AI takes aim at shale margins improvement
Poor levels of profitability in shale are prompting a search to use artificial intelligence (AI) to improve the efficiency of production and improve the bottom line
Low oil and gas prices and a failure to throw off free cash flow (FCF) amid continuing high opex costs have hugely taken the lustre off the investment case for US shale over the last year. So, the quest to improve margins has grown ever more important, with AI solutions now being pursued to add to the shale producer toolbox.
“There is too much use of existing techniques that are understood,” but which may not be optimal in cost terms, says David Cosby, founder of Longview, Texas-based oil shale R&D firm Shale Tech. One way to raise shale profitability is to focus on production rather than drilling, says Cosby, who works with Ambyint, a US firm which uses AI to optimize the artificial lift and production of shale.
Decisions on lift types are mostly taken by humans, but AI can quickly call up the relevant data points to optimise production and adapt to snags, says Cosby. “Well conditions do not stay static over time”, and human monitoring of this is time-consuming and expensive. Liquid, whether water, oil or condensation, always needs to be removed from shale wells, says Cosby, and different types of lift are required depending on the well.
Ambyint uses AI to optimise lift techniques, and this can lead to margin improvements of 10-20pc, according to Cosby. The frequency of rod pump failures can also be reduced using AI, he continues. He also points to fluid compression through techniques such as that offered by Midland, Texas oil services firm Compressco as being much cheaper than rod pump methods. “Combined with AI, this can be a game-changer for the industry.”
Harder to quantify are the potential gains from cutting down on drilling frequency as production is optimized. A 2019 Ambyint white paper argues that expanding the number of wells from a central pad is a popular strategy that can create a false sense of efficiency. Ambyint says that adding too many wells to a pad reduces the output of each well, due to well interference and pressure loss. “Those who have focused on production have been able to avoid drilling new wells—this can be taken further,” Cosby says.
Rapid growth in the US shale industry has left shareholders largely unrewarded. Investors over the last decade would have done better to buy a US benchmark index rather than put their money into shale. Weak profitability lies behind that underperformance.
According to Rob West at energy technology consultancy Thunder Said Energy (TSE), fragmented supply chains such as US shale present potential for cost savings through blockchain, which is now starting to be implemented in oil and gas procurement for the first time. TSE calculates that this could cut shale costs by up to 10pc as "smart contracts" replace labour-intensive, error-prone and hard-to-audit manual contracting.
“Combined with AI, [fluid compression] can be a game-changer for the industry” David Cosby
Research from consultancy McKinsey in August argues that shale producers, despite sustained investment, have been dogged by persistent negative FCF. The most productive use of capital often lies in maintaining existing wells, McKinsey argues. Independents must therefore must focus on operational performance rather than volume, and production monitoring through AI techniques can help raise profitability, the research says.
“The industry is evolving from brute-force fracking to customised fracs in order to maximize well productivity,” says Pavana Gadde, CEO of Houston shale well analysis firm Shale Value. “The focus on gaining deeper insight into the sub-surface has accelerated given the low commodity price environment.” Her company combines reservoir science with AI and machine learning to improve shale profitability.
The next set of improvements in well productivity are “likely to come from the industry’s advancement in analysis of the reservoir fluid behaviour,” says Gadde. This will prompt operators to better manage and optimise well spacing, she continues. She sees the main current obstacle as the fact that rock, fluids and other variables are different for each well, so the data cannot yet be used to feed AI models.