Capturing Value Through Machine Learning, Shell Adapts The Era of Low Oil Prices

Exploring how Shell has leveraged machine learning to adapt the era of low oil prices through predictive maintenance, optimization and safety applications

Oil prices, driven by million variables including political tensions, inventions, and regular supply-demand relationships, have been very volatile. This also translates into fluctuating operating profit margins for nonstate-owned oil majors as shown in Table 1. Oil majors are among the most diversified in the oil industry, being present across upstream, midstream and downstream and thus capturing profit swings in the value chain. Nonetheless, operating margins and cash flows dropped significantly amidst low oil prices (2015-2016). As revenues were typically constrained by the capacity of operations and the market price, oil companies focused on costs to boost operating margins through efficient management of current processes without compromising safety[i].

Planning to use technology from C3 IoT and Azure to efficiently manage the cost of goods sold and exploration/drilling costs (non-human addressable cost buckets as revealed in Table 2) in the short term while improving safety[ii], Shell launched a deliberate strategy of using machine learning (ML) across its entire operation as stated by Satya Nadella’s keynote speech at Ignite 2018[iii].

In the short term, Shell leverages ML for predictive maintenance and production optimization to reduce COGS. First, collecting sensor data from the equipment in the field, Shell can predict when equipment will fail and replaces it before it breaks[iv]. Preventing unplanned downtime of its assets has lowered costs and saved around 3.5 million barrels in lost production since the implementation[v]. Second, as mentioned by Craig Walker, CIO of Shell Downstream, Shell leverages ML for feedback selection, which brings vast value to refineries[vi] by ensuring the different blends (from 1,500 fields[vii]) do not cause operational problems, and for demand forecasting[viii]. Doing so, Shell understands market patterns more accurately and plans its production to maximize the value generated by an oil barrel. To reduce exploration/drilling costs, the company uses real-time data coming from drill bits to chart more accurate courses for wells to boost productivity and reduce drill wear&tear[ix].

In the medium term, Shell projects to rollout ML widely in its operations to improve safety in retail sites. For example, wrong fueling and smoking can threaten the safety of both customers and employees. Daniel Jeavons, GM of Shell data science, mentions that with computer vision, Shell developed ML models that use labeled footage of dangerous behaviors to train itself and continuously improve, instead of analyzing hours of CCTV footage manually. Using this, Shell intends to intervene with safety threatening issues immediately.[x]

Shell should think critically to scale ML to health, safety, and environment use. For short term, Shell can leverage its predictive capabilities to better manage its carbon footprint. Through ML, Shell can identify key variables that cause emissions, predict emissions beforehand and adjust emissions before violations occur[xi]. It may also help Shell reduce the costs incurred due to environmental regulations[xii]. In the medium term, Shell may disseminate its safety improvement initiative in retail sites to non-customer facing operations such as exploration and refining. Using image recognition systems to reduce hazardous situations, Shell can intervene with safety threatening issues for its employees. Predicting the level of corrosion in pipelines and pipes within a plant, Shell would have the opportunity to replace pipes before they start leaking hazardous liquids[xiii].

All above-mentioned uses of ML are applicable due to the abundance of data. However, the question “Can ML be leveraged to eliminate next high risk, low probability event such as Deepwater Horizon with less relevant data?[xiv]” remains unanswered.

(Word Count: 800)

[i] Sarah Kent, “Shell Prepares for ‘Lower Forever’ Oil Prices,” Wall Street Journal, July 27, 2017, [], accessed November 2018.

[ii] Steven Norton, “Shell Announces Plans to Deploy AI applications at Scale,” Wall Street Journal, Sep 20, 2018, [], accessed November 2018.

[iii] Satya Nadella, “Ignite Vision Keynote,” speech given at Microsoft Ignite 2018 conference, Orlando, FL, September 24, 2018.

[iv] Ed Crooks, “The week in Energy AI and Oil,” Financial Times, Sep 25, 2018, [], accessed November 2018.

[v] Shell, “Digital Innovation,”, accessed in November 2018.

[vi] McKinsey Energy Insights, “Optimization,”, accessed in November 2018.

[vii] Ivanhoe, L. F, and G G. Leckie, -“Global oil, gas fields, sizes tallied, analyzed,” Oil and Gas Journal. Feb. 15, 1993, pp. 87-91

[viii] Maxine-Laurie Marshall, “Shell taps into AI to streamline operations and refine customer-centricity,” Global Intelligence for the CIO, Aug 2017, [], accessed November 2018.

[ix] Steven Norton, “Shell Announces Plans to Deploy AI applications at Scale,” Wall Street Journal, Sep 20, 2018, [], accessed November 2018.

[x] Shell, “Machine Vision – Helping Shell Retail achieve Goal Zero,” Youtube, published Sep 26, 2018, [], accessed November 2018

[xi] ABB, “Using artificial intelligence to reduce emissions,”, accessed November 2018

[xii] Timothy Gardner, “U.S. rule to cut toxic emissions at refineries,” Reuters, September 29, 2015, [], accessed November 2018.

[xiii] Daniel Schlecht, Philipp A. Gerbert and Sylvain Santamarta (The Boston Consulting Group), “Artificial Intelligence: The Ultimate Disrupting Force for Oil and Gas Companies,” SPE Norway Volume 3 2017 September

[xiv] Simen Eldevik, “AI + Safety,” Position Paper, DNV GL, published on 28 Aug 2018


Future-Proofing Financial Services: Barclays Accelerator Program and Open Innovation


John Deere: Betting the Farm on Machine Learning

Student comments on Capturing Value Through Machine Learning, Shell Adapts The Era of Low Oil Prices

  1. I think it is interesting how Shell is able to use Machine Learning in the contexts of both worker safety and emissions reduction and can envisage other linkages to its overall supply chain. To your question, it would definitely be difficult to do so solely on this basis in the short term. The crux of machine learning is indeed the dataset as you point out and I believe setting out to create a comprehensible library of data (image/ audio recognition, harnessing the Internet of Things) would be one of the key steps to determine how impactful this initiative will be. That said, through a supervised or semi- supervised learning approach, I believe there are benefits in complementing the existing procedures with these new techniques of safety intervention.

  2. the Article does a great job at listing and explaining the various ways through which machine learning is being utilized in the oil and gas industry. through uptime optimization, sales forecasting, safety standards observance and emission control machine learning is clearly a trend worth noting in the context of the industry’s future.
    It would have been interesting if the post also discussed how another mega trend in this industry, shale, will either amplify or mute the impact of machine learning.

Leave a comment