, applied to process design, eliminates the concept of unpredictability and consigns unpredictability to its proper place — as a product of unknown or unmined information. The ability to transform the abundant data life sciences companies have at their disposal to decision-making support tools and process automation is what is needed to fulfil the potential of the life sciences industry.

About the Author

Chuck Piccirillo | CEO and Founder