As president and founder of Novi Labs, we are often considered as an alternative to building similar solutions in house. Clearly, there are pluses and minuses with either approach, but often we see Oil & Gas operators embarking on a “build” path for their analytics journey without a full appreciation of what “building” really means. So, the purpose of this article is to point some reasons why a “buy” approach should be given serious consideration.
1. Oil & Gas Companies Are NOT Software Development Shops
While we have seen some pretty amazing technical advances in the industry, it is rare that an Oil & Gas company is able to successfully build software that reaches the target audience in a way that truly affects business outcomes. I worked in the industry, so I have a unique perspective on this. Building software is hard, it requires specialized workflows and a lot of experience and knowledge to do it well, just like drilling oil & gas wells. Converting petroleum engineers to software engineers is as likely to be successful as converting software engineers to petroleum engineers. Experience in an engineering discipline is not hot swappable.
2. Software Platforms Arbitrage Out Risk
Inevitably, when you build software, you make mistakes. A software company is built to manage and curtail these risks, it is just part of their DNA. When software companies make mistakes and build software that customers do not use, they remove it from their platform and eat the R&D expense as risk capital gone bad. In most large companies, these inevitable early missteps and failures in early days can kill programs. So, when you “buy” software, you are in effect arbitraging out this risk, as what you are getting represents the software engineering experiments that worked, and what you are not getting is all the software engineering experiments that did not work. Someone else (the software company) paid for that R&D and ate it when it didn’t work out.
Outside of normal “swing and miss” software development risks, the other major risk often overlooked is that internal build programs often rely on a very small group of internal employees that know how the solution is cobbled together. They know where all the bodies are buried. What happens if they leave, and go to a competitor? I have seen the shambles left behind at several of our prospects when this happens; it can be extremely disruptive if a solution that is relied on for business decision making suddenly has no supporting cast. Software companies manage flight risk, and if they do it well, there is very little risk associated with one engineer leaving, which, in effect, curtails that risk for their customers.
3. Hiring Top Notch Data Scientists and Computer Scientists Is Really Tough
There are really two problems here. First, top tier computer scientists and data scientists do not typically want to work in industry or for consulting firms unless the circumstances are exactly right. They prefer either software startup companies, or, they prefer working for large tech companies that are entirely built around software engineering. Mental models of top notch engineers are focused around solving high scale problems. One company’s problems do not typically represent the type of scale that they find interesting.
4. Building a Full Stack Artificial Intelligence Solution Means You Have to Solve Three Very Hard Problems
AI solutions are only as good as the data that they are built based on. To deploy AI driven software successfully, you have to solve 1) a data aggregation & management problem; 2) a model optimization problem; and 3) an interactivity problem. AI Models in and of themselves are useless without software to interact with them that is particularly focused on a business workflow. Failure to execute well on either of these three problems means that your overall program is subject to massive risk.
5. The Difficulty of Sustainability is Not Taken Into Account
Let’s say you are successful solving the three very hard problems mentioned above. Your reward for doing this is that you inherit the responsibility to sustain all three of those going forward. AI models degrade in effectiveness and accuracy naturally over time. In energy analytics, we find that it is necessary to retrain the model based on latest data fairly frequently so lessons learned from new data can be immediately applied to investment decision. So, the implication is that your platform must flow data through, regenerate and validate new models build on that new data and expose those insights to your customers in as non-disruptive a way as possible.
Outside of these “core five” considerations, there are the financial and timing aspects to consider as well. Top tier data driven E&P companies that have built successful programs internally will tell you that the cost of maintaining these programs long term is far more than they expected. And, it might take them years to build, fail, and build again until they get it right. I would argue that the cost of buying is far less than the cost of building, and a bought solution should deliver value much faster while arbitraging out significant risks.
Here is how Novi applies AI to Oil & Gas Shale Planning workflows, if you want to learn more, read this article