Adding AI to the Toolbox

Technical and financial analyses in the electric power industry are generally implemented either with tightly controlled but inflexible coded applications or loosely controlled but flexible spreadsheets. With new Artificial Intelligence (AI) code writing tools will the middle ground become more popular providing the best of both approaches?

Organizations across many industries use 3rd party software for corporate-wide workflows (e.g. enterprise resource planning (ERP) programs by Oracle and others) especially for financial results and public reporting. Across several organizations implementing these ERPs, I found many downsides (inflexible, insufficient customization to the business, expensive, cumbersome interface, etc.) but one important upside which is consistent control over the workflow.

The more common the workflow (e.g. financial reporting), the more 3rd party software options are available. Industry-specific workflows such as power generation site screening or production forecasts have fewer 3rd party software options so companies also develop in-house programs. Whether 3rd party or in-house, the same trade-off exists i.e. these applications enable consistency and accuracy with tight control of revisions to the program (i.e. version control), but the programs are often inflexible and cumbersome. In-house software also carries a risk of abandonment when the programmer leaves the company.

Spreadsheets are the alternative to coded programs. Spreadsheets do not require knowledge of a programming language and provide flexibility, but that flexibility means there is essentially no version control. Across an organization, spreadsheets will usually have a wide range of assumptions, methodology and errors.

Can a hybrid middle ground approach offer a better alternative blending AI assisted code with spreadsheets to provide control and consistency but also efficiency and flexibility?

I was interested in trying. Many people by now have been impressed by what AI infused large language models (LLMs) can produce. LLMs are particularly adept at writing in computer languages. I used Anthropic’s Claude within VS Code to write Python code to access a large publicly available EIA data set and output an Excel file with plant performance benchmarks for peer projects. My code writing skills went dormant with the advent of Excel, but this was amazingly easy. I also got to experience AI hallucinations and AI’s sycophantic demeanor. Beyond just a code writer, Claude has some understanding of the context of solving a problem.

The peer project benchmarks goal of this specific example was driven by a common issue I have experienced. Whether in project development, M&A, or asset management it is necessary to establish peer project benchmarks. In the spreadsheet world, a user could cherry pick benchmarks or creatively filter out unwanted comparisons or simply do an incomplete analysis. If 3rd party software was used, the program would either be customizable which introduces the same issues as spreadsheets or non-customizable which often makes it too rigid to be useful. In-house programs could perform this simple task, but given the resources that were required to develop and maintain in-house programs they typically were designed to be all-encompassing to capture multiple workflows until they become bloated, inflexible and complex. With AI assisted code writing, simple modules can be developed that provide consistent accurate results. Organizations can be assured that key parameters supporting a decision are developed with a consistent accurate method while the project team can still utilize flexible spreadsheets to efficiently complete an analysis around those key parameters.

With AI code writing tools, an in-house programmer can be much more productive supporting many more task-specific programs, and the simplicity and consistency of the coding will make staffing transitions smoother. Instead of a single overall cumbersome in-house program, modules can be built to do subtasks while inputs and outputs can be customized easily to integrate with the overall work flow. The programs can be easy to modify, but version control can still be maintained.

3rd party software is the right choice in some applications, and spreadsheets have their place. But, in-house programs mostly written by AI could enable many more middle ground options for an organization enabling better quality, consistency, and accuracy of the analyses supporting decisions.