In the 2016 Accenture and Microsoft oil and gas digital trends survey two-thirds of oil and gas professionals said analytics is one of the most important capabilities for transforming their company, though only 13% said their organization has fully mature analytics capabilities. Meanwhile, International Data Corp.’s recent predictions claim that by 2020, 80% of large oil and gas companies will run their business with help from a cognitive/ artificial intelligence agent capable of learning, reasoning and solving complex problems.

Clearly, there are high aspirations and challenges to overcome. Tessella has been through this journey with other industries that are now using analytics, machine learning and cognitive computing to effectively reduce costs and increase profits. The company is increasingly applying these learnings to the petroleum industry.

Five rules for success

Throughout such projects five clear rules consistently emerge that must be addressed to deliver value from data analytics.

1. Focus on business outcomes, not data. Successful analytics programs start by identifying what the business wants to achieve and what decisions must be made to do so. A drilling data project should start by first identifying the business outcome and then secondly the associated decision. For example, determine the circumstances to preventing a stuck drillstring. Then assess what data and technology are needed to inform decisions. A company may find that it was throwing away the data it needed most—data that would be missed if it started a project by looking at what data the company has. Tessella calls this approach “decisions first, data last.”

2. Have a big vision but focus on quick wins. Many data projects fail because they are too big and take too long to deliver value, leading senior teams to lose interest. Many data projects start as IT megaprojects to consolidate data without any plan for what to do with it. Data projects must have a pragmatic execution plan with milestones designed to demonstrate early success. Focus on multiple smaller projects that identify specific datasets and streams from sensors. Tessella’s stuck drillstring example may include pressure, torque, hook load, weight on bit, mud flow, ROP and rpm combined with historical data on pressure test assessments and ROP analysis. Analytics can then be applied in real time, delivering the fast-actionable results that will better support engineers.

3. Who, when and how will data be acted upon. Data success requires an understanding of who will use the data, when the information is needed and how they engage with the insights being provided. Resulting insights can then be presented in an appropriate manner for the decision maker. When there is a need to act quickly, even trained experts can’t immediately spot problems within hundreds of thousands of datapoints. It is important that the output is displayed clearly and focused on presenting key information, both raw and derived, so experts can quickly assess the situation, apply their knowledge and take the appropriate action.

4. Replace silos with translators and collaboration. Business transformational data projects transcend traditional organizational boundaries. Companies need to adopt newly evolved structures, creating a culture where data scientists are in direct contact with the business functions, the IT departments and the domain experts (engineers, geologists, etc.) to whom they are providing insights.

Teams must be led by someone with a strong understanding of both business context and technical challenges. These are vital translators who can speak the language of business and data science. This is particularly important in the oil and gas industry, where many data need to move across silos (e.g., geophysics data acquired in exploration is valuable for drilling assessment). And this is not just a technical challenge; many data owners and custodians are fiercely protective of their data. Data teams must include people who understand these data and will use them responsibly to help convince the custodians of the benefit of sharing.

5. Take a scientific approach to data science. Many analytics strategies fail because industries put technology first, and the oil and gas industry is guiltier than most. Companies invest in an analytics platform—a black box—that may rapidly identify trends in their datasets. However, these correlations may not be meaningful in a business context. To deliver real insight, the reasons for these correlations need to be fully understood.

This is where a scientific approach comes in. Data science teams that understand the data and the industry issue being investigated can design and execute finely controlled experiments that eliminate variability and hidden biases from the many data feeds, thereby dismissing accidental correlations and reducing unnecessary averaging of results. Through this approach they can identify clear lines of causation between the decisions and outcomes, providing confidence and trust in the analytics and delivered insights.

Applying the rules to drilling operations

Drilling operations offer a great example of the data science opportunity in practice. The business improvement is clear: when a drill, casing or completion string gets physically stuck in the wellbore it can cost from $10,000 to $1 million per day in nonproductive time.

Drilling generates high volumes of diverse data. Throughout the process a variety of sensors transmit data at different intervals and often in different formats. A clear approach is needed or the industry will rapidly drown in a mass of irrelevant data and fail to spot the signal in the noise. Translators are critical—those who can talk the language of the drilling engineering specialists to identify and interpret the combination of variables that is meaningful to them.

Equally vital is scientific rigor in the selection and implementation of the appropriate analytical tools. Given the data volumes and complexity of the model, machine learning is an obvious approach. For machine learning it is vital that appropriately deep and diverse training datasets are selected. Otherwise, the predictive capabilities of the system will be poor. It is important to establish confidence in the model by thoroughly validating the system outputs on the correct datasets as well as understanding the inaccuracies and potential biases in these data.

Success requires an understanding of the drilling engineer’s activity—who will use the data, how they work, when the information is needed and how they engage with the insights provided. Again, the translator comes into play.

These five rules provide a useful starting point. Learned from long-standing experience in oil and gas and other data-focused industries such as pharmaceuticals, space and defense, these rules are applicable to any advanced data project.

Naturally, this is a quick run-through and is based on a real project that took several iterations and had much greater complexity. There is an urgent need to approach data science more strategically, identifying what decisions need to be made, quickly finding the specific data that can support those decisions and designing a rigorous analytics approach to establish clear evidence of the link between a decision and the desired outcome.

Comments