AI Election Test: A Glimpse into the Future of Strategic Insight

OKLAHOMA CITY, OK – August 05, 2026 – In a move that blended high-stakes politics with technological audacity, reporters from The Oklahoman sat down in a conference room less than a week before the state's June 16 primary with a unique challenge. They asked Saxum, a local strategic consultancy, to use its proprietary AI tool, Foundry DST, to call a slate of races before a single vote was cast. The results of this live-fire exercise offer a compelling window into the power, pitfalls, and profound institutional implications of predictive artificial intelligence.

By publicly testing its county-level sentiment model, Saxum not only demonstrated remarkable predictive capabilities but also embraced a level of transparency rare in the opaque world of AI development. The mixed results provide a crucial case study for how such tools are poised to reshape decision-making far beyond the campaign trail.

The Digital Crystal Ball: A Test of Predictive Power

The most striking success for Foundry DST came in its forecast for State Question 832, a contentious measure to raise Oklahoma's minimum wage. The day before the election, the model projected that 55.42% of voters would reject the proposal. The final certified result was 55.38% opposed—a breathtakingly narrow gap of just four hundredths of a percentage point. In its post-election analysis, Saxum described the outcome as "remarkably close to the model."

This wasn't an isolated success. Foundry DST also correctly called the Republican primaries for lieutenant governor and attorney general. Furthermore, it accurately projected that Mike Mazzei would secure a spot in the August gubernatorial runoff. The model’s granularity was on full display, correctly identifying Mazzei's base of support in rural western counties and rival Chip Keating's strength in the suburban counties surrounding Oklahoma City.

Built on a foundation of roughly 15,000 data points—spanning everything from census figures to tribal enrollment across Oklahoma’s 77 counties—Foundry DST synthesizes outputs from three distinct AI models to generate a single consensus forecast with an attached confidence score. This architecture is designed to provide what Saxum calls "anticipatory sentiment," allowing organizations to test how a message or policy will land before it goes public, moving strategy from guesswork to evidence-based forecasting.

When Rational AI Meets Real Voters

For all its successes, the model also had a significant and instructive miss. While it correctly placed Mike Mazzei in the gubernatorial runoff, it incorrectly projected that he would face Chip Keating. Instead, Gentner Drummond surprised many by finishing first, with Mazzei a close second and Keating falling to third.

In a candid post-mortem, Saxum didn't hide from the error. Hart Brown, the firm's President of AI & Transformation, attributed the miss to a classic failure mode in artificial intelligence. "It started acting in what we call a rational actor theory," he told The Oklahoman. The model over-weighted what would logically and rationally benefit voters most, failing to fully account for the complex, often less-than-rational drivers of human behavior.

Saxum's analysis identified two key factors the model underweighted: the surprising effectiveness of candidate Jake Merrick's messaging in his home region and, critically, a late surge of voter sentiment around the issue of foreign land ownership that buoyed Drummond, who had higher name recognition. The AI, in its logical purity, missed the emotional and behavioral currents that ultimately decided the race. In response, Saxum has already reinforced the behavioral side of the model, reduced the influence of rational choice theory, and recalibrated how it profiles candidates to focus more on voter response and less on candidate platforms.

Accountability in the Algorithm

The most significant aspect of the Oklahoma experiment may not be the technology itself, but the philosophy guiding it. In a field rife with 'black box' algorithms and proprietary secrecy, Saxum’s decision to open its model to public scrutiny and openly discuss its failures is a notable deviation. This commitment to transparency is a cornerstone of the company's approach.

"Used the right way, tools like this are an opportunity, not a risk, a way to communicate with more precision and more respect for the people you're trying to reach," said Peter Farrell, CEO of Saxum. "That means using it openly, with real accountability, and with people, not algorithms, making the final call."

This human-centric approach is baked into Foundry DST's process. It is not a self-serve platform; every analysis is reviewed by an expert analyst who applies local context and human judgment to the AI's output. This 'human-in-the-loop' system is designed to mitigate the inherent biases and blind spots of a purely algorithmic model, creating a partnership between machine efficiency and human wisdom. Brown has even suggested a general guideline of 70% human oversight to 30% AI efficiency to manage risk effectively, a principle that resonates amid growing ethical concerns about AI's role in influencing public discourse.

Beyond the Ballot Box: The New Frontier for Sentiment Analysis

While the political arena provided a dramatic proving ground, the true institutional impact of tools like Foundry DST lies in their application across a wide array of sectors. Saxum reports that demand is surging from industries well beyond politics, including marketing, public affairs, government, energy, and healthcare.

For any large organization, the ability to stress-test a major announcement, policy rollout, or marketing campaign at a granular, county-by-county level represents a paradigm shift. It transforms communications from a reactive art to a proactive science, allowing leaders to identify potential backlash, refine messaging for specific communities, and allocate resources with far greater precision. This capability addresses what Saxum identifies as a fundamental gap left by traditional market research, which is often too slow or broad to inform fast-moving strategic decisions.

By providing predictive, localized intelligence, these advanced AI systems offer a powerful tool for de-risking critical business and policy decisions. The lessons learned from a primary election in Oklahoma are now informing a new frontier of strategic communication, where understanding community sentiment is the key to building trust and achieving institutional goals.