Mastering the art of analytics: a Groundhog Day story of the exploration and production sector

Being data-driven is the new war cry of the E&P sector these days. Feverish frenzy around Digital and highly publicized business successes of the data-driven firms from the other sectors are driving most of the E&P firms’ fascination with Big Data, analytics or IOT these days. E&P leaders or senior executives, on their part, are also coming out in support of wider adoption of analytics and need to do a lot more with their data.

But beyond all the intent and excitement, not much has changed to suggest that E&P sector is on its path to becoming data-driven. Here is why I think so and most of it is based on my last 12 years of observation of the sector analytics initiatives.

Decoding data-driven

While there is no universal definition of what being data-driven truly means, various publications and experts’ commentaries on the subject identify three main acts that define data-driven firms.

1: They manage their data well

2: They leverage teamwork and wide variety of data sources to generate insights

3: They approach analytics objectively, acknowledging and addressing the human biases

Let’s look at these in the context of E&P sector.

First Act

Act 1 needs no introduction having been a proverbial thorn in the E&P firms’ flesh and deservedly received significant attention, if not action, for last as many years as one can remember. Data gurus have warned us—and rightly so—enough about the perils of poor data quality in anything even remotely to do with analytics. So any further discussion on this will be preaching to the choir.

Second Act

In the world of big data, Variety, and not Volume, is what drives the business value of the analytics, according to the experts. There is a reason why it’s the case. New correlations from analyzing different data sources (both structured and unstructured) can create new knowledge of performance drivers—a key to insight generation. While Variety is gaining greater currency in the consumer-facing sectors such as retail, it’s not a common practice in the E&P sector.

I can think of two reasons for that.

First, data sources being analyzed could belong to multiple departments which, in many E&P firms, don’t naturally share data or collaborate with each other. This limits the opportunity to develop broader perspective into a business problem, necessary to explore a range of the performance drivers. For example, through collaboration with the technical teams, supply chain teams can develop better understanding of how different plant configurations, field characteristics or operational activity patterns influence the material demand, as opposed to only analyzing their own datasets.

Second is the unknown or poorly understood causalities in new correlations that are bound to show up when analytics involves multiple data sources. In a science-driven E&P sector, engineers are predisposed to look for causalities in the correlations in order to trust the analytics. So any correlations which engineers can’t explain with their domain knowledge are likely to be dismissed.

For instance, in equipment failure prediction analytics, operations teams feel comfortable exploring the correlations between the past failures and equipment performance data because causality is well established there.  But when it comes to exploring the correlations between equipment failures and, let’s say, weather cycles or workforce demographic patterns, causalities may be less obvious to the engineers. Such “unusual” correlations require deeper analysis and experimentation to verify the causalities and many E&P managers may not have the appetite, analytics acumen or resources (read data scientists) for such experimentations.

Third Act

Which brings us to the third act: navigating the human biases in the analytics.

We all have been told to mind the dangers of “garbage in, garbage out” with computers, but “bias in, bias out” could soon replace that advise in the age of Big Data.

Experts argue that knowledge gaps and individual incentives can create hidden biases in both the collection and analysis of the data compromising the analytics results. They also recommend addressing these biases through experimentation (ref to Act 2), research and training.

In E&P sector, given the typical data uncertainties and intuitive decision making styles, biases can occur naturally and are well documented in the SPE literature. Left unchecked, they can limit the adoption or quality of the analytics efforts especially when the latter are misaligned with the managerial incentives.

For example, in the E&P capital projects, pro-project sanction bias of the engineering teams is described as one of the main reasons for overly optimistic production forecasts and poor concept selection. In this scenario, any attempt to bring analytics to improve the accuracy of project evaluation is likely to meet resistance if it reduces the chances of project getting approved and in turn career advancement of the individuals.

Managerial incentives also rub off on their teams in the way they approach the analytics. Technical teams incentivized on oil gains from well intervention opportunities, may focus their analytics efforts on identifying expensive drilling and workovers targets than exploring cheaper production optimization alternatives on the surface. Operations teams compensated for meeting the production targets may sidestep the competing advice from equipment predictive model if it incurs production loss. The unfortunate BP Macondo incident is an example of this behavior where the rig staff, under pressure to make up for the lost drilling time, reportedly misinterpreted the negative pressure test.

Groundhog Day Story?

Many would agree that mastering these “softer” aspects of analytics is as critical to being data-driven as harder aspects such as skills or technology. However, in my experience, “soft” is seldom acknowledged, much less addressed, in the analytics initiatives in the E&P sector. Many E&P firms still tend to approach analytics as a tool to solve tactical—and often one-off—problems than a philosophy of doing things. This mind-set has led to analytics projects being IT group or departmental endeavors than leadership-driven initiatives.

It’s not hard to see why so many sector analytics pilots after promising starts have failed to deliver. Without hands-on and committed leadership involvement, initial momentum from successful, if at all, pilots is lost as soon as the projects run into the issues described above.

Are the things looking any different on the front line? Not much I’m afraid.

E&P workforce, especially the earlier generation, still identifies analytics with the first principle methods such as reservoir simulation or engineering models while viewing data sciences based methods (e.g. statistics, machine learning) with skepticism. The current crop is more open to the latter but has a steep learning curve to climb in large part due to the absence of formal data sciences learning programs in the sector or academia. And while conferences are good forums for knowledge exchange, E&P analytics conferences are fast losing their novelty. Once you have attended a few, you have probably seen it all.

In one such conference, sitting in a big data session, I overheard a couple of participants sigh and mutter “the same old stuff!” Another one at the end of the session called the whole event as Groundhog Day in an obvious reference to a popular movie (with the same title) where the lead protagonist is forced to live the same day over and over again. Even though ironic, I thought that was an apt summary of the current state of the analytics in the E&P sector.

But let me end with a note of optimism. In the movie, at the end of the day (literally), hero emerges transformed and enlightened. I hope E&P sector ends its Groundhog Day on the same note and emerges truly data-driven.