5 Reasons Why ML & AI Have Let Us Down in O&G Operations

It’s time to get real: AI and ML are overused buzzwords. Ever heard of an incredible black box AI engine supported by ML applied to rich data leading to enhanced decision making that will save you time & make you money?

Umm … 🤮 Let’s move on from that.

For the past decade, technology companies have falsely promised AI & ML applications that have underdelivered on value and customer satisfaction.

Before we get to the current issues, I want to clarify that ML is a subset of AI. It is a method of training algorithms to learn how to make decisions (that’s the AI).

#1: ML & AI applications are not a part of daily operations.

Machine learning models should be set up to allow operators to interact continuously with “human in the loop” (HITL) feedback for the model. But here’s the kicker: no technology company is ACTUALLY doing this in O&G operations. They might say they are, but their “ML & AI” are actually humans behind the curtain. Uh.. what? 😱

ML & AI applications are isolated platforms that operators rarely use. It takes a back seat to their core workflows, which causes a disconnect and prevents model improvements. So, how do we get accurate ML & AI in O&G operations? The answer is pretty simple: leverage the knowledge of the field staff & engineers through thoughtful field data capture.

#2: Inputs to ML & AI are missing context, causing low-quality outputs.

There’s a common misconception that users are too “stupid” to understand the mathematical complexity of the magic “black box AI engine” and they should accept the outputs as they are. Here is a paradigm shift for you… I believe leveraging the users’ knowledge is the only way ML & AI can have accurate outputs. 🤯

We all know the old saying of “💩 in, 💩 out.” The outputs of ML & AI in operations are currently inaccurate because we are limiting ourselves to “dumb” time series data streams. We can make this time series data “rich” by prompting users to label the trends with context, like plunger replacement, frac hit, or compressor speed adjusted. Labeled time series inputs can achieve more accurate model outputs.

#3: Current “ML & AI” solutions are focused on minor pieces of Operations.

ML & AI tech companies in O&G operations have focused on artificial lift (AL) optimization which is a problem, but it is not THE problem. This is just the problem outside tech from silicon valley has chosen to focus on. Optimization of AL makes up less than 20% of the daily field and virtual touches. 80% of the work is focused on facilities, automation, measurement, maintenance, scheduling, interventions, and well troubleshooting.

Tech companies will claim they have to focus on a small problem to be effective. But they also choose to believe the problem they are solving is the most important and ignore the overwhelming amount of work outside of their niche solution.

If the industry hasn’t solved the most fundamental problem in operations — accurate work detection & distribution — how can we graduate to a higher level problem, like AL optimization? It’s like focusing on building the roof of a house when the foundation isn’t even poured!

#4: ML & AI is only as beneficial as the human experience it is based on.

Technology companies bias their ML & AI on the experiences of one or two people, which are typically contractors or an “SME” that briefly worked in operations.

These products never fully meet customer needs because they fail to evolve significantly beyond version one. Why not leverage the knowledge of the hundreds of field staff & engineers that are your customers? It is unrealistic to build a product on the experience of a few.

#5 ML & AI lack explainability.

Have you ever asked a technology company to explain how their AI & ML works? The response is usually, “well, we look at these 40+ variables and optimize based on these 10 variables, it’s complicated, but trust us, it works.” 🤬

How can anyone expect users to trust their product without being able to explain it simply? Without trust, usage is low, leading to a low value. Operations should be given radical transparency and explainability of the technology, much like this example that illustrates a complicated technology to users.

There are five levels of intelligence:

1. Smart
2. Intelligent
3. Brilliant
4. Genius
5. Simple

If tech companies know what they are doing AND are doing what they say they are doing, they should be able to explain the technology simply, right? Think about it: why do technology companies have so many analysts or customer success staff?5 Reasons Why ML & AI Have Let Us Down in O&G Operations

A new approach to ML & AI.

Tasq makes shit that works for operations.

How? We lived the problems first hand, and we were fed up with the current state of industry technology solutions, so we made our own solution. Tasq is the work flow of daily operations, we just leverage ML to help it run efficiently & effectively.

Contact us for a demo and learn how we can help solve daily operations work on multiple levels.

…Or reach out to us in 2 years after dropping $3 million on in-house solutions that cause more disruption than good.

We’re here to help either way.

‍