Building a Weighted Scoring Model

A weighted scoring model is the workhorse of site selection. Its logic is simple, which is a strength: state the factors that matter, decide how much each matters, grade every candidate site on each factor, and combine the grades into a single comparable score. The discipline is not in the arithmetic, which any spreadsheet handles, but in the choices you make at each step.

Start with the factors. For a foundational model, the six core variables, power, water, fiber, land, incentives, and community, are a sound backbone, though in practice each is decomposed into sub-factors. Power might split into availability, cost, reliability, and carbon profile; community might split into current sentiment, regulatory posture, and prior data-center experience. The goal is a set of factors specific enough to score meaningfully but few enough to keep the model legible. A matrix with sixty factors is not more rigorous than one with fifteen; it is usually less, because it buries the decisive variables under a mass of trivia and invites false precision.

Next, assign weights. This is the most consequential and most abused step. Weights encode your priorities, and they should reflect the reality established in Module 1: power dominates. A defensible model gives power the largest single share, with the exact figure depending on the workload and strategy, and distributes the remainder across the others according to their genuine importance to this project. The critical discipline is to set the weights before you look at any specific site's scores. If you weight after scoring, you will unconsciously tune the weights to favor the site you already prefer, and the model becomes a rationalization rather than an analysis. Write the weights down, justify each one in a sentence, and freeze them.

Then score each site on each factor, typically on a consistent scale such as one to five or one to ten. Scoring demands evidence, not impression. A power score should rest on the utility's stated capacity and timeline, the interconnection-queue position, and actual price data, not on a developer's optimism. A community score should rest on the jurisdiction's recent regulatory actions, public hearings, and any moratoria, not on a friendly lunch with an economic-development officer. The quality of a scoring model is determined almost entirely at this step, because a precise weighting applied to sloppy scores yields a precise, wrong answer. Where evidence is thin, record the uncertainty rather than papering over it with a confident number.

Finally, combine. Each factor's score is multiplied by its weight, the products are summed, and the sites are ranked by total. The output is a single figure per site that lets you line up candidates and see, at a glance, which lead and which trail. That figure is genuinely useful: it makes priorities explicit, it exposes disagreement (two analysts who score the same site differently can find exactly which factor they dispute), and it produces an auditable record that survives the scrutiny of an investment committee.

But the combined score must be read with a clear understanding of what it is and is not. It is a structured summary of judgments, not an objective measurement. Change the weights and the ranking can flip; change a few scores within their plausible range and a close race reverses. This sensitivity is not a flaw to hide but information to surface, which is the subject of the next lesson. Faraday's read: a scoring model earns its keep not by producing a winner but by making the reasoning behind the winner inspectable. If someone challenges your top-ranked site, the model lets you answer precisely: it leads because power availability, weighted heavily and scored on firm utility commitments, outweighs its weaker incentive package. That is a defensible answer. "It felt right" is not.

Two practical refinements separate a mature model from a naive one. First, distinguish factors that should be scored on a gradient from factors that should act as gates, pass or fail conditions that no amount of strength elsewhere can offset; the next lesson develops this. Second, resist the urge to over-engineer. The value of the model lies in forcing explicit, evidence-based judgments about a small number of variables that truly matter, and in making those judgments comparable across sites. A clean model with honest weights and well-sourced scores beats an elaborate one every time. Build the simple version first, populate it with real evidence, and add complexity only where a genuine decision hinges on it. The point of the framework is not to remove human judgment but to organize it, expose it, and hold it accountable.