For the people out there in the trenches, pulling together immense permitting packages, coordinating with agencies, and running through walls to actually get projects built, diagrams like the one below probably feel a bit ridiculous.
Nonetheless, I know I'm not alone in seeing awesome diagrams like this one…
Permitting flowchart. Credit: Association of General Contractors
…and thinking to myself, “AI must be helpful here, right?”
This was the topic of a recent discussion I had with a forward-thinking executive responsible for permitting large scale energy infrastructure.
The conversation inspired us to put our thoughts down on paper to answer the question: Which challenges in the process of developing or permitting large-scale infrastructure are suitable for AI to solve?
At Blumen, we have a unique lens into the common problems across the industry and we’ve chased down the good ones and the bad ones alike. As a result, we feel like we have a good sense of what makes a permitting problem suitable to the current state of the art in AI, LLMs, and machine learning.
We believe there are two distinct classes of problems AI can be helpful with in permitting and development. We call them: Mosaic Problems and Invisible Ink Problems.
A problem is defined by (1) the current condition of the problem, (2) the traits of the data available to solve the problem, and (3) the desired business outcome you can expect from solving that problem using AI. I know that sounds pretty abstract so I’ll provide examples below.
Mosaic problems are made up of scattered disparate pieces of information stitched together to make something meaningful.
Mosaic problems save time, cost, and risk by processing a bunch of knowable information.
Problem traits: labor-intensive data gathering and reporting on facts.
Data traits: the information needed to solve the problem exists today as:
project files or institutional data (structured and unstructured)
Desired Business Outcome: a synthesized deliverable that pulls together the above data in an accurate and desired format in 1/100 of the time and 1/10 the cost of alternatives.
Mosaic example: Critical issues analysis
Problem: identifying all of the potential fatal flaws or critical issues facing a prospective project requires a lot of manual work analyzing geographic data, project information, and regulatory text. Synthesizing the results is expensive and time consuming. It also limits the number of projects that can be screened rigorously in a given year.
Data: Publicly available geospatial data, project specs and files, and federal, state, and local regulations.
Solution: A system that uses AI to automatically check all of the regulations that might impact this project against the geographic and project triggers for this site.
Business Outcome: A critical issues analysis report in 1/100 of the time and 1/10 the cost of the manual alternative.
Invisible ink problems look like hidden patterns we need to find creative ways to uncover using a bunch of past examples.
Invisible ink problems save significant time and rework by predicting non-obvious or unknowable outcomes.
Problem traits: unpredictable reviews with ambiguous rules and lots of regulator/agency discretion; these details are not reflected explicitly in documented rules or data.
Data traits: existing factual data has limited usefulness but a significant number of past filings and outcomes are publicly available.
Desired Business Outcome: A statistical overview and synthesis of past filings and outcomes that provides high predictability of regulator/agency outcomes and can be used to form a deliverable the agency will accept.
Invisible ink example: Construction and operations plan for submittal
Problem: the written regulations that outline what is required in a construction and operations plan are table stakes and regulators often augment these regulations with interpretations (based on experience) for the level of detail that must be provided in order to secure a license to build and operate. These subjective determinations of “sufficiency” for an application change over time, which creates cost and schedule uncertainty. Risk from an unpredictable back and forth with regulators ultimately increases the cost to ratepayers and delays the deployment of technology that can help address the climate crisis.
Data: ~15 past COPs are publicly available.
Solution: an AI system that learns patterns from past filings and, using factual site and geodata, gives an outline for a report that mirrors the hidden traits of past accepted filings.
Business Outcome: a COP that incorporates the “invisible ink” of augmented requirements that saves time and cost by avoiding multiple rounds of regulatory feedback.
A particularly cool attribute of "Invisible ink" problems is that they can occasionally understand latent factors driving decisions better than a human can – the golden goose of regulatory risk. One of my favorite examples of this is WOTUS-ML (the topic of a future article), which predicts Section 404 jurisdictional determination outcomes across the country with high accuracy.
Whether you find yourself confronting one of these problems and you want to take a crack at it OR you were curious enough to get this far but you didn’t understand any of this, give us a call.
Talk soon,
Hannes
It’s time to build. We’re here to help.
Thank you! We will contact you shortly.
Oops! Something went wrong while submitting the form.
.png)
%201.svg){kind=icon}
